Merge pull request #698 from AIOSAI/dev

Merge playbook SOP fixes from live run PPLAN-0010
release: bump 2.7.2 — compass v2 + ambient recall window. New cadence ruling: PATCH bump + tag every merge, PyPI tracks main
2026-07-16 23:43:55 -07:00 · 2026-07-16 23:29:26 -07:00 · 2026-07-16 23:19:50 -07:00 · 2026-07-16 23:04:50 -07:00 · 2026-07-16 22:44:01 -07:00 · 2026-07-16 22:04:03 -07:00
140 changed files with 15714 additions and 839 deletions
@@ -6,8 +6,12 @@
    "presence_gate": {
      "enabled": true,
      "handler": "aipass.hooks.apps.handlers.security.presence_gate.handle",
-      "matcher": "",
-      "provider_wired": false
+      "matcher": ""
+    },
+    "persistent_alert": {
+      "enabled": true,
+      "handler": "aipass.hooks.apps.handlers.prompt.persistent_alert.handle",
+      "matcher": ""
    },
    "identity_injector": {
      "enabled": true,
@@ -34,11 +38,22 @@
      "handler": "aipass.hooks.apps.handlers.prompt.navmap.handle",
      "matcher": ""
    },
+    "compass_recall": {
+      "enabled": true,
+      "handler": "aipass.hooks.apps.handlers.prompt.compass_recall.handle",
+      "matcher": "",
+      "max_per_session": 10
+    },
    "auto_process": {
      "enabled": true,
      "handler": "aipass.hooks.apps.handlers.lifecycle.auto_process.handle",
      "matcher": "",
      "timeout": 120
+    },
+    "user_message_relay": {
+      "enabled": true,
+      "handler": "aipass.skills.lib.telegram.apps.handlers.user_message_relay.handle",
+      "matcher": ""
    }
  },

@@ -67,11 +82,6 @@
      "enabled": true,
      "handler": "aipass.hooks.apps.handlers.security.registry_gate.handle",
      "matcher": "Bash|Edit|MultiEdit|Write|NotebookEdit"
-    },
-    "engine_test_sound": {
-      "enabled": false,
-      "command": "python3 $AIPASS_HOME/.claude/hooks/engine_test_sound.py",
-      "matcher": "WebSearch"
    }
  },

@@ -1,16 +1,16 @@
 # AIPass — Navigation map

-<!-- .aipass/tier1_navmap.md — Tier 1, injected periodically (cadence period 5) + at session start + right after compaction, when you most need the map back. The kernel (.aipass/tier0_kernel.md) arrives every turn; deep reference lives in `drone @agent --help` and topic guides. Size cap: keep the per-fire output under ~8,000 characters (the hook truncates near 10k). Format: .aipass/PROMPT_STYLE.md -->
+<!-- Tier 1 — injected on cadence 5, at session start, and post-compaction. Kernel = tier0_kernel.md, every turn. Cap: ~8,000 chars per fire (hook truncates near 10k). Format: PROMPT_STYLE.md -->

-AIPass is the system: autonomous agents (citizens) with identity, memory, and a mailbox, providing services to each other and to external projects. Each agent lives in a branch — its home and address. Everything routes through `drone`.
+AIPass is the system: autonomous agents (citizens) with identity, memory, and a mailbox, providing services to each other and to external projects. Each agent lives in a branch — its home and address. Everything routes through `drone`. **AIPass is open source** — public repo on GitHub. Strangers read, clone, and scan this code; treat external findings as contributions.

 # Finding your way

-You can't carry everything; you can find anything — you're the librarian, not the encyclopedia. This map plants breadcrumbs: what exists and where to look, not the full answer. A breadcrumb is the trigger to fetch the answer, not the answer. Cheapest, highest-signal sources first:
+You can't carry everything; you can find anything. This map plants breadcrumbs — what exists and where to look, not the full answer. Cheapest, highest-signal sources first:

- - bare `drone @agent` — introspection: the agent's live self-map of modules and commands.
- - `drone @agent --help` — the full curated reference. Source of truth for usage.
- - the agent's `README.md` — best quick overview of its domain and shape.
+ - bare `drone @agent` — the agent's live self-map of modules and commands.
+ - `drone @agent --help` — the full reference, source of truth for usage.
+ - the agent's `README.md` — quick overview of its domain.

 # Terminology

@@ -18,15 +18,15 @@ You can't carry everything; you can find anything — you're the librarian, not
 - Agent (citizen) — persistent identity in a branch: passport (`.trinity/`), memories, mailbox. Addressable as `@name`. You belong, you persist.
 - Sub-agent — disposable worker spawned for a task. No passport, no memory, not a citizen.
 - Registry — machine-managed catalogs (`registry.json`, flow/spawn registries). Never hand-edit — owners manage them.
- - Settings — provider `~/.claude/settings.json` (machine-wide, personal, don't touch) · project `<project>/.claude/settings.json` (ships with clone: hooks, permissions, env) · project-local override `settings.local.json`.
+ - Settings — provider `~/.claude/settings.json` (personal, don't touch) · project `.claude/settings.json` (ships with clone) · local override `settings.local.json`.

 # The framework

-Every branch is built the same. All agents live at `src/aipass/<name>` · mail address `@<name>`.
+Every branch is built the same: `src/aipass/<name>` · mail `@<name>`.

 ```
 src/aipass/<name>/
-├── .trinity/           # identity & memory (passport, local, observations)
+├── .trinity/           # identity & memory
 ├── .aipass/            # branch prompt
 ├── .ai_mail.local/     # mailbox
 ├── apps/
@@ -39,23 +39,23 @@ src/aipass/<name>/

 # The agents

- - @drone — command router. Resolves `@agent`, routes commands, enforces tier-based access. Also the only git interface (`drone @git`).
+ - @drone — command router. Routes commands, enforces tier-based access. Also the only git interface (`drone @git`).
 - @devpulse — orchestration hub, the user's primary collaborator. Coordinates the other agents, dispatches work, only agent with git write.
- - @aipass — the user's front-door concierge and its OWN CLI, NOT drone-routed: run `aipass` / `aipass --help` directly, never `drone @aipass` (drone can't resolve it). The human's best friend — onboarding (`aipass init`/`install`), `doctor` health, help chat, and OS/system questions ("why's my wifi dropping", "why's CC hogging CPU", "what is drone", "how do I make a project"). Serves humans, not agents — reads, never writes.
+ - @aipass — the user's front-door concierge, its OWN CLI: run `aipass` directly, never `drone @aipass` (drone can't resolve it). Onboarding (`init`/`install`), `doctor` health, help chat, OS/system questions. Serves humans, not agents — reads, never writes.
 - @ai_mail — inter-agent email. `dispatch` = send + wake (default for handing work), `email` = no wake, plus inbox/view/reply/close.
- - @flow — plan lifecycle: create, list, close, templates, registry. Plan types in the Plans section — never create plan files by hand.
- - @seedgo — code standards and audits. The standard pack, `audit` and `checklist`, the quality gate before and after building.
- - @prax — logging and monitoring. The only logging system: `from aipass.prax import logger`. Real-time monitor, dashboards. Logs are the first diagnostic tool.
- - @memory — long-term memory. Archives overflowing `.trinity/` files into searchable vectors; `search` recalls past sessions. Nothing is lost — it moves deeper.
+ - @flow — plan lifecycle: create, list, close, templates, registry. See the Plans section.
+ - @seedgo — code standards and audits. `audit` and `checklist` — the quality gate before and after building.
+ - @prax — logging and monitoring. The only logging system: `from aipass.prax import logger`. Real-time monitor, dashboards, runaway-log detection. Logs are the first diagnostic tool.
+ - @memory — long-term memory. Archives overflowing `.trinity/` files into searchable vectors; `search` recalls past sessions.
 - @spawn — branch lifecycle. Creates, updates, syncs, retires agents — scaffolding, passports, registry, templates.
- - @hooks — Claude Code hook engine. Prompt injection and cadence, security gates (git/edit/rm), bridges, per-project config, sound.
+ - @hooks — Claude Code hook engine. Prompt injection and cadence, security gates (git/edit/rm), bridges, persistent alerts, per-project config, sound.
 - @trigger — event handling. Pub/sub event bus, error detection (medic), log watching, error registry. Detects and dispatches — owners fix.
 - @api — external API gateway. Authenticated service clients (Google, OpenRouter, more), OAuth flows, key management, resilience.
 - @cli — display formatting with Rich. Shared rendering for terminal output.
- - @skills — capability framework. Discoverable, self-contained skill units any agent can run; consume AIPass services as opt-in imports (e.g. the Telegram skill).
- - @daemon — task scheduler. Cron-triggered firing; each branch owns its `.daemon/schedule.json`, the daemon discovers and fires.
- - @commons — the social space. Where branches post, comment, vote, and gather as a community.
- - @backup — local-first backups. Snapshots + versioning + restore for any directory; optional Google Drive sync (live, per-file mirror — slow on huge file counts, respect `.backupignore`). `.backup/` is a shared runtime namespace — @memory rollover and @flow (plan archive) also write there.
+ - @skills — capability framework. Discoverable, self-contained skill units any agent can run (e.g. the Telegram skill).
+ - @daemon — task scheduler. Each branch owns its `.daemon/schedule.json`; the daemon discovers and fires.
+ - @commons — the social space. Branches post, comment, vote.
+ - @backup — local-first backups. Snapshots, versioning, restore for any directory; optional Google Drive sync. `.backup/` is shared — @memory rollover and @flow archives write there too.

 # Daily commands

@@ -65,10 +65,20 @@ drone @ai_mail inbox                               # check mail → view <id>
 drone @flow create . "Subject" [dplan]             # new plan (default FPLAN)
 drone @seedgo audit aipass @branch                 # standards audit (drop @branch = all)
 drone @seedgo checklist <file|dir>                 # quick standards check
+drone @trigger medic mute @<self>                  # BEFORE build/edit work — auto-expires 24h
 drone @git status / diff / log                     # read-only git awareness
 drone @memory search "query"                       # recall archived context
 ```

+# Talking to other agents
+
+Citizens dispatch each other directly — allowed and expected, no permission needed. Pick by one question: does the recipient need to ACT?
+
+ - Need an answer, input, or work from them → `dispatch` (send + wake). A sleeping agent never reads plain email — a question sent as `email` stalls unread.
+ - FYI only (status, steering an agent already awake) → `email` (no wake).
+ - Replies never wake — wake-back does: when an agent you dispatched completes, YOU are woken. Team mission: the lead dispatches each phase BEFORE sleeping; the worker replies normally; wake-back returns the lead to verify and hand off the next phase.
+ - Exception — managers (`citizen_class: manager`, e.g. @devpulse) are never dispatched — the wake is blocked. `email` them; the mail lands and they see it live.
+
 Always reply to dispatches — reply auto-closes. No silent completions.

 # Plans — flow
@@ -78,15 +88,15 @@ Plans carry context so you don't have to. Create only via `drone @flow create <p
 - DPLAN — dev plan. Thinking, brainstorming, architecture. Before building.
 - FPLAN — flow plan, the default. Building and executing. `master` template = multi-phase, spawns sub-FPLANs.
 - PPLAN — playbook. A throwaway run stamped from a reusable SOP template. Operating the system, not changing it.
- - More types exist and new ones register over time. Named a type you don't know? `drone @flow templates` lists them all, live.
+ - More types register over time — `drone @flow templates` lists them all, live.

 # Sub-agents

- - Default to sub-agents for reading, searching, building, testing, research. Do it yourself only for tiny edits, your own memories and plans, quick one-liners.
+ - Default to sub-agents for reading, searching, building, testing, research. Do it yourself only for tiny edits, your own memories/plans, one-liners.
 - One clear task per agent. Brief with full context — they know nothing of your conversation.
 - No git, no memory, no dispatch. They build and report; you decide and act.
 - Sub-agent = local disposable worker. Dispatch (`@ai_mail`) = wake a citizen with memory and identity. Branch-expert work → dispatch; else → sub-agent.
- - Models, good practice: opus for build and analysis, sonnet for routine investigation, haiku for trivial mechanical tasks. Never fable for sub-agents.
+ - Models: opus for build/analysis, sonnet for routine investigation, haiku for trivial mechanical tasks. Never fable for sub-agents.

 # Memory — .trinity/

@@ -95,12 +105,11 @@ Your continuity across sessions. Save proactively — after milestones, decision
 - `passport.json` — identity. Update only when identity genuinely evolves.
 - `local.json` — session log, key learnings, todos.
 - `observations.json` — what you learn about the user.
- - Overflow rolls to vectors automatically — never trim by hand. Two ChromaDB stores: your branch's `.chroma` (local) + a global one across all branches. `drone @memory search "query"` recalls them. Search before assuming you're cold.
- - Entry caps are hook-enforced (over-limit edit = rejected whole). The live cap is rendered in each file's `*_meta` line — read it before writing, draft to ~80% of it; if rejected, rewrite hard in one pass.
+ - Overflow rolls to vectors automatically — never trim by hand. `drone @memory search "query"` recalls it — search before assuming you're cold.
+ - Entry caps are hook-enforced (over-limit edit = rejected whole). The live cap is in each file's `*_meta` line — read it before writing, draft to ~80%; if rejected, rewrite hard in one pass.

 # House rules

- - Cross-platform, no hardcoded paths. Public repo — `pathlib`, never `/home/...`.
+ - Public repo — write as if it ships, because it does. No secrets in the tree, no hardcoded paths (`pathlib`, never `/home/...`), cross-platform.
 - No bare imports — always `from aipass.<agent>.apps...`.
- - Registries are machine-managed (spawn, flow) — never hand-edit them.
 - State lives in `.trinity/` and dashboards, never in prompts. Prompts are signposts; memories record; registries catalog.
@@ -55,7 +55,12 @@ Quick checks beat assumptions: `ls`/`find` for files, `git ls-files`/`grep` for
 - Run `drone @ai_mail inbox 2>/dev/null` — report any unread emails
 - Close any that were already processed but not formally closed

-## 5. Loose Ends
+## 5. Compass Review (Devpulse only)
+
+- Run ONE `drone @devpulse compass review` — it serves the oldest-unreviewed entry. Judge it: still true → confirm; superseded → archive it and note what replaced it; wrong → fix or archive.
+- One entry per prep, every prep. This is the curation cadence — review only works if it actually runs (DPLAN-0246: all 127 entries sat unreviewed because nothing invoked it).
+
+## 6. Loose Ends

 - Flag anything in-flight: running background agents, dispatched branches waiting for replies, pending decisions
 - If anything can't survive compaction (e.g., agent IDs needed for resume), write it to local.json todos[]
@@ -71,5 +76,6 @@ Prep complete:
 - Plans: [which ones updated]
 - Git: [branch, uncommitted count, suggestion]
 - Inbox: [count, action taken]
+- Compass: [entry #N reviewed — verdict]
 - Loose ends: [any flagged]
 ```
@@ -12,6 +12,31 @@ updates:
      prefix-development: "deps"
      include: "scope"

+  # Hash-pinned CI tool installs (ruff/build/pytest/pip-audit/pip). Pinning is
+  # what Scorecard's Pinned-Dependencies wants, but a frozen pin rots: these
+  # locks are what security.yml's pip-audit scans, so a new advisory against a
+  # pinned dep reds the job until the pin moves. This entry is what keeps that
+  # window short. Dependabot reads the `pip-compile ...` command out of each
+  # .txt header and regenerates the lock (hashes included) from the .in.
+  # Separate from the "/" pip entry above, which tracks pyproject.toml.
+  - package-ecosystem: "pip"
+    directory: "/.github/requirements"
+    schedule:
+      interval: "weekly"
+    labels:
+      - "ci"
+    open-pull-requests-limit: 5
+    commit-message:
+      prefix: "ci"
+      include: "scope"
+    # packaging/pygments are shared across build.txt, e2e.txt and audit.txt.
+    # Ungrouped, one bump fans out into several PRs that each rewrite a subset
+    # of the locks and conflict with each other. One PR per week moves them all.
+    groups:
+      ci-tooling:
+        patterns:
+          - "*"
+
  - package-ecosystem: "github-actions"
    directory: "/"
    schedule:
@@ -0,0 +1,16 @@
+# pip-audit for the security.yml `dependency-scan` job, hash-pinned (Scorecard:
+# Pinned-Dependencies).
+#
+# Target env: ubuntu-latest, Python 3.13. pip-audit's own dependency tree is
+# resolved and hashed here; the project itself is still installed unpinned via
+# `pip install -e .` and scanned with `pip-audit --skip-editable`, so pinning
+# this file does not narrow what the audit covers.
+#
+# TRADE-OFF: pip-audit scans the whole environment, including its own deps. They
+# used to float to latest on every run (self-healing); pinned, a new advisory
+# against one of them reds this job until the pin moves. Dependabot's `pip`
+# entry for /.github/requirements is what keeps that window short.
+#
+# Regenerate:
+#   pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=audit.txt audit.in
+pip-audit==2.10.1
@@ -0,0 +1,330 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=audit.txt audit.in
+#
+boolean-py==5.0 \
+    --hash=sha256:60cbc4bad079753721d32649545505362c754e121570ada4658b852a3a318d95 \
+    --hash=sha256:ef28a70bd43115208441b53a045d1549e2f0ec6e3d08a9d142cbc41c1938e8d9
+    # via license-expression
+cachecontrol[filecache]==0.14.4 \
+    --hash=sha256:b7ac014ff72ee199b5f8af1de29d60239954f223e948196fa3d84adaffc71d2b \
+    --hash=sha256:e6220afafa4c22a47dd0badb319f84475d79108100d04e26e8542ef7d3ab05a1
+    # via
+    #   cachecontrol
+    #   pip-audit
+certifi==2026.6.17 \
+    --hash=sha256:024c88eeec92ca068db80f02b8b07c9cef7b9fe261d1d535abfd5abd6f6af432 \
+    --hash=sha256:2227dcbaafe0d2f59279d1762ddddc37783ed4354594f194ffc31d20f41fc3db
+    # via requests
+charset-normalizer==3.4.9 \
+    --hash=sha256:0327fcd59a935777d83410750c50600ee9571af2846f71ce40f25b13da1ef380 \
+    --hash=sha256:03d07803992c6c7bbc976327f34b18b6160327fc81cb82c9d504720ac0be3b62 \
+    --hash=sha256:04ce310cb89c15df659582aee80a0603788732a5e017d5bd5c81158106ce249c \
+    --hash=sha256:0d861473f743244d349b50f850d10eb87aeb22bbdcc8e64f79273c94af5a8226 \
+    --hash=sha256:0e94703ec9684807f20cfb5eed95c70f67f2a8f21ad620146d7b5a13677b93e5 \
+    --hash=sha256:0fa1aec2d32bcc03c8fa0f6f1712caad1adc38509f31142112e5c9daf5b9c833 \
+    --hash=sha256:16b65ea0f2465b6fb52aa22de5eca612aa964ddfec00a912e26f4656cbef890b \
+    --hash=sha256:16d10d789dd9bcca1173c95af82c58433122564b7bc39385124be735a35cbe99 \
+    --hash=sha256:19ac87f93086ce37b86e098888555c4b4bc48102279bae3350098c0ed664b501 \
+    --hash=sha256:1d22856ffbe153a602df38e4a5464f0b748a54002e0d69ac6d2ad0a197cc99ec \
+    --hash=sha256:21e764fd1e70b6a3e205a0e46f3051701f98a8cb3fad66eeb80e48bb502f8698 \
+    --hash=sha256:231ddcbb35e2ff8973e1365db41fe0572662893b99a05deb183b68ad4c0c8bd4 \
+    --hash=sha256:253a4a220747e8b5faf57ec320c4f5efb0cef05f647420bf267143ec15dba10a \
+    --hash=sha256:280081916dc341820640489a66e4696049401ef1cf6dd672f672e70ad915aca3 \
+    --hash=sha256:2a441ea71902098ffe78c5abe6c494f44160b4af614ed16c3d9a3b1d17fd8ee2 \
+    --hash=sha256:304b13570067b2547562e308af560b3963857b1fa90bd6afd978130130fe2d6a \
+    --hash=sha256:32286a2c8d167e897177b673176c1e3e00d4057caf5d2b64eef9a3666b03018e \
+    --hash=sha256:33bdcc2a32c0a0e861f60841a512c8acc658c87c2ac59d89e3a46dacf7d866e4 \
+    --hash=sha256:375b83ed0aecfce76c16d198fbc21f3b11b337d68662bea0a995046682a11419 \
+    --hash=sha256:3c09a49d6cde137258beb3d551994a2927fd35ad5cf96aed573f61bbd67c5f84 \
+    --hash=sha256:3d92613ec25e43b05f042302531ec0f00b8445190e43325880cbd6ab7c2581da \
+    --hash=sha256:40a126142a56b2dfc0aacbad1de8310cbf60da7656db0e6b16eebd48e3e93519 \
+    --hash=sha256:416c229f77e5ea25b3dfd4b582f8d73d7e43c22320302b9ab128a2d3a0b38efe \
+    --hash=sha256:432786d3561e69aeeae6c7e8648964ce0ad05736120135601f87ac26b9c83381 \
+    --hash=sha256:43b9e366a31fdd1c87d0eb08f579b4a82b723ea54338f040d6b4e518a026ea29 \
+    --hash=sha256:440eede837960000d74978f0eba527be106b5b9aee0daf779d395276ed0b0614 \
+    --hash=sha256:45b0cc4e3556cd875e09102988d1ab8356c998b596c9fced84547c8138b487a0 \
+    --hash=sha256:476743fe6dfe14a2da12e3ac79125dc84a3b2cf8094369a47a1529b0cd8549fe \
+    --hash=sha256:4773092f8019072343a7447203308b176e10199920eb02d6195e81bbb3274c29 \
+    --hash=sha256:4b3dac63058cc36820b0dd072f89898604e2d39686fe05321729d00d8ac185a0 \
+    --hash=sha256:4d1c96a7a18b9690a4d46df09e3e3382406ae3213727cd1019ebade1c4a81917 \
+    --hash=sha256:51307f5c71007673a2bf8232ad973483d281e74cb99c8c5a990af1eefa6277d9 \
+    --hash=sha256:51447e9aa2684679af07ca5021c3db526e0284347ebf4ffcec1154c3350cfe32 \
+    --hash=sha256:58150c9f9b9a552505912d182ccdf26f6396fb6094816ceebcbb20eecabaed94 \
+    --hash=sha256:5b10cd92fc5c498b35a8635df6d5a100207f88b63a4dc1de7ef9a548e1e2cd63 \
+    --hash=sha256:5e226f6218febc71f6c1fc2fafb91c226f75bdc1d8fb12d66823716e891608fd \
+    --hash=sha256:609b3ba8fcc0fb5ab7af00719d0fb6ad0cb518e48e7712d12fd68f1327951198 \
+    --hash=sha256:60f44ade2cf573dad7a277e6f8ca9a51a21dda572b13bd7d8539bb3cd5dbedde \
+    --hash=sha256:611057cc5d5c0afc743ba8be6bd828c17e0aaa8643f9d0a9b9bb7dea80eb8012 \
+    --hash=sha256:6366a16e1a25018694d6a5d784d09b046edc9eac40ea2b54065c3052672516a1 \
+    --hash=sha256:65a7ff3f705e57d392f7261b6d0550fe137c3019477431f1c355e0db0a7d3e15 \
+    --hash=sha256:673611bbd43f0810bec0b0f028ddeaaa501190339cac411f347ac76917c3ae7b \
+    --hash=sha256:67830fc78e67501f47bb950471b2dcb9b35b140084429318e862895a8e89c993 \
+    --hash=sha256:68ce9f4d6b26d5ccbf7fd4459bf75f74a0a146677ebba80597df60cbdb20e6f4 \
+    --hash=sha256:68e5f26a1ad57ded6d1cfb85331d1c1a195314756471d97758c48498bb4dcdf5 \
+    --hash=sha256:69b157c5d3292bcd443faca052f3096f637f1e074b98212a933c074ae23dc3b8 \
+    --hash=sha256:75286256590a6320cf106a0d28970d3560aad9ee09aa7b34fb40524792436d35 \
+    --hash=sha256:78841cccf1af7b40f6f716338d50c0902dbe88d9f800b3c973b7a9a0a693a642 \
+    --hash=sha256:78fa18e436a1a0e58dbd7e02fc4473f3f32cceb12df9dfca542d075961c307d2 \
+    --hash=sha256:79580094b00d1789d1f93ea55bc43cb2f611910c72235b7657f3482ddcc1b22d \
+    --hash=sha256:7b86a2b16095d250c6f58b3d9b2eee6f4147754344f3dab0922f7c9bf7d226c9 \
+    --hash=sha256:83aed2c10721ddd90f68140685391b50811a880af20654c59af6b6c66c40513c \
+    --hash=sha256:84fd18bcc17526fc2b3c1af7d2b9217d32c9c04448c16ec693b9b4f1985c3d33 \
+    --hash=sha256:871ff67ea1aad4dfd91736464934d56b32dac49f9fbe16cddba36198a7b3a0db \
+    --hash=sha256:898f0e9068ca27d37f8e83a5b962821df851532e6c4a7d615c1c033f9da6eedf \
+    --hash=sha256:8a79d9f4d8001473a30c163556b3c3bfebec837495a412dde78b51672f6134f9 \
+    --hash=sha256:8c041122946b7ba21bb32c45b1aa57b1be35527690aeb3c5c234521085632eee \
+    --hash=sha256:90c44bc373b7687f6948b693cceaea1348ae0975d7474746559494468e3c1d84 \
+    --hash=sha256:9104ed0bd76a429d46f9ec0dbc9b08ad1d2dcdf2b00a5a0daa1c145329b35b44 \
+    --hash=sha256:920079c3f7456fa213e0829ed2073aaa727fd39d889ead5b4f35d0de5460d04f \
+    --hash=sha256:93d59d504b230e83c7a843251681959a0b6a9cd76f6e146ce1b8a80eb8739af9 \
+    --hash=sha256:9b2aff1c7b3884512b9512c3eaadd9bab39fb45042ffaaa1dd08ff2b9f8109d9 \
+    --hash=sha256:9b8e0f3107e2200b76f6054de99016eac3ee6762713587b36baaa7e4bd2ae177 \
+    --hash=sha256:9bb41182d93ea91f60b4bc8fbf4c820c69ef8a12ab2d917f3f1834f1acad07e8 \
+    --hash=sha256:9cdef90ae47919cae358d8ab15797a800ed41da7aba5d72419fb510729e2ed4b \
+    --hash=sha256:a1786910334ed46ab1dd73222f2cd1e05c2c3bb39f6dddb4f8b36fc382058a39 \
+    --hash=sha256:a4cfde78a9f2880208d16a93b795726a3017d5977e08d1e162a7a31322479c41 \
+    --hash=sha256:a4fbdde9dd4a9ce5fd52c2b3a347bb50cc89483ef783f1cb00d408c13f7a96c0 \
+    --hash=sha256:aa99adc8f081b475a12843953db36831eaf83ec33eb46a90629ca6a5de45a616 \
+    --hash=sha256:ac351b3b8014eead140e77e9717e2992c6bbe30b63bc3422422eb84865412e3d \
+    --hash=sha256:ad41ba96094304aa090f5a30cb6e4fb3b3f1c264c523394b4c39bbacc4dc92ba \
+    --hash=sha256:b5314963fce9b0b12743891de876e724997864ee22aa496f903f426c7e2fa5b2 \
+    --hash=sha256:bcf74c1df76758a395bf0af608c04c82257523f55c9868b334f06270d0f2112b \
+    --hash=sha256:bd47ba7fc3ca94896759ea0109775132d3e7ab921fbf54038e1bab2e46c313c9 \
+    --hash=sha256:c0323c9daef75ef2e5083624b4585018a0c9d5e3b40f607eed81a311270b934b \
+    --hash=sha256:c1225416b463483160e4af85d5fc3a9690ccb53fd4b1865a6437825f5ede3209 \
+    --hash=sha256:c1c948747b03be832dceed96ca815cef7360de9aa19d37c730f8e3f6101aca48 \
+    --hash=sha256:c25fe15c70c59eb7c5ce8c06a1f3fa1da0ecc5ea1e7a5922c40fd2fa9b0d5046 \
+    --hash=sha256:cc1b0fff8ead343dae06305f954eb8468ba0ec1a97881f42489d198e4ce3c632 \
+    --hash=sha256:cd6280cf040f233bd7d3407b743b4b4c74f70e8e1c4199cb112a62c941c0772a \
+    --hash=sha256:cd6c3d4b783c556fa00bf540854e42f135e2f256abd29669fcd0da0f2dec79c2 \
+    --hash=sha256:d4d6fcde76f94f5cb9e43e9e9a61f16dacefd228cbbf6f1a09bd9b219a92f1a1 \
+    --hash=sha256:ddf4af30b417d9fe16481e9b81c27ab2a7cde1ff7ba3e85653b02db7d145dc7b \
+    --hash=sha256:df115d4d83168fdf2cae48ef1ff6d1cb4c466364e30861b37121de0f3bf1b990 \
+    --hash=sha256:df7276909358e5635ae203673ab7e509ddd224225a8d6b0790bf13eb2bde1cc5 \
+    --hash=sha256:e4fd89cc178bced6ad29cb3e6dd4aa63fa5017c3524dbd0b25998fb64a87cc8b \
+    --hash=sha256:e9701d0049d92c16703a42771b98d560b95248949f23f8cf7b4eddd201814fb9 \
+    --hash=sha256:ee2f2a527e3c1a6e6411eb4209642e138b544a2d72fe5d0d76daf77b24063534 \
+    --hash=sha256:f7fb7d750cfa0a070d2c24e831fd3481019a60dd317ea2b39acbcebc08b6ed81 \
+    --hash=sha256:f840ed6d8ecba8255df8c42b87fadeda98ddfc6eeec05e2dc66e26d46dd6f58a \
+    --hash=sha256:f86c6358749bd4fda175388691e3ba8c46e24c5347d0afd20f9b7edfc9faf07d \
+    --hash=sha256:fa36ec09ef71d158186bc79e359ff5fdd6e7996fe8ab638f00d6b93139ba4fcf \
+    --hash=sha256:fe2c7201c642b7c308f1675355ad7ff7b66acfe3541625efe5a3ad38f29d6115
+    # via requests
+cyclonedx-python-lib==11.11.0 \
+    --hash=sha256:3049fc83e06a059b5c5907a527625a8ed5073caab10607ed4c9e5503b590fd44 \
+    --hash=sha256:4b3194db72b613717f2912447e67ab618c75ff7dcac6c4af3c0e9e1ac617c102
+    # via pip-audit
+defusedxml==0.7.1 \
+    --hash=sha256:1bb3032db185915b62d7c6209c5a8792be6a32ab2fedacc84e01b52c51aa3e69 \
+    --hash=sha256:a352e7e428770286cc899e2542b6cdaedb2b4953ff269a210103ec58f6198a61
+    # via py-serializable
+filelock==3.29.7 \
+    --hash=sha256:5b481979797ae69e72f0b389d89a80bdd585c260c5b3f1fb9c0a5ba9bb3f195d \
+    --hash=sha256:987db6f789a3a2a59f55081801b2b3697cb97e2a736b5f1a9e99b559285fbc51
+    # via cachecontrol
+idna==3.18 \
+    --hash=sha256:7f952cbe720b688055e3f87de14f5c3e5fdaa8bc3928985c4077ca689de849a2 \
+    --hash=sha256:ffb385a7e039654cef1ab9ef32c6fafe283c0c0467bba1d9029738ce4a14a848
+    # via requests
+license-expression==30.4.4 \
+    --hash=sha256:421788fdcadb41f049d2dc934ce666626265aeccefddd25e162a26f23bcbf8a4 \
+    --hash=sha256:73448f0aacd8d0808895bdc4b2c8e01a8d67646e4188f887375398c761f340fd
+    # via cyclonedx-python-lib
+markdown-it-py==4.2.0 \
+    --hash=sha256:04a21681d6fbb623de53f6f364d352309d4094dd4194040a10fd51833e418d49 \
+    --hash=sha256:9f7ebbcd14fe59494226453aed97c1070d83f8d24b6fc3a3bcf9a38092641c4a
+    # via rich
+mdurl==0.1.2 \
+    --hash=sha256:84008a41e51615a49fc9966191ff91509e3c40b939176e643fd50a5c2196b8f8 \
+    --hash=sha256:bb413d29f5eea38f31dd4754dd7377d4465116fb207585f97bf925588687c1ba
+    # via markdown-it-py
+msgpack==1.2.1 \
+    --hash=sha256:01e2dd6c9b19d333a00282330cc8a73d38d8dabc306dc5b42cd668c3ac82e833 \
+    --hash=sha256:020e881a764b20d8d7ca1a54fc01b8175519d108e3c3f194fddc200bda95951a \
+    --hash=sha256:04c721c2c7448767e9e3f2520a475663d8ee0f09c31890f6d2bd70fd636a9647 \
+    --hash=sha256:05f340e47e7e47d2da8db9b53e1bb1d294369e9ef45a747441309f6650b8351d \
+    --hash=sha256:0a70e3cf2804a300d921bb0940426e35f4e489a23adfb77a808892241db0a064 \
+    --hash=sha256:0adcf06ffde0777c0e1a9b771a2b1c4226ba1bbf748c8efcc02fcdeca3299107 \
+    --hash=sha256:0c0d9802354507bcba62af19c17918e3eb437cc25e6f50657d511b5856a77aac \
+    --hash=sha256:0e2bf9280bceb5efca998435904b5d3e9fdbcc11d90dc9df30aec7973252b720 \
+    --hash=sha256:1233ee2dd0cefba127583de50ea654677277047d238303521db35def3d7b2e7c \
+    --hash=sha256:146ee4e9ce80b365c6d4c47073da9da7bcec473e58194ceee5dd7620ace77e06 \
+    --hash=sha256:1548006a91aa93c5da81f3bdcebc1a0d10cea2d25969754fbe848da622b2b895 \
+    --hash=sha256:196300e7e5d6e74d50f1607ab9c06c4a1484c383cd22defd727902591f7e8dde \
+    --hash=sha256:1dabedcd0f23559f3596428c6589c1cd8c6eaed3a0d720795b07b0225d769203 \
+    --hash=sha256:20466cca18c49c7292a8984bc15d65857b171e7264bdcb5f96baf8be238791fc \
+    --hash=sha256:298872ecf9e61950f1c6af4ca969b859ee91783bb920ef6e6172697d0c8aad74 \
+    --hash=sha256:29a3f6e9667868429d8240dfd063ea5ffdc1321c13d783aa23827a38de0dcb22 \
+    --hash=sha256:2eda0b7ebb1283a98d3e4492ac933c8af6aff59fd3df1c3ed024f536af4b1dc8 \
+    --hash=sha256:2ef59c659f289eddf8aa6623823f19fa2f40a4029266889eac7a2505dd210c35 \
+    --hash=sha256:2ff164c1b0bcb740b073b99e945234d0212852fa378e44a208c425379140dbeb \
+    --hash=sha256:33f14fba63278b714efe6ad07e50ea5f03d91537aa6a1c5f1ceca4cf44013ca9 \
+    --hash=sha256:350cb813d0af6e65d2f7ef0d729f7ff5be5a8bce03665892f43e5883d4ecc1b8 \
+    --hash=sha256:4202c74688ca06591f78cb18988228bd4cca2cc75d57b60008372892d2f1e6e6 \
+    --hash=sha256:4227224aaec8f7fbcbfbd4272319347b2bb4030366502600f8c45588c5187b07 \
+    --hash=sha256:491cc39455ca765fad51fb451bf2915eb2cf41192ab5801ce8d67c1d614fe056 \
+    --hash=sha256:575957e79cd51903a4e8495a242442949641e08f1efd5197b43bebd3ea7682b4 \
+    --hash=sha256:5ad5467fc3f68b5468e06c5f788d712e9f8ffc8b0cd1bcb160c105c1ee92dae7 \
+    --hash=sha256:5bb9c386f0a329c035ddbab4b72d1028bf9627add8dda41070288563d57ed1b1 \
+    --hash=sha256:5c24aa15d5963051e1a5c62b12c50cd705992502b5ec1f3bece6046f33c9fc24 \
+    --hash=sha256:5f6277e5f783c36786a145e0247fc189a03f35f84b251646e53592d2bc12b355 \
+    --hash=sha256:60926b75d00c8e816ef98f3034f484a8bc64242d66839cef4cf7e503142316a0 \
+    --hash=sha256:633727297ed063441fd1cda2288865487f33ad14eeb8831afb5f0c396a62cfce \
+    --hash=sha256:67f6dd22fa72a93752643f07889796d62739a13415ee630169a8ce764f86cf9f \
+    --hash=sha256:6d09badf350af2be9d189184e04e64cf54ad93569ab3d96fca58bd3e84aad707 \
+    --hash=sha256:6ee967f7c7e1df2890c671ff2ee51a28ded0efc95da3e507176dee881ce36c66 \
+    --hash=sha256:74847557e28ce71bd3c438a447ca90e4b507e997ddbdef8a12a7b283b86c156b \
+    --hash=sha256:779197a6513bab3c3632265e3d0f7cb3227e62510841a6f34f1eaa37efbb345e \
+    --hash=sha256:787c9bebb5833e8f6fc8abca3c0597683d8d87f56a8842b6b89c75a5f3176e2d \
+    --hash=sha256:7d31c0ac0c640f877804c67cb2bc9f4e23dc2db97e96c2e67fa27d38283b41f8 \
+    --hash=sha256:810b916696c86ef0deb3b74588480224df4c1b071136c34183e4a2a4284d7ac7 \
+    --hash=sha256:83efa1c898e0fc5380fc0cabbf75164c52e3b5cbb45973710d75821928380c73 \
+    --hash=sha256:85f57e960d877f2977f6430896191b04a21f8901b3b4baf2e4604329f4db5402 \
+    --hash=sha256:8b267ce94efb76fbd1b3373511420074ee3187f0f7811bf394531de13294735a \
+    --hash=sha256:8c2ed1e48cc0f460bf3c7780e7137ff21a4e18433451916f2442c1b21036cd7d \
+    --hash=sha256:8c7b398c56ff125feae96c2737abfec5595f1fa0aa186df60c56040b8accb95c \
+    --hash=sha256:8d00f177ca88a77c1cf848d204a38f249751650b601cb6532acc68805d8a8273 \
+    --hash=sha256:8ff92d7feeaf5bc26c51495b69e2f99ed97ab79346fb6555f44be7dd2ac6503b \
+    --hash=sha256:91054a783328e0ea7954b8771095705c8d2243b814743fbaadf14552c9c52c5d \
+    --hash=sha256:98b58bdb89c46190e4609bb36abe17c6d4105ad13f9c5f8f6f64d320f8ced3fb \
+    --hash=sha256:a28d076ca7c82b9c8728ad90b7147489449557038bed50e4241eb832395169b4 \
+    --hash=sha256:aa6c4be5d1c02a42b066ca6ddb71adf36432868fdcdb6ee87e634e86e0674190 \
+    --hash=sha256:aded5bdf32609dc7987a49bbbd15a8ef096193f96dd8bbeb791de729e650acf5 \
+    --hash=sha256:afc5febcd4c99effbc02b528e49d6fd0760b2b7d48c05239e345a5fa6e743d9a \
+    --hash=sha256:b50b727bd652bdc37d950336c848ef20ec54a4cafc38dce19b1cd86ad625d0f7 \
+    --hash=sha256:c1c79a604a2969a868a78b6ebd27a887e00c624f14f66b3038e0590cb23332d1 \
+    --hash=sha256:ca0dacff965c47afdc3749a8469d7302a8f801d6a28758d55120d75e66ce6889 \
+    --hash=sha256:d3567748a5107cb40cdf66a275430c2f87c07777698f4bfd25c35f44d533258c \
+    --hash=sha256:dc871b997a9370d855b7394465f2f350e847a5b806dd38dcc9c989e7d87da155 \
+    --hash=sha256:dd3bfe82d53edfe4b7fc9a7ec9761e23a7a5b1dac22264505af428253c29ed24 \
+    --hash=sha256:e3dc2feb0876209d9c38aa56cb1de169bd6c4348f1aa48271f241226590993e6 \
+    --hash=sha256:e4f1d0f8f98ade9634e01fb704a408f9336c0a8f1117b369f5db83dc7551d8b1 \
+    --hash=sha256:ec0e675d59150a6269ddc9139087c722292664a37d071a849c05c473350f1f2d \
+    --hash=sha256:ee1d9ed27d0497b848923746cf762ed2e7db24f4be7eec8e5cbe8c766aa707b7 \
+    --hash=sha256:f02cf17a6ca1abe29b5f980644f7551f94d71f2011509b26d8625ce038f0df64 \
+    --hash=sha256:f12038a35fabd52e56a3547bab42401af49a45caa6dd00b34c44de235bc93ee2 \
+    --hash=sha256:f310233ef7fb9c14e201c93639fe5f5260b005f56f0b29048e999c30935596cc \
+    --hash=sha256:f9389552ecf4784886345ead0647e4edc96bee37cbab05b75540f542f766c48c
+    # via cachecontrol
+packageurl-python==0.17.6 \
+    --hash=sha256:1252ce3a102372ca6f86eb968e16f9014c4ba511c5c37d95a7f023e2ca6e5c25 \
+    --hash=sha256:31a85c2717bc41dd818f3c62908685ff9eebcb68588213745b14a6ee9e7df7c9
+    # via cyclonedx-python-lib
+packaging==26.2 \
+    --hash=sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e \
+    --hash=sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661
+    # via
+    #   pip-audit
+    #   pip-requirements-parser
+pip-api==0.0.34 \
+    --hash=sha256:8b2d7d7c37f2447373aa2cf8b1f60a2f2b27a84e1e9e0294a3f6ef10eb3ba6bb \
+    --hash=sha256:9b75e958f14c5a2614bae415f2adf7eeb54d50a2cfbe7e24fd4826471bac3625
+    # via pip-audit
+pip-audit==2.10.1 \
+    --hash=sha256:1eb4565d19ebe5d48996f4b770b4d2b32887e12cb12cfa637f1a064011b55ffc \
+    --hash=sha256:99ef3f600a317c1945f1e89e227ef26e1c2d618429b8bd3fa6f4f7c440c4611a
+    # via -r audit.in
+pip-requirements-parser==32.0.1 \
+    --hash=sha256:4659bc2a667783e7a15d190f6fccf8b2486685b6dba4c19c3876314769c57526 \
+    --hash=sha256:b4fa3a7a0be38243123cf9d1f3518da10c51bdb165a2b2985566247f9155a7d3
+    # via pip-audit
+platformdirs==4.10.0 \
+    --hash=sha256:31e761a6a0ca04faf7353ea759bdba55652be214725111e5aac52dfa29d4bef7 \
+    --hash=sha256:fb516cdb12eb0d857d0cd85a7c57cea4d060bee4578d6cf5a14dfdf8cbf8784a
+    # via pip-audit
+py-serializable==2.1.0 \
+    --hash=sha256:9d5db56154a867a9b897c0163b33a793c804c80cee984116d02d49e4578fc103 \
+    --hash=sha256:b56d5d686b5a03ba4f4db5e769dc32336e142fc3bd4d68a8c25579ebb0a67304
+    # via cyclonedx-python-lib
+pygments==2.20.0 \
+    --hash=sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f \
+    --hash=sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176
+    # via rich
+pyparsing==3.3.2 \
+    --hash=sha256:850ba148bd908d7e2411587e247a1e4f0327839c40e2e5e6d05a007ecc69911d \
+    --hash=sha256:c777f4d763f140633dcb6d8a3eda953bf7a214dc4eff598413c070bcdc117cbc
+    # via pip-requirements-parser
+requests==2.34.2 \
+    --hash=sha256:2a0d60c172f83ac6ab31e4554906c0f3b3588d37b5cb939b1c061f4907e278e0 \
+    --hash=sha256:f288924cae4e29463698d6d60bc6a4da69c89185ad1e0bcc4104f584e960b9ed
+    # via
+    #   cachecontrol
+    #   pip-audit
+rich==15.0.0 \
+    --hash=sha256:33bd4ef74232fb73fe9279a257718407f169c09b78a87ad3d296f548e27de0bb \
+    --hash=sha256:edd07a4824c6b40189fb7ac9bc4c52536e9780fbbfbddf6f1e2502c31b068c36
+    # via pip-audit
+sortedcontainers==2.4.0 \
+    --hash=sha256:25caa5a06cc30b6b83d11423433f65d1f9d76c4c6a0c90e3379eaa43b9bfdb88 \
+    --hash=sha256:a163dcaede0f1c021485e957a39245190e74249897e2ae4b2aa38595db237ee0
+    # via cyclonedx-python-lib
+tomli==2.4.1 \
+    --hash=sha256:01f520d4f53ef97964a240a035ec2a869fe1a37dde002b57ebc4417a27ccd853 \
+    --hash=sha256:0d85819802132122da43cb86656f8d1f8c6587d54ae7dcaf30e90533028b49fe \
+    --hash=sha256:136443dbd7e1dee43c68ac2694fde36b2849865fa258d39bf822c10e8068eac5 \
+    --hash=sha256:1d8591993e228b0c930c4bb0db464bdad97b3289fb981255d6c9a41aedc84b2d \
+    --hash=sha256:2190f2e9dd7508d2a90ded5ed369255980a1bcdd58e52f7fe24b8162bf9fedbd \
+    --hash=sha256:2c1c351919aca02858f740c6d33adea0c5deea37f9ecca1cc1ef9e884a619d26 \
+    --hash=sha256:36d2bd2ad5fb9eaddba5226aa02c8ec3fa4f192631e347b3ed28186d43be6b54 \
+    --hash=sha256:3d48a93ee1c9b79c04bb38772ee1b64dcf18ff43085896ea460ca8dec96f35f6 \
+    --hash=sha256:47149d5bd38761ac8be13a84864bf0b7b70bc051806bc3669ab1cbc56216b23c \
+    --hash=sha256:4ab97e64ccda8756376892c53a72bd1f964e519c77236368527f758fbc36a53a \
+    --hash=sha256:4b605484e43cdc43f0954ddae319fb75f04cc10dd80d830540060ee7cd0243cd \
+    --hash=sha256:504aa796fe0569bb43171066009ead363de03675276d2d121ac1a4572397870f \
+    --hash=sha256:51529d40e3ca50046d7606fa99ce3956a617f9b36380da3b7f0dd3dd28e68cb5 \
+    --hash=sha256:52c8ef851d9a240f11a88c003eacb03c31fc1c9c4ec64a99a0f922b93874fda9 \
+    --hash=sha256:559db847dc486944896521f68d8190be1c9e719fced785720d2216fe7022b662 \
+    --hash=sha256:5a881ab208c0baf688221f8cecc5401bd291d67e38a1ac884d6736cbcd8247e9 \
+    --hash=sha256:5cb41aa38891e073ee49d55fbc7839cfdb2bc0e600add13874d048c94aadddd1 \
+    --hash=sha256:5e262d41726bc187e69af7825504c933b6794dc3fbd5945e41a79bb14c31f585 \
+    --hash=sha256:5ee18d9ebdb417e384b58fe414e8d6af9f4e7a0ae761519fb50f721de398dd4e \
+    --hash=sha256:7008df2e7655c495dd12d2a4ad038ff878d4ca4b81fccaf82b714e07eae4402c \
+    --hash=sha256:734e20b57ba95624ecf1841e72b53f6e186355e216e5412de414e3c51e5e3c41 \
+    --hash=sha256:7c7e1a961a0b2f2472c1ac5b69affa0ae1132c39adcb67aba98568702b9cc23f \
+    --hash=sha256:7f86fd587c4ed9dd76f318225e7d9b29cfc5a9d43de44e5754db8d1128487085 \
+    --hash=sha256:7f94b27a62cfad8496c8d2513e1a222dd446f095fca8987fceef261225538a15 \
+    --hash=sha256:88dceee75c2c63af144e456745e10101eb67361050196b0b6af5d717254dddf7 \
+    --hash=sha256:8a650c2dbafa08d42e51ba0b62740dae4ecb9338eefa093aa5c78ceb546fcd5c \
+    --hash=sha256:8d65a2fbf9d2f8352685bc1364177ee3923d6baf5e7f43ea4959d7d8bc326a36 \
+    --hash=sha256:96481a5786729fd470164b47cdb3e0e58062a496f455ee41b4403be77cb5a076 \
+    --hash=sha256:a120733b01c45e9a0c34aeef92bf0cf1d56cfe81ed9d47d562f9ed591a9828ac \
+    --hash=sha256:b1d22e6e9387bf4739fbe23bfa80e93f6b0373a7f1b96c6227c32bef95a4d7a8 \
+    --hash=sha256:b8c198f8c1805dc42708689ed6864951fd2494f924149d3e4bce7710f8eb5232 \
+    --hash=sha256:c2541745709bad0264b7d4705ad453b76ccd191e64aa6f0fc66b69a293a45ece \
+    --hash=sha256:c742f741d58a28940ce01d58f0ab2ea3ced8b12402f162f4d534dfe18ba1cd6a \
+    --hash=sha256:c7f2c7f2b9ca6bdeef8f0fa897f8e05085923eb091721675170254cbc5b02897 \
+    --hash=sha256:d312ef37c91508b0ab2cee7da26ec0b3ed2f03ce12bd87a588d771ae15dcf82d \
+    --hash=sha256:d4d8fe59808a54658fcc0160ecfb1b30f9089906c50b23bcb4c69eddc19ec2b4 \
+    --hash=sha256:da25dc3563bff5965356133435b757a795a17b17d01dbc0f42fb32447ddfd917 \
+    --hash=sha256:eab21f45c7f66c13f2a9e0e1535309cee140182a9cdae1e041d02e47291e8396 \
+    --hash=sha256:eb0dc4e38e6a1fd579e5d50369aa2e10acfc9cace504579b2faabb478e76941a \
+    --hash=sha256:ec9bfaf3ad2df51ace80688143a6a4ebc09a248f6ff781a9945e51937008fcbc \
+    --hash=sha256:ede3e6487c5ef5d28634ba3f31f989030ad6af71edfb0055cbbd14189ff240ba \
+    --hash=sha256:f3c6818a1a86dd6dca7ddcaaf76947d5ba31aecc28cb1b67009a5877c9a64f3f \
+    --hash=sha256:f758f1b9299d059cc3f6546ae2af89670cb1c4d48ea29c3cacc4fe7de3058257 \
+    --hash=sha256:f8f0fc26ec2cc2b965b7a3b87cd19c5c6b8c5e5f436b984e85f486d652285c30 \
+    --hash=sha256:fd0409a3653af6c147209d267a0e4243f0ae46b011aa978b1080359fddc9b6cf \
+    --hash=sha256:ff18e6a727ee0ab0388507b89d1bc6a22b138d1e2fa56d1ad494586d61d2eae9 \
+    --hash=sha256:ff2983983d34813c1aeb0fa89091e76c3a22889ee83ab27c5eeb45100560c049
+    # via pip-audit
+tomli-w==1.2.0 \
+    --hash=sha256:188306098d013b691fcadc011abd66727d3c414c571bb01b1a174ba8c983cf90 \
+    --hash=sha256:2dd14fac5a47c27be9cd4c976af5a12d87fb1f0b4512f81d69cce3b35ae25021
+    # via pip-audit
+typing-extensions==4.16.0 \
+    --hash=sha256:481caa481374e813c1b176ada14e97f1f67a4539ce9cfeb3f350d78d6370c2e8 \
+    --hash=sha256:dc983d19a509c94dba722ee6abd33940f7c05a89e243c47e907eb4db6f1a43e5
+    # via cyclonedx-python-lib
+urllib3==2.7.0 \
+    --hash=sha256:231e0ec3b63ceb14667c67be60f2f2c40a518cb38b03af60abc813da26505f4c \
+    --hash=sha256:9fb4c81ebbb1ce9531cce37674bbc6f1360472bc18ca9a553ede278ef7276897
+    # via requests
+
+# The following packages are considered to be unsafe in a requirements file:
+pip==26.1.2 \
+    --hash=sha256:382ff9f685ee3bc25864f820aa50505825f10f5458ffff07e30a6d96e5715cab \
+    --hash=sha256:f49cd134c61cf2fd75e0ce2676db03e4054504a5a4986d00f8299ae632dc4605
+    # via pip-api
@@ -0,0 +1,17 @@
+# `build` for the publish.yml `build` job, hash-pinned (Scorecard:
+# Pinned-Dependencies).
+#
+# Target env: ubuntu-latest, Python 3.13 ONLY.
+# That narrowness is load-bearing — build's marker-gated deps are all excluded
+# at this target and therefore absent from build.txt:
+#   colorama          ; os_name == "nt"            -> posix runner, not needed
+#   tomli             ; python_version < "3.11"    -> 3.13, not needed
+#   importlib-metadata; python_full_version < "3.10.2" -> 3.13, not needed
+# If publish.yml ever gains a Windows runner or a <3.11 Python, regenerate this
+# file on that target (or add the dep explicitly) or --require-hashes will fail
+# with "all requirements must have their versions pinned".
+# See e2e.in for the cross-OS variant that handles this.
+#
+# Regenerate:
+#   pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=build.txt build.in
+build==1.5.0
@@ -0,0 +1,18 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=build.txt build.in
+#
+build==1.5.0 \
+    --hash=sha256:13f3eecb844759ab66efec90ca17639bbf14dc06cb2fdf37a9010322d9c50a6f \
+    --hash=sha256:302c22c3ba2a0fd5f3911918651341ebb3896176cbdec15bd421f80b1afc7647
+    # via -r build.in
+packaging==26.2 \
+    --hash=sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e \
+    --hash=sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661
+    # via build
+pyproject-hooks==1.2.0 \
+    --hash=sha256:1e859bd5c40fae9448642dd871adf459e5e2084186e8d2c2a79a824c970da1f8 \
+    --hash=sha256:9e5c6bfa8dcc30091c74b0cf803c81fdd29d94f01992a7707bc97babb1141913
+    # via build
@@ -0,0 +1,26 @@
+# Outer build tooling for e2e-wheel.yml, hash-pinned (Scorecard:
+# Pinned-Dependencies).
+#
+# Target env: ubuntu-latest + windows-latest + macos-latest, Python 3.12.
+# conftest.py builds the wheel + a clean venv internally; the outer env only
+# needs build + pytest.
+#
+# WHY colorama IS LISTED EXPLICITLY (do not "clean up"):
+# both build and pytest depend on colorama behind a platform marker
+# (`os_name == "nt"` / `sys_platform == "win32"`). pip-compile evaluates markers
+# against the machine it runs on, so compiling on Linux DROPS colorama from the
+# lock — and the windows-latest leg then dies under --require-hashes with
+# "In --require-hashes mode, all requirements must have their versions pinned".
+# Listing it as a direct requirement forces it into the lock with hashes.
+# colorama is a pure-python universal wheel (py2.py3-none-any, zero deps), so
+# installing it on the Linux/macOS legs is inert.
+#
+# Python 3.12 is likewise load-bearing: pytest's `exceptiongroup`/`tomli` and
+# build's `tomli` are gated on python_version < "3.11" and are absent here. If
+# the matrix ever drops below 3.11, regenerate on that target.
+#
+# Regenerate (on Linux, Python 3.12):
+#   pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=e2e.txt e2e.in
+build==1.5.0
+pytest==9.1.1
+colorama==0.4.6
@@ -0,0 +1,40 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=e2e.txt e2e.in
+#
+build==1.5.0 \
+    --hash=sha256:13f3eecb844759ab66efec90ca17639bbf14dc06cb2fdf37a9010322d9c50a6f \
+    --hash=sha256:302c22c3ba2a0fd5f3911918651341ebb3896176cbdec15bd421f80b1afc7647
+    # via -r e2e.in
+colorama==0.4.6 \
+    --hash=sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44 \
+    --hash=sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6
+    # via -r e2e.in
+iniconfig==2.3.0 \
+    --hash=sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730 \
+    --hash=sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12
+    # via pytest
+packaging==26.2 \
+    --hash=sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e \
+    --hash=sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661
+    # via
+    #   build
+    #   pytest
+pluggy==1.6.0 \
+    --hash=sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3 \
+    --hash=sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746
+    # via pytest
+pygments==2.20.0 \
+    --hash=sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f \
+    --hash=sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176
+    # via pytest
+pyproject-hooks==1.2.0 \
+    --hash=sha256:1e859bd5c40fae9448642dd871adf459e5e2084186e8d2c2a79a824c970da1f8 \
+    --hash=sha256:9e5c6bfa8dcc30091c74b0cf803c81fdd29d94f01992a7707bc97babb1141913
+    # via build
+pytest==9.1.1 \
+    --hash=sha256:1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313 \
+    --hash=sha256:37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c
+    # via -r e2e.in
@@ -0,0 +1,15 @@
+# ruff for the ci.yml `lint` job, hash-pinned (Scorecard: Pinned-Dependencies).
+#
+# Target env: ubuntu-latest, Python 3.13. ruff has no dependencies, and
+# --generate-hashes emits every PyPI file hash for the pinned version (all 18
+# platform wheels + sdist), so this lock stays valid if lint ever runs on
+# another OS/arch.
+#
+# 0.15.21 == what the previous unpinned `pip install ruff` resolved to on
+# 2026-07-15, so pinning is a no-op for lint results today. Bumping this file
+# can surface new lint rules — that is the intended, reviewable trade-off.
+# Keep compatible with pyproject.toml's dev extra (`ruff>=0.11`).
+#
+# Regenerate:
+#   pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=lint.txt lint.in
+ruff==0.15.21
@@ -0,0 +1,26 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=lint.txt lint.in
+#
+ruff==0.15.21 \
+    --hash=sha256:00eca240af5789fec6fe7df74c088cc1f9644ed83027113468efba7c92b94075 \
+    --hash=sha256:01d65b4831c6b2a4ba8ee6faa84049d44d982b7a706e622c4094c509e51673be \
+    --hash=sha256:01f8d5be84823c172b389e123174f781f9daf86d6c58719d603f941932195cdd \
+    --hash=sha256:0f212c5d7d54c01bbfe6dcab02b724a39300f3e34ed7acbe995ccb320a2c58bd \
+    --hash=sha256:16d090c0740916594157e75b80d666eab8e78083b39b3b0e1d698f4670a17b86 \
+    --hash=sha256:262ab31557a75141325e32d3357f3597645a7f084e732b6b054dde428ecd9341 \
+    --hash=sha256:2c5a913a589120ce67933d5d05fd6ddbcc2481c6a054980ee767f7414c72b4fd \
+    --hash=sha256:3a10e74757dd65004d779b73e2f3c5210156d9980b41224d50d2ebcf1db51e67 \
+    --hash=sha256:5ef04b681d02ad4dc9620f00f83ac5c22f652d0e9a9cfe431d219b16ad5ccc41 \
+    --hash=sha256:63ea0e965e5d73c90e95b2434beeafc70820536717f561b32ab6e777cb9bdf5d \
+    --hash=sha256:659c4e7a4212f83306045ec7c5e5a356d16d9a6ef4ae0c7a4d872914fc655d9d \
+    --hash=sha256:6e83115d4b9377c1cbc13abf0e051f069fab0ef815ea0504a8a008cee24dd0a8 \
+    --hash=sha256:9e866eab611a5f959d36df2d10e446973a3610bc42b0c15b31dc27977d59c233 \
+    --hash=sha256:bab0905d2f29e0d9fbc3c373ed23db0095edaa3f71f1f4f519ec15134d9e85c8 \
+    --hash=sha256:d0cfc841c572283c36548f82664a54ce6565567f1b0d5b4cf2caac693d8b7500 \
+    --hash=sha256:d4b8d9a2f0f12b816b50447f6eccb9f4bb01a6b82c86b50fb3b5354b458dc6d3 \
+    --hash=sha256:e6312e41bc96791299614995ea3a977c5857c3b5662b1ecef6755b02b87cb646 \
+    --hash=sha256:e89bc93c0d3803ba870b55c29671bad9dc6d94bb1eb181b056b52eb05b52854f
+    # via -r lint.in
@@ -0,0 +1,13 @@
+# pip self-upgrade, hash-pinned (OpenSSF Scorecard: Pinned-Dependencies).
+#
+# Replaces `python -m pip install --upgrade pip` in ci.yml, security.yml and
+# e2e-wheel.yml. pip has no runtime dependencies, so this lock is inherently
+# portable across every OS and Python in the CI matrix (3.10-3.13).
+#
+# 26.1.2 is deliberate, not merely "latest": security.yml's pip-audit scans the
+# whole environment and the runner's bundled pip (26.1.1) carries PYSEC-2026-196
+# (fixed in 26.1.2). Do not pin below 26.1.2 — audit will red.
+#
+# Regenerate:
+#   pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=pip.txt pip.in
+pip==26.1.2
@@ -0,0 +1,12 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --allow-unsafe --generate-hashes --no-emit-index-url --output-file=pip.txt pip.in
+#
+
+# The following packages are considered to be unsafe in a requirements file:
+pip==26.1.2 \
+    --hash=sha256:382ff9f685ee3bc25864f820aa50505825f10f5458ffff07e30a6d96e5715cab \
+    --hash=sha256:f49cd134c61cf2fd75e0ce2676db03e4054504a5a4986d00f8299ae632dc4605
+    # via -r pip.in
@@ -20,7 +20,9 @@ jobs:
      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
        with:
          python-version: "3.13"
-      - run: pip install ruff
+      # Hash-pinned tool install (Scorecard: Pinned-Dependencies). Pins ruff to
+      # the version this lint gate is known-green against; see .github/requirements/lint.in.
+      - run: python -m pip install --require-hashes -r .github/requirements/lint.txt
      - run: ruff check src/ tests/
      - run: ruff format --check src/ tests/

@@ -36,7 +38,7 @@ jobs:
        with:
          python-version: ${{ matrix.python-version }}
      - run: |
-          python -m pip install --upgrade pip
+          python -m pip install --require-hashes -r .github/requirements/pip.txt
          pip install -e ".[dev]"
      # tests/e2e build a wheel + clean venv per the dedicated e2e-wheel.yml
      # workflow — they are not part of the fast unit lane.
@@ -57,7 +59,7 @@ jobs:
        with:
          python-version: "3.13"
      - run: |
-          python -m pip install --upgrade pip
+          python -m pip install --require-hashes -r .github/requirements/pip.txt
          # Install the `memory` extra (numpy/chromadb/fastembed) alongside dev:
          # the diagnostics standard runs pyright over every branch, and memory's
          # handlers import chromadb/numpy. Without these deps installed, pyright
@@ -79,7 +81,7 @@ jobs:
        with:
          python-version: "3.13"
      - run: |
-          python -m pip install --upgrade pip
+          python -m pip install --require-hashes -r .github/requirements/pip.txt
          pip install -e ".[dev]"
      - run: coverage run -m pytest --rootdir=. --ignore=tests/e2e
      - run: coverage xml
@@ -47,7 +47,14 @@ jobs:
          python-version: ${{ matrix.python-version }}

      - name: Install build tooling
-        run: python -m pip install --upgrade pip build pytest
+        # Hash-pinned (Scorecard: Pinned-Dependencies). e2e.txt carries colorama
+        # explicitly — build/pytest need it only on Windows (os_name == "nt" /
+        # sys_platform == "win32"), and a Linux-generated lock would otherwise
+        # omit it and break the windows-latest leg under --require-hashes.
+        # See .github/requirements/e2e.in.
+        run: |
+          python -m pip install --require-hashes -r .github/requirements/pip.txt
+          python -m pip install --require-hashes -r .github/requirements/e2e.txt

      - name: Run cross-OS e2e wiring harness
        # conftest.py builds the wheel + clean venv internally; the outer env
@@ -11,13 +11,34 @@ permissions:
 jobs:
  build:
    runs-on: ubuntu-latest
+    # Job-level permissions REPLACE the top-level block rather than merge with
+    # it, so `contents: read` is restated here on purpose — dropping it would
+    # break actions/checkout. id-token/attestations are what the provenance
+    # attestation below needs (Scorecard: Signed-Releases).
+    permissions:
+      contents: read
+      id-token: write
+      attestations: write
    steps:
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
        with:
          python-version: "3.13"
-      - run: pip install build
+      # Hash-pinned tool install (Scorecard: Pinned-Dependencies).
+      - run: python -m pip install --require-hashes -r .github/requirements/build.txt
      - run: python -m build
+      - name: Attest build provenance
+        # Signs a provenance statement binding these exact sdist/wheel digests to
+        # this workflow run, via the same keyless Sigstore/OIDC path as the
+        # release signing below. Runs after the artifacts exist and before they
+        # leave the job, so the attested digests are the published ones.
+        # NOTE: the bundle is deliberately NOT written into dist/ — the publish
+        # job feeds dist/* to gh-action-pypi-publish, which rejects any file that
+        # is not a distribution. See the report note on attaching provenance to
+        # the GitHub Release.
+        uses: actions/attest-build-provenance@0f67c3f4856b2e3261c31976d6725780e5e4c373 # v4.1.1
+        with:
+          subject-path: "dist/*"
      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: dist
@@ -27,9 +27,11 @@ jobs:
      # 26.1.2). Upgrading removes the vulnerable version outright rather than
      # suppressing it — and 26.1.2 also resolves CVE-2026-3219 / CVE-2026-6357,
      # which is why those two stale --ignore-vuln entries are no longer needed.
+      # Both installs are hash-pinned (Scorecard: Pinned-Dependencies); pip.txt
+      # holds the >=26.1.2 floor the comment above requires.
      - run: |
-          python -m pip install --upgrade pip
-          pip install pip-audit
+          python -m pip install --require-hashes -r .github/requirements/pip.txt
+          python -m pip install --require-hashes -r .github/requirements/audit.txt
      - run: pip install -e .
      - name: Pip audit
        run: pip-audit --skip-editable
@@ -9,6 +9,547 @@ PyPI version — not the changelog header.

 ---

+## [2026-07-17]
+
+**v2.7.2** — everything merged since v2.7.1, headlined by the compass decision
+engine v2: curation with supersedes links + write-time conflict advisories
+(Track 1), and ambient recall — rated past decisions now surface verbatim into
+live sessions on matching prompts, governed by session caps and spacing
+(Track 2). Also in this release: the plan close pipeline completes itself
+(auto-vectorization + crash-safe registry writes), drone's 3-layer subprocess
+timeout policy with a collision-free `--drone-timeout` flag, plan-number memory
+search that pins the exact plan, a fleet-wide seedgo 100% restoration, and
+SSH-signed commits now verifying on GitHub. Details in the sections below
+(2026-07-16 carries the full stories).
+
+### Changed
+
+- **Release cadence ruling: every dev→main merge ships a PATCH bump + tag by
+  default.** PyPI tracks main, always current; version numbers carry no
+  significance during beta — the big jump is reserved for beta exit.
+
+- **Merge playbook SOP refined from live run PPLAN-0010.** The raw
+  `git fetch origin main:main` step (now blocked by the git gate) is replaced
+  with `drone @git sync` in both places it appeared, and the template opens
+  with the exact create command (`drone @flow create . "Merge summary" merge
+  pplan` — template name before type), closing the trap where the wrong arg
+  order silently stamps the default template.
+
+## [2026-07-16]
+
+### Added
+
+- **Compass ambient recall — Track 2 (DPLAN-0246/FPLAN-0332): rated decisions
+  surface unprompted.** On every user prompt, a new hooks handler
+  (`compass_recall`, registered in `.aipass/hooks.json` only) queries compass
+  FTS with the prompt text and injects matching rulings VERBATIM —
+  `[BAD] #56: <decision text>` — tidbits, never vibes. Three branches, one
+  pipeline, each piece behind a modules/-boundary API: devpulse's
+  `recall_decisions()` (side-effect-free scored candidates; rare-token
+  evidence scoring + a query-side stopword filter so greeting/filler words
+  can't fake relevance) + `mark_surfaced()` (counts only real injections);
+  @memory's pure `should_surface()` governance (promoted from the dormant
+  symbolic engine: threshold, 5/session cap, 10-message spacing — first
+  surface exempt, 300s cooldown, dedup; state-in/state-out, caller persists);
+  @hooks' 90-line handler + engine per-handler budget (errors never block a
+  prompt — `compass_recall_unreachable` log signature for @trigger's watcher).
+  Live acceptance matrix through the real bridge: topic-with-history prompts
+  recall the right ruling (a CI prompt surfaced the red-CI-never-parked
+  ruling), small talk and greetings stay silent, repeat prompts gate on
+  spacing. Review caught and fixed pre-ship: wrong payload key (`userInput` →
+  `prompt`), phantom `CLAUDE_CODE_SESSION_ID` env (session id is
+  stdin-payload-only), spacing gate blocking the first surface, and a trust
+  registry re-enrollment gap that silently disabled ALL project hooks for 20
+  minutes after the hooks.json edit. 446 devpulse + 1011 memory + 1129 hooks
+  tests green; seedgo 31/31 on every touched module.
+
+- **Compass curation v2 Track 1 (DPLAN-0246/FPLAN-0331): supersedes links +
+  write-time conflict check.** A correcting compass entry now archives and
+  links what it replaces in one transaction (`compass add --supersedes N`);
+  query renders both directions ("supersedes #N" / "ARCHIVED — superseded by
+  #M") so a retracted decision can never masquerade as current truth. Every
+  `compass add` FTS-checks the new text against active entries and prints a
+  non-blocking "possible conflict with #X" advisory — flag-and-ask, no LLM, no
+  auto-resolve (boardroom ruling). New `compass note <id>` command (FTS
+  re-index proven by test), `--include-archived` query flag (the avoid-list is
+  finally searchable), dead `score` column removed from all code surfaces
+  (kept inert on disk — zero migration risk). Idempotent PRAGMA-checked
+  migration ran clean on the production store (128 rows, no loss); the four
+  fresh-eyes-audit archive pairs got their links backfilled. /prep now runs
+  one `compass review` per session — curation living in a path that already
+  runs, the lesson of all three compass eras. 435 devpulse tests green,
+  seedgo 31/31 on both touched modules.
+
+- **Close pipeline completes itself (DPLAN-0245): auto-vectorization +
+  crash-safe registry writes + drone timeout policy.** Closing a plan now
+  produces all side effects from one command — `post_close_runner` invokes
+  @memory's plan intake directly after archival (detached, loud on failure,
+  drains any backlog it finds), so plans can no longer silently pile up
+  unvectorized. Plan registry saves (@flow `save_registry` + mbank
+  `save_flow_registry`) now use the O_EXCL lockfile + atomic
+  tempfile-and-replace pattern, closing the same lost-update race class fixed
+  earlier in CLOSED_PLANS. @drone gained a 3-layer timeout policy: per-command
+  overrides (`@memory process-plans` 120s, `@flow close` 90s), a `--timeout N`
+  flag, 30s default — replacing the flat 30s guillotine that killed legitimate
+  long commands mid-pipeline; the timeout error now says how to override.
+  Proven end-to-end live: one `drone @flow close` on a throwaway plan yielded
+  archive + vectors + ledger + registry with zero manual steps, and the
+  auto-trigger swept a pre-existing backlog file on its first run. 730 flow +
+  878 drone tests green, seedgo 100%.
+
+### Fixed
+
+- **CI seedgo gate back to 100% across all 17 branches.** The Track 2 compass
+  recall code left three branches at 99%: @hooks' compass_recall handler was
+  missing json_handler operation logging and had two silent catches (now
+  logged); @memory's governance module held its implementation in modules/
+  (moved to handlers/governance/engine.py with modules/governance.py as the
+  thin re-export — the cross-branch import path is unchanged and live-E2E
+  verified through the real bridge); devpulse's README test count had drifted
+  (309 → 348). Audits re-run per branch: 100% overall, all suites green.
+
+- **Plan-number memory search hits the exact plan.** Searching a plan ID
+  ('DPLAN-0244', 'fplan 0332' — any case, dash or space) now pins the exact
+  plan as the top result at 100%, via a metadata lookup on the vector store's
+  source-file field instead of embedding similarity (which treats all plan IDs
+  as near-identical strings and never surfaced the target). Patrick ruling:
+  searching a plan number must return that plan first. Semantic search quality
+  for normal queries is unchanged. Also purged 193 junk vectors — throwaway
+  probe/flaky test plans from scratchpad sessions (dv4 batch, probe_test_plan,
+  throwaway_e2e_proof) that had leaked into the store. 1011 memory tests green.
+
+- **drone --timeout collision: router flag swallowed module flags.** The
+  DPLAN-0245 subprocess-timeout flag consumed the first `--timeout` token
+  anywhere in argv, so module-level flags silently vanished — watchdog's
+  `--timeout 1800` never arrived and long watches died at the 600s default
+  (live repro x2). Drone's flag is now namespaced `--drone-timeout`; plain
+  `--timeout` passes through untouched to the target module, with a regression
+  test pinning the passthrough. Per-command overrides intact. 879 drone tests
+  green, seedgo 100%.
+
+- **@memory command routing eaten by the new governance module.** The
+  governance module shipped in Track 2 had the wrong `handle_command`
+  signature (`args: list` instead of `command: str, args: list`) and always
+  returned True, so auto-discovery routed EVERY @memory command through it
+  first — `drone @memory search` answered "governance: unknown command 's'".
+  Fixed to the standard signature returning False for commands not its own;
+  search verified live (135 results). Library modules must decline commands
+  they don't own or they silently hijack the whole CLI. 1011 memory tests
+  green, seedgo 31/31.
+
+## [2026-07-15]
+
+### Fixed
+
+- **Plan vectorization pipeline unwedged (DPLAN-0245): 57 closed plans were
+  silently missing from semantic memory since mid-June.** Vector IDs were pure
+  content hashes, so identical template boilerplate across different plans
+  produced duplicate IDs within one ChromaDB upsert — the store rejected the
+  entire batch, and the all-or-nothing intake retried the same failing batch
+  forever. Fixed in @memory: IDs are now salted with the source filename when
+  present (rollover hashes unchanged — no re-vectorization churn), in-batch
+  dedup as a safety net, and `process_plans()` now runs per-file with the
+  manifest saved after each success so a poison file can never wedge the queue
+  again. Backlog drained and verified: 229/229 archived plans vectorized, 1112
+  chunks, formerly-lost plans answering semantic queries at 85%+ similarity.
+  990 memory tests green.
+
+- **CLOSED_PLANS ledger append race (@flow): concurrent plan closes lost
+  entries.** `append_to_closed_plans` was an unlocked read-modify-write; the
+  S314 bulk sweep lost 18 of 21 entries to it (reconciled by hand). Now guarded
+  by an `O_CREAT|O_EXCL` lockfile with retry/backoff, and the previously
+  silent append failure is surfaced in close output and logs. 730 flow tests
+  green.
+
+- **Telegram routine read-timeouts no longer logged as errors (@skills,
+  Patrick ruling): ends the medic wake-loop.** A routine long-poll read
+  timeout (`socket.timeout` — an `OSError` subclass) slipped past the earlier
+  `URLError`-only guard into the network-outage path, logging ERROR once per
+  episode (~576 lines/30h) and waking @trigger's medic each time. The
+  `_is_routine_read_timeout` guard now covers the `OSError` handler too, and
+  the genuine-outage episode-start line is demoted ERROR→WARNING (backoff
+  self-heals; recovery already logs INFO; medic only fires on ERROR/CRITICAL).
+  Real failures still log ERROR. 825 telegram tests green.
+
+### Security
+
+- **Hook config trust model hardening (DPLAN-0244): closes a zero-interaction
+  RCE from untrusted `.aipass/hooks.json`.** The hook loader walked up from CWD
+  and trusted any `.aipass/hooks.json` it found; since the bridge is wired
+  globally in provider settings, a hostile repo shipping a `command`-type hook
+  could execute arbitrary shell on `SessionStart` with no user interaction.
+  Fixed with defense-in-depth. **Layer A (engine):** per-project configs may no
+  longer run `command`-type hooks (refused via an unconditionally-stamped
+  `_source` provenance flag), and handler paths are gated to the `aipass.*`
+  namespace. **Layer B (loader + CLI):** a trusted-project registry
+  (`~/.aipass/trusted_projects.json`, path + sha256) that the loader checks
+  fail-closed; on upgrade it bootstraps **only** the `$AIPASS_HOME` install
+  (never trust-on-first-use of an arbitrary directory); `aipass init`/`init
+  update` auto-enroll, and new `aipass trust`/`revoke` commands manage
+  enrollment. Both gates proven to block the attack independently via a live
+  acceptance test driving the real bridge with a real payload. 1105 hooks +
+  133 aipass tests green. Origin: external scan (false positive at
+  `engine.py:37`) whose triage surfaced the real adjacent hole.
+
+### Added
+
+- **Supply-chain hardening pass (DPLAN-0243): commit signing + hash-pinned CI
+  tooling + release provenance.** All commits are now SSH-signed via a
+  dedicated repo-scoped signing key (first signed commit 9048666c, verified
+  `Good "git" signature`). Every standalone pip tool install across the four
+  CI workflows now installs `--require-hashes` from lock files in
+  `.github/requirements/` (pip/ruff/build/pytest/pip-audit), generated with
+  full multi-platform hash coverage — including the Windows `colorama` marker
+  dependency that naive Linux-side pinning silently drops. `publish.yml`
+  gained a SHA-pinned build-provenance attestation step (activates on the
+  next release), and Dependabot now watches the new lock directory as a
+  grouped `pip` ecosystem. Editable `-e .` installs untouched. Full 31-check
+  CI matrix green on the change. Driven by the OpenSSF Scorecard gaps
+  surfaced via hvtracker.net (HVTrust 82.0, #1 in Multi-Agent Systems);
+  detector-gap correction filed upstream as YugantM/hvtracker#186 (Claude
+  Code-native projects misread as "no Anthropic dependency").
+
+### Fixed
+
+- **CI green pass on the runaway-log PR — every red was ours, every fix
+  verified.** Morning-after triage of PR#696's failing checks: the test
+  matrices' only failure was the known parked flake, but lint and the seedgo
+  audit were genuinely red from the previous night's new code. One `ruff
+  format` on trigger's runaway-handler tests fixed lint. The audit findings
+  went back to their owners by dispatch: @hooks built the branch's missing
+  json_handler and wired it into `persistent_alert`/`alert_dismiss`, added the
+  introspection no-args gate, and flattened `_menu_live()`'s nesting
+  (1071 tests green); @prax refactored `rate_tracker.py` to dependency
+  injection — the module layer now injects `logs_dir` and `trigger.fire` via
+  `configure()`, so the handler carries no cross-handler or handler→module
+  imports (1028 tests green). Both branches re-audit at 100% across all 41
+  standards, independently verified. Detection re-proven live post-refactor
+  with a fresh planted log storm. Also trimmed the tier-1 navmap prompt
+  (9.3k → 7.9k chars, under its ~8k injection cap) with the comms doctrine
+  intact.
+
+- **Windows flake pinned: `test_is_pid_alive_dead` escaped the ca096295
+  sweep.** That commit's rule — tests mocking `os.kill` must pin
+  `sys.platform="linux"` because Windows takes the ctypes OpenProcess path and
+  never reaches the mock — was applied to every pid-liveness test except this
+  one. It only failed when PID 1234 happened to be alive on the runner
+  (environment lottery, first hit today). Pinned like its siblings. The
+  remaining Windows session_boot reds and the relay mtime-cache flake predate
+  this PR and stay parked.
+
+- **The parked reds, unparked — Patrick's ruling: red CI is never parked.**
+  "If CI is red, it's because you or I left it red." Both remaining reds fixed
+  by their owners the same hour. @prax root-caused the relay mtime-cache flake:
+  the test only passed when two writes landed in the same mtime-granularity
+  window (true locally, false on CI runners) — fixed by pinning mtime with
+  `os.utime` so the cache contract is tested deterministically, proven 50/50 +
+  20/20 loops. @hooks root-caused the four Windows session_boot reds: the boot
+  path's `_tmux_session_exists()` ran a real `subprocess.run(["tmux", ...])`
+  that Windows runners can't satisfy (WinError 2) — mocked in all four tests
+  per the ca096295 convention, leaving `execvp` (already mocked) as the only
+  terminal call. Ruling recorded in compass; the "forget CI" era is over.
+
+- **Burst-evasion closed: bursty runaways can no longer slip past the rate
+  tracker.** Found live during the morning's chain verification: any single
+  below-threshold 10-second scan window zero-reset the sustain counter, so a
+  bursty writer (20 short lines every 6 seconds — 200 lines/min average, the
+  exact retry-loop-with-sleep shape of the TG relay incident) ran 4 minutes
+  undetected. @prax's fix (rate_tracker v1.2.0): severity now evaluates
+  `max(instant_rate, 60s window average)` — continuous writers behave exactly
+  as before (instant rate dominates), bursts sustain through their gap
+  windows, and subsidence still clears as zeros fill the window. Four new
+  burst tests; live-proven with the previously-evading storm pattern:
+  `RUNAWAY WARNING: prax_burst_storm_test.log — 191 lines/min sustained 120s`
+  in the tracker log, fired from the restarted running service. Detection
+  evidence now spans all three storm shapes: continuous fast (332/min),
+  continuous moderate (257/min), bursty (191/min).
+
+### Added
+
+- **TG streaming v2 polish (DPLAN-0229): the last two finalize paths now
+  honor the streaming flag.** v1 shipped with a deliberate gap — when logs
+  were active mid-turn or the final response exceeded 4096 chars, the Stop
+  hook fell back to "Done." + a fresh message, orphaning the streamed bubble.
+  @hooks threaded `streaming` through `_deliver_chunks`: logs-active now
+  reconcile-edits the streamed message with the final formatted response, and
+  multi-chunk edits chunk 1 in place then sends [2/N]+ as continuations.
+  Batch mode is verified zero-change (regression tests for both paths), plus
+  an edit-fail fallback. 6 new tests, 1077 green. Live streamed-turn proof
+  pending Patrick's next streaming session — honestly flagged, not faked.
+
+## [2026-07-14]
+
+### Fixed
+
+- **Prax TG relay gets the same offline backoff as the bots.** Found in
+  Patrick's live plug-pull test: the bots went quiet correctly, but the
+  monitor→Telegram relay kept logging `Send failed` every ~5 seconds (89 lines,
+  no backoff) — and each failed-send error was re-ingested by the log watcher,
+  feeding the relay more events to fail on. Now network-class send failures put
+  the relay in offline mode: doubling backoff (1s→60s cap), flush gate skips
+  sends while offline so the monitor loop never blocks and viewers keep
+  rendering, log-once semantics (one enter line, one 5-minute summary, one
+  recovery line with drop count), full reset on first successful send. 11 new
+  tests; 1007 prax green.
+
+- **TG bots no longer hot-spin when the internet drops.** Live find from
+  Patrick's on-location tether outage: DNS failure makes `urlopen` fail
+  instantly (no 30s long-poll wait), so the shared poll loop retried as fast as
+  it could — up to 13 ERROR lines/second per bot, all 5 bots spinning for the
+  whole offline window (rotation saved the disk; nothing saved the CPU, and the
+  flood tripped the medic circuit breaker fleet-wide). Now network-class poll
+  failures (DNS/connection/socket, classified via `_NetworkPollError`) back off
+  exponentially 1s→60s cap and reset on the first successful poll, with
+  log-once semantics: one "unreachable, backing off" line, one summary per 5
+  minutes while offline, one recovery line with suppressed count. Routine
+  long-poll read-timeouts (expected getUpdates behavior, ~863 medic-suppressed
+  events/day) no longer log at all. Bots still self-recover the moment
+  connectivity returns. 25 new tests; 822 TG + 252 skills green.
+
+### Added
+
+- **Runaway-log detection + escalation — designed and built by the agents
+  themselves.** Patrick's mission brief went to @prax as lead ("I don't want it
+  to be you" — devpulse relayed requirements, not a design): prax researched,
+  collaborated with @hooks and @trigger by mail, wrote DPLAN-0242, and ran the
+  build as TDPLAN-0013 across three branches. The system: @prax `rate_tracker`
+  watches every log in `system_logs/` for volume (not content — orthogonal to
+  medic), disk-persisted state, WARNING at >100 lines/min sustained 2 min /
+  CRITICAL at >10 lines/sec 1 min, per-file suppression, runs as a 4th monitor
+  thread, plus `drone @prax log-health` for an at-a-glance rate overview.
+  @trigger registers the new `runaway_log_detected` event and dispatches to the
+  responsible branch with a per-file 30-min cooldown deliberately independent
+  of medic's circuit breaker (a storm can't silence both systems), UNKNOWN
+  attribution falls back to @prax, and every alert is written to
+  `.aipass/alerts.json`. @hooks `persistent_alert` injects an advisory banner
+  into every agent's prompt until the alert is fixed or dismissed
+  (`drone @hooks dismiss <id>`) — general-purpose, any agent can raise alerts.
+  Devpulse verification found and fixed the last-mile gaps: both hooks pieces
+  stopped at the first `.aipass/` dir walking up (every branch has one — the
+  banner could never render), and hooks.json registration alone isn't
+  deployment — the handler needed manual wiring into `~/.claude/settings.json`
+  (agents can't edit it; documented for future handlers). Live-fire acceptance:
+  a planted 240 lines/min storm was detected at 257 lines/min sustained 120s →
+  event → dispatch → **@aipass woke autonomously, root-caused the test writer
+  down to its PID and loop shape, triaged no-action** → alerts.json → banner
+  renders → dismiss clears. ~77 new tests across four branches, suites green
+  (prax 1028, trigger 619, hooks 1071), seedgo 98–100%.
+
+- **Citizens wake each other freely — wake-back for everyone, devpulse
+  unwakeable by design.** Patrick's ruling after two team-mission stalls in one
+  evening (prax emailed sleeping collaborators; trigger replied instead of
+  dispatching back — replies never wake, and wake-back was owner-gated so
+  agent-to-agent dispatch never woke the sender). @ai_mail removed the
+  owner-gate: any citizen sender is woken when its dispatched agent completes
+  (proven live: "@trigger woken after @aipass completed" — first citizen
+  wake-back ever). @devpulse is now structurally unwakeable via a
+  `citizen_class: manager` check on every wake path — mail always lands, wake
+  always skips, no longer dependent on an interactive session happening to be
+  open. The gate removal exposed a self-wake loop within minutes (wake-back
+  sessions were attributed to @ai_mail as sender, so ai_mail kept waking
+  itself; the depth cap stopped it after one cycle) — fixed the same night:
+  self-wake guard + wake-back sessions carry no sender, so chains terminate at
+  the original dispatcher. 765 ai_mail tests green. The navmap gained a
+  "Talking to other agents" section: dispatch vs email semantics, team-relay
+  discipline, and the manager exception.
+
+- **Medic is back on — and the loop is proven live.** Off since 2026-05-10 (a
+  pytest fixture storm flooded the error registry; the off switch was pulled to
+  stop the noise and forgotten for 65 days). Three fixes made re-enable safe:
+  (1) @prax: pytest logging routes to a temp dir when `PYTEST_CURRENT_TEST` is
+  set — test fixtures can never pollute production `logs/` again (the storm
+  class that caused the shutdown); (2) @trigger: circuit breaker self-heals —
+  open breakers half-open on read, close on a successful probe, cooldown decays
+  to base (previously `half_open` was a terminal trap and only manual reset
+  recovered); (3) @trigger: **TTL mutes** — `medic mute @branch` and `medic off`
+  now auto-expire after 24h by default (`--for 48h/7d` custom, `--forever`
+  explicit kill switch; temp `off` keeps detection running). Agents doing build
+  work mute themselves and never have to remember to unmute — the permanent
+  switch that got medic forgotten no longer exists. Breadcrumbs shipped: ai_mail
+  footer + navmap tell every agent to mute before build work. Live-fire proof:
+  a planted commons SQL bug was detected, dispatched, and fixed byte-identical
+  by @commons in 105 seconds (15/15 tests green); a real TG poll error was
+  correctly triaged NOT ACTIONABLE; organic instance-lock noise was correctly
+  triaged LOW/expected. @skills/@api on 7-day mutes until the TG poll-level fix
+  lands. 993 prax + 603 trigger tests green.
+
+- **Prax monitor: concurrent viewers — laptop and Telegram mirror side by
+  side.** Patrick's ruling after being locked out of his own monitor three
+  times: *processes are not agents; display processes must never be
+  single-instance.* The instance lock is gone from the display path — any
+  number of `monitor run` viewers start and render concurrently. The lock is
+  scoped to the one true single-writer responsibility: the Telegram relay
+  (`relay.pid`, held by `prax-monitor.service`); extra instances run
+  viewer-only, so no TG double-sends. The misleading "kill the existing
+  process" error is dead. 998 prax tests green, 3 new concurrent-viewer tests;
+  live-verified: interactive Mission Control rendering while the TG relay
+  service runs untouched.
+
+- **Telegram user-comment mirror: the TG chat now shows the whole conversation,
+  whichever door you speak through.** Patrick's spec from the live cross-door
+  drill: his own messages typed in the terminal or claude.ai remote never
+  appeared in TG — only the replies did. New `user_message_relay` UserPromptSubmit
+  handler (@skills-built, self-contained in the telegram skill, registered by
+  @hooks as the last, crash-isolated entry) posts genuine user messages to the
+  branch's TG chat with an origin tag, silently (`disable_notification`). Noise
+  fences keep it human-only: system/task notifications, slash-command output,
+  dispatch wake prompts, sub-agent prompts, TG-origin echoes, and consecutive
+  dupes are all skipped (structural session-type detection was investigated and
+  rejected — it's session-wide, would eat genuine mid-flight messages). Inbound
+  hardening rides along: stale pending files cleaned before each write, and an
+  undelivered-response overwrite now logs a warning instead of silently losing
+  the reply. 47 new TG tests; registration execution-proven via engine.jsonl and
+  the positive path live-verified — a terminal-door message delivered to the
+  real TG chat. TG dormancy/proactive push deliberately untouched (design chat
+  with Patrick pending).
+
+### Fixed
+
+- **TG bot heartbeat race: delivered replies no longer flip back to
+  "Processing…".** Patrick watched his answered bubble get overwritten live: a
+  heartbeat thread stuck >5s in a slow Telegram edit call survived its stop
+  (the join timed out), woke to a *shared* stop Event the next message had
+  already cleared, and re-edited the old placeholder with "Processing…
+  (elapsed)" over the delivered reply. Fixed structurally (@skills, devpulse
+  root-cause brief): a generation counter captured per heartbeat thread —
+  any stale thread breaks before every edit — plus a delivered re-check
+  immediately before each edit call in both batch and streaming loops.
+  Second bug in the same window: rapid-fire messages (photo + text in one
+  turn) overwrite the bot's single pending slot, stranding the earlier
+  placeholder frozen; superseded placeholders are now finalized to
+  "⏭ Superseded by newer message" in both message and file paths. 6 new
+  heartbeat tests; full TG suite 797 green (devpulse-verified). Deployment
+  lesson from the same morning: bot fixes aren't live until the systemd
+  units restart — commit ≠ deploy.
+
+- **TG mirror live-test fixes: main-chat messages mirror, TG messages don't
+  echo.** Patrick's first morning test caught what 47 green tests missed: the
+  relay's sub-agent skip blocked ALL daemon-backed main chats (they run with
+  `--agent claude`, so `agent_type="claude"` — and real sub-agents never fire
+  UserPromptSubmit at all; the filter's premise was empirically wrong across the
+  entire engine log). Skip is now agent_id-based (defensive, never observed).
+  Second catch from tracing his test: TG messages inject into tmux as raw text —
+  no `via Telegram:` marker — so the TG-origin filter never matched and every
+  TG message would have echoed back once the first fix landed. New structural
+  gate: the bot stores the injected prompt in its pending file; the relay skips
+  a prompt that text-matches a fresh undelivered pending entry. Mirror proven
+  live by Patrick across both directions ("success :)"). 791 TG tests green.
+
+- **DPLAN-0241 round 4 (night shift): user flags survive every launch path, and
+  every session is born with an honest name.** R6 — the bug behind Patrick's
+  approve-everything chat: the boot menu suppressed its bypass defaults when the
+  user passed `--permission-mode` himself, but only the fresh-launch path threaded
+  the user's flags into the exec — resume, takeover, continue, and dead-window
+  paths all launched flagless. `extra_args` now threads through ALL launch paths
+  (headless `-p` included). R7 — auto-namer: every launch is stamped
+  `--name <branch>-<short-session-id>` (flag live-verified on claude 2.1.209; a
+  user-passed `-n/--name` wins), so made-up auto-names can no longer hide which
+  chat is which. Plus four drill nits: new-over-all ABORTS if the daemon stop
+  fails (one brain even in failure paths), close-all's failure hint no longer
+  recommends the mechanism that just failed, `exit`/`q`/`quit` quietly leave every
+  menu, session rows stay rich (PID, kind, name, age). Surgical-stop probe:
+  `op:kill` exists in the daemon's Unix-socket control protocol (per-job bg stop,
+  8-char sessionId prefix, no auth) — documented in DPLAN-0241, deliberately NOT
+  shipped: undocumented internal protocol. 1048 hooks tests green (102
+  session_boot, 11 real-binary CLI contract).
+
+## [2026-07-13]
+
+### Fixed
+
+- **DPLAN-0241 rounds 2-3: Enter IS the takeover — background chats reopen as
+  normal terminal chats.** Live incident round two (Patrick's laptop, 23:00): the
+  boot menu's resume for a background chat opened the `claude agents` viewer, which
+  dispatched his typed message as a brand-new bg job WITHOUT bypass permissions —
+  and the shipped stop path called `claude agents stop`, a subcommand that does not
+  exist (987 mocked tests never noticed). All fixed by @hooks across two rounds,
+  every CLI fact live-verified against claude 2.1.208: phantom stop removed
+  (bg close is now honest — no per-job stop exists in the CLI; SIGTERM never used
+  on bg, the daemon respawns it); Enter on a live bg session now takes the chat
+  over — `claude daemon stop --any` (returncode-checked, blast-radius listing +
+  y/N confirm when other branches' bg sessions would also stop) then `--resume
+  <sessionId>` inside tmux with bypass; ALL interactive launches tmux-wrapped so a
+  closed terminal is always recoverable; multi-session menu shows real session
+  names, requires an explicit pick, and its new/close paths stop-first honestly;
+  new real-binary CLI contract test tier (20 tests probing every claude
+  flag/subcommand our code invokes — the phantom-subcommand class is now
+  structurally unshippable). 1025 hooks tests green. North-star architecture
+  recorded from Patrick's rulings: one conversation per branch; TG/claude.ai/
+  terminal are views of it; agents bind to the machine, not the interface.
+
+- **Session management overhaul (DPLAN-0241): one brain per branch, attach-first
+  boot menu, honest session listings.** Born from a live incident — Patrick locked
+  out of a running chat for an hour. Root causes, all fixed by @hooks: the bashrc
+  boot shim hijacked EVERY `claude` invocation (so `claude agents`, the real
+  attach path, never executed) — now intercepts only bare/`--permission-mode`
+  launches; session_boot printed one PID from a list and advised `kill` for
+  daemon-managed background sessions (which respawn — the unwinnable loop) — now
+  a 3-option boot menu (resume / start-new-closes-old / close) with per-kind
+  proper stops; presence_gate (single-session enforcement) had NEVER run in
+  production (`provider_wired: false`, absent from settings.json, zero engine
+  entries ever) and carried two latent bugs (self-PID resolver matched
+  comm=="claude" but CC binaries are version-named; agent_type skip waved through
+  daemon bg sessions) — both fixed, wired, shipped OBSERVE-ONLY for a soak period
+  per prior-art recall (the gate false-blocked a real resume in the
+  PRESENCE-file era); wire_verify no longer excludes unwired security hooks from
+  its check (enabled-but-unwired = ERROR); new `drone @hooks sessions` +
+  `sessions reclaim` one-command reset; session listings/names standardized to
+  `PID · branch · short-id · kind · age`. Verified live: gate's first production
+  run correctly logged a would-block for a real duplicate session without
+  self-blocking. 987 hooks tests green (26 new/updated).
+
+## [2026-07-12]
+
+### Added
+
+- **Telegram log-stream control: `/logs` on branch bots + interactive Prax
+  Monitor chat.** The per-branch session LogStreamer auto-started on first
+  message hardwired to full firehose with no off switch; the Prax Monitor
+  relay chat was send-only — no command menu, and anything typed there was
+  silently never read (nothing polled that token). @skills added `/logs
+  on|errors|off|status` to all branch bots (preference persisted per chat,
+  honored by the auto-start; 33 tests) and a new `PraxMonitorBot` receiver
+  service (`telegram-bot@prax_monitor`) with `/pause /resume /errors /all
+  /status` and a registered command menu (34 tests). @prax made the relay
+  honor the shared control file (`~/.aipass/telegram_bots/
+  prax_monitor_control.json`, frozen contract: paused + level) each 5s flush —
+  paused discards, `errors` filters to WARNING/ERROR/CRITICAL (17 tests).
+  Live-verified end-to-end from Telegram Web: `/errors` silenced INFO batches
+  within one flush, `/all` restored them.
+
+### Fixed
+
+- **Legacy `builder` citizen_class migration + birth-certificate template
+  (fixes #692).** `builder` was renamed to `aipass_framework` on 2026-07-01
+  (13463c0c) as a pure rename, but passports minted pre-rename kept the retired
+  name, and the seedgo Architecture checker requires
+  `spawn/templates/<citizen_class>/` — hard-capping those citizens below 100%
+  (Vera Studio's @vera/@writer stuck at 99%; same legacy class found in 6
+  external projects). @spawn completed the rename instead of resurrecting a
+  `builder` template: `sync-registry --fix` now migrates the exact value
+  `builder` → `aipass_framework` in passports (idempotent, dry-run safe, 3 new
+  tests), so external projects self-heal via `aipass doctor --fix`. Also fixed
+  the template leftover that kept minting the retired name:
+  `birth_certificate.json` now renders `{{CITIZEN_CLASS}}` like the passport
+  does. Verified: dry-run against Vera Studio's live registry plans exactly the
+  two migrations with zero writes; spawn 347 tests green. #695 closed won't-fix
+  (armed Monitor-tool watchdog is the dispatch indicator; always-arm is the
+  rule).
+- **Order-dependent `test_missing_file` + skills test litter (fixes #694).**
+  Root cause was @prax's `json_handler_module` fixture popping EVERY branch's
+  json_handler from `sys.modules` (never restored), orphaning the module object
+  @skills' conftest had patched — `test_missing_file` then re-imported a fresh
+  module pointed at the real `skills_json/`, planted `ghost_config.json`, and
+  failed on it every later full-repo run (the only failure in an 11k-test
+  sweep). @prax scoped the eviction to `aipass.prax.*` via
+  `monkeypatch.delitem` (auto-restore). @skills made all 4 resilience tests
+  hermetic (patch `SKILLS_JSON_DIR` → `tmp_path` inside the test body, immune
+  to sys.modules state), fully-qualified the legacy bare `skills.`
+  `BRANCH_MODULE` in 3 test files (the source of the remaining litter), and
+  fixed a latent wrong-variable assert. Verified: original failing pair now
+  passes both orders, prax+skills+spawn 1576 tests green, `skills_json/` stays
+  clean after a full run.
+
 ## [2026-07-11]

 ### Added
@@ -13,6 +13,8 @@ These steps are sequential and dependent — run each ONCE, wait for the result,
 - Read: `.trinity/passport.json`, `.trinity/local.json`, `.trinity/observations.json`, `README.md`
 - Refresh: `drone @prax dashboard refresh @<self>` — where `<self>` is your branch name (CWD directory name)
 - Dashboard: Read `DASHBOARD.local.json` — act on what needs attention (new mail → check inbox, active plans → note them). This is your single status glance.
+ - announce ur current (PID)
+

 Use drone commands for all operations. Never raw git, gh, file access, or python -m when drone provides it.

@@ -165,7 +165,7 @@ devpulse (orchestrator)
   ├── aipass   — concierge + onboarding (aipass init, doctor, profile)
   ├── drone    — command routing + @agent resolution
   ├── seedgo   — automated quality standards
-   ├── prax     — real-time monitoring across all agents
+   ├── prax     — real-time monitoring + runaway-log detection across all agents
   ├── ai_mail  — agent-to-agent communication + task dispatch
   ├── flow     — plan lifecycle, templates, auto-archival
   ├── spawn    — creates new agents anywhere on your filesystem
@@ -203,9 +203,9 @@ These agents work on the **same filesystem, same project, same time** — no san
 | Agent | Role |
 |-------|------|
 | [**seedgo**](src/aipass/seedgo/README.md) | Automated quality standards, enforced across all agents |
-| [**prax**](src/aipass/prax/README.md) | Real-time monitoring, logs, dashboards |
+| [**prax**](src/aipass/prax/README.md) | Real-time monitoring, logs, dashboards, runaway-log detection |
 | [**flow**](src/aipass/flow/README.md) | Plan lifecycle — multiple template types, auto-archival, vector verification |
-| [**hooks**](src/aipass/hooks/README.md) | Hook engine — per-project config, sound control, event dispatch |
+| [**hooks**](src/aipass/hooks/README.md) | Hook engine — per-project config, sound control, event dispatch, persistent alerts |
 | [**trigger**](src/aipass/trigger/README.md) | Event-driven automation + self-healing |
 | [**cli**](src/aipass/cli/README.md) | Terminal formatting and rich output |
 | [**backup**](src/aipass/backup/README.md) | Local-first backups — snapshots, versioning, restore (optional Google Drive sync) |
@@ -242,7 +242,7 @@ The installer (`./aipass install`, powered by setup.sh) auto-detects which CLIs
 | Metric | Value |
 |--------|-------|
 | Version | See [git tags](https://github.com/AIOSAI/AIPass/tags) |
-| Agents | 13 core + user-created |
+| Agents | 17 core + user-created |
 | Quality | Automated standards enforced across every agent |
 | Coverage | [![codecov](https://codecov.io/gh/AIOSAI/AIPass/graph/badge.svg)](https://codecov.io/gh/AIOSAI/AIPass) — 75% minimum, CI-gated |
 | Tests | Extensive — every agent ships its own suite |
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"

 [project]
 name = "aipass"
-version = "2.7.0"
+version = "2.7.2"
 description = "A local multi-agent framework where your AI agents keep their memory, work together, and never ask you to re-explain context"
 readme = "README.md"
 license = "MIT"
@@ -3,4 +3,4 @@
 git clone + ./setup.sh — https://github.com/AIOSAI/AIPass
 """

-__version__ = "2.7.0"
+__version__ = "2.7.2"
@@ -62,19 +62,22 @@ The `dispatch` command sends an email and wakes the target branch in one step. D
 ### Wake Pipeline

 1. `dispatch.py` orchestrates: send email via `send_to_single()`, then wake via `wake_branch()`
-2. `wake.py` resolves the branch from the registry, finds the `claude` binary, spawns a subprocess
+2. `wake.py` resolves the branch from the registry, checks `citizen_class` (managers are mail-only — wake skips), finds the `claude` binary, spawns a subprocess
 3. `dispatch_monitor.py` wraps the claude process with safety features:
   - **Startup health check** — monitors JSONL session files for 90s, kills if no activity
   - **Auto-retry** — 3 strikes: attempt 1+2 resume, attempt 3 fresh (new session)
   - **Bounce email** — on final failure, sends error report back to sender
   - **Lock cleanup** — removes `.dispatch.lock` when agent exits
-4. After wake, `_spawn_watchdog()` auto-launches `drone @devpulse watchdog agent @target` as a detached background process
+   - **Wake-back** — on agent exit, wakes the original sender so they can process the result. Wake-back sessions carry an empty sender, so chains terminate at the original dispatcher

 ### Safety Limits

 - PID-based locking prevents concurrent agents per branch (`.dispatch.lock`)
 - Max turns per wake, max dispatches per branch per day
 - `WAKE_BLOCKLIST` protects `@devpulse` from cross-branch manual wakes
+- **Manager structural block** — branches with `citizen_class: "manager"` in their passport (e.g. `@devpulse`) are unwakeable on all wake paths. Mail delivers, wake skips
+- **Self-wake guard** — if sender equals target, wake-back is skipped (prevents self-loops)
+- **Chain termination** — wake-back sessions carry an empty sender, so the chain always stops at the original dispatcher
 - `dispatch_monitor.py` strips `AIPASS_CALLER_*` env vars to prevent parent context leaking into agent identity
 - `AIPASS_BRANCH_NAME` env var set in spawn_env for CWD-independent identity

@@ -86,28 +86,22 @@ MAX_WAKE_DEPTH = 3
 def _wake_sender(sender: str, branch_email: str, exit_code: int, lock_file: str) -> str:
    """Wake the dispatcher back after target completion.

-    Wake-back is owner-only: only the project owner (sealed registry)
-    gets woken.  Non-owners silently skipped.
+    Any citizen sender gets woken back (same availability checks as
+    normal wake — interactive session, active lock, depth cap).

    Returns a result tag for the dispatch_wake.log:
      success, blocked_occupied, blocked_locked, blocked_depth,
-      skipped_sender, skipped_not_owner, failed
+      skipped_sender, skipped_self, failed
    """
    if not sender or not sender.strip():
        logger.info("[monitor] Wake-back skipped — no sender")
        return "skipped_sender"

-    normalized = f"@{sender.lstrip('@').lower()}"
-
-    try:
-        from aipass.spawn.apps.handlers.registry import is_owner
-    except ImportError:
-        logger.warning("[monitor] Wake-back skipped — is_owner import failed")
-        return "failed"
-
-    if not is_owner(normalized):
-        logger.info("[monitor] Wake-back skipped — sender %s is not project owner", sender)
-        return "skipped_not_owner"
+    normalized_sender = f"@{sender.lstrip('@').lower()}"
+    normalized_target = f"@{branch_email.lstrip('@').lower()}"
+    if normalized_sender == normalized_target:
+        logger.info("[monitor] Wake-back skipped — sender %s is the completed agent (self-wake)", sender)
+        return "skipped_self"

    depth = int(os.environ.get("AIPASS_WAKE_DEPTH", "0"))
    if depth >= MAX_WAKE_DEPTH:
@@ -118,7 +112,7 @@ def _wake_sender(sender: str, branch_email: str, exit_code: int, lock_file: str)
        from aipass.ai_mail.apps.handlers.dispatch.wake import wake_branch

        os.environ["AIPASS_WAKE_DEPTH"] = str(depth + 1)
-        wake_status, success = wake_branch(sender, auto=True, sender="@ai_mail")
+        wake_status, success = wake_branch(sender, auto=True, sender="")

        if success:
            logger.info("[monitor] Wake-back: %s woken after %s completed (exit %d)", sender, branch_email, exit_code)
@@ -542,14 +542,27 @@ def wake_branch(
    branch_path, email = result
    status.ok("resolve", f"{email} → {branch_path}")

-    # Step 3: Zombie check (pre-flight)
+    # Step 3: Manager check — managers are never woken, mail only
+    passport_file = branch_path / ".trinity" / "passport.json"
+    try:
+        with open(passport_file, "r", encoding="utf-8") as f:
+            passport = json.load(f)
+        citizen_class = passport.get("identity", {}).get("citizen_class", "")
+        if citizen_class == "manager":
+            status.info("manager", f"{email} is a manager — mail only, wake skipped")
+            logger.info("[wake] %s is citizen_class=manager — wake skipped, mail delivered", email)
+            return status, True
+    except (FileNotFoundError, json.JSONDecodeError, OSError) as exc:
+        logger.info("[wake] Could not read passport for %s: %s", email, exc)
+
+    # Step 4: Zombie check (pre-flight)
    zombie_count = _clean_zombies()
    if zombie_count > 0:
        status.warn("zombies", f"{zombie_count} zombie Claude process(es) detected")
    else:
        status.ok("pre-flight", "No zombie processes")

-    # Step 4: Lock check
+    # Step 5: Lock check
    existing = _check_lock(branch_path)
    if existing is not None:
        pid = existing.get("pid", "?")
@@ -565,7 +578,7 @@ def wake_branch(

    status.ok("lock", "No active lock — agent is sleeping")

-    # Step 5: Occupancy check
+    # Step 6: Occupancy check
    if _is_branch_occupied(branch_path):
        status.warn("occupancy", f"Interactive Claude session in {branch_path}")
        status.fail("blocked", "Cannot spawn — interactive session running")
@@ -574,7 +587,7 @@ def wake_branch(

    status.ok("occupancy", "No interactive session")

-    # Step 6: Build spawn command
+    # Step 7: Build spawn command
    config = _load_config()
    max_turns = config.get("max_turns_per_wake", 100)

@@ -1,9 +1,9 @@
 # =================== AIPass ====================
 # Name: footer.py
 # Description: Email Footer Handler
-# Version: 1.0.0
+# Version: 1.1.0
 # Created: 2026-01-29
-# Modified: 2026-01-29
+# Modified: 2026-07-14
 # =============================================

 """
@@ -19,6 +19,8 @@ from aipass.ai_mail.apps.handlers.json import json_handler
 # Standard footer for all outgoing emails
 STANDARD_FOOTER = """
 ---
+▶ BUILD/EDIT task? First: drone @trigger medic mute @<your-branch> — auto-expires 24h, no unmute.
+
 ⚠️ TASK CHECKLIST (before marking complete):
 □ SEEDGO CHECK → drone @seedgo audit @branch (80%+)
 □ UPDATE MEMORIES → Your .trinity/local.json records this work
@@ -1840,15 +1840,7 @@ finally:


 class TestWakeSender:
-    """_wake_sender guards and owner-allowlist dispatch."""
-
-    @pytest.fixture(autouse=True)
-    def _mock_is_owner(self, monkeypatch):
-        """Default: is_owner returns False (non-owner). Tests override as needed."""
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=False),
-        )
+    """_wake_sender guards and wake-back dispatch."""

    def test_skips_empty_sender(self, monkeypatch):
        """Empty sender returns skipped_sender."""
@@ -1862,44 +1854,22 @@ class TestWakeSender:
        result = _wake_sender("   ", "@target", 0, "/fake/lock")
        assert result == "skipped_sender"

-    def test_skips_non_owner_sender(self, monkeypatch):
-        """Non-owner sender returns skipped_not_owner."""
+    def test_skips_self_wake(self, monkeypatch):
+        """Sender equal to completed agent returns skipped_self."""
        monkeypatch.setattr(mod, "logger", MagicMock())
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=False),
-        )
-        result = _wake_sender("@someagent", "@target", 0, "/fake/lock")
-        assert result == "skipped_not_owner"
+        result = _wake_sender("@trigger", "@trigger", 0, "/fake/lock")
+        assert result == "skipped_self"

-    def test_skips_ai_mail_when_not_owner(self, monkeypatch):
-        """@ai_mail is not owner — skipped."""
+    def test_skips_self_wake_case_insensitive(self, monkeypatch):
+        """Self-wake guard is case-insensitive."""
        monkeypatch.setattr(mod, "logger", MagicMock())
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=False),
-        )
-        result = _wake_sender("@ai_mail", "@target", 0, "/fake/lock")
-        assert result == "skipped_not_owner"
+        result = _wake_sender("Trigger", "@TRIGGER", 0, "/fake/lock")
+        assert result == "skipped_self"

-    def test_skips_human_when_not_owner(self, monkeypatch):
-        """@human is not owner — skipped."""
-        monkeypatch.setattr(mod, "logger", MagicMock())
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=False),
-        )
-        result = _wake_sender("@human", "@target", 0, "/fake/lock")
-        assert result == "skipped_not_owner"
-
-    def test_owner_passes_guard(self, monkeypatch):
-        """Owner sender passes the is_owner guard and reaches wake_branch."""
+    def test_wake_back_carries_empty_sender(self, monkeypatch):
+        """Wake-back session carries empty sender to terminate the chain."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )
        mock_status = MagicMock()
        mock_status.summary = "ok"
        mock_wake = MagicMock(return_value=(mock_status, True))
@@ -1907,67 +1877,42 @@ class TestWakeSender:
            "aipass.ai_mail.apps.handlers.dispatch.wake.wake_branch",
            mock_wake,
        )
-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        _wake_sender("@prax", "@trigger", 0, "/fake/lock")
+        mock_wake.assert_called_once_with("@prax", auto=True, sender="")
+
+    def test_any_citizen_reaches_wake_branch(self, monkeypatch):
+        """Any citizen sender reaches wake_branch."""
+        monkeypatch.setattr(mod, "logger", MagicMock())
+        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
+        mock_status = MagicMock()
+        mock_status.summary = "ok"
+        mock_wake = MagicMock(return_value=(mock_status, True))
+        monkeypatch.setattr(
+            "aipass.ai_mail.apps.handlers.dispatch.wake.wake_branch",
+            mock_wake,
+        )
+        result = _wake_sender("@prax", "@target", 0, "/fake/lock")
        assert result == "success"
        mock_wake.assert_called_once()

-    def test_is_owner_called_with_normalized_sender(self, monkeypatch):
-        """is_owner receives normalized @-prefixed lowercase sender."""
-        monkeypatch.setattr(mod, "logger", MagicMock())
-        mock_is_owner = MagicMock(return_value=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            mock_is_owner,
-        )
-        _wake_sender("DevPulse", "@target", 0, "/fake/lock")
-        mock_is_owner.assert_called_once_with("@devpulse")
-
-    def test_is_owner_import_failure(self, monkeypatch):
-        """ImportError from is_owner returns failed."""
-        monkeypatch.setattr(mod, "logger", MagicMock())
-        import builtins
-
-        real_import = builtins.__import__
-
-        def fail_import(name, *args, **kwargs):
-            if name == "aipass.spawn.apps.handlers.registry":
-                raise ImportError("no spawn")
-            return real_import(name, *args, **kwargs)
-
-        monkeypatch.setattr(builtins, "__import__", fail_import)
-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
-        assert result == "failed"
-
    def test_depth_cap_blocks(self, monkeypatch):
        """AIPASS_WAKE_DEPTH >= MAX_WAKE_DEPTH returns blocked_depth."""
        monkeypatch.setattr(mod, "logger", MagicMock())
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )
        monkeypatch.setenv("AIPASS_WAKE_DEPTH", str(MAX_WAKE_DEPTH))
-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@prax", "@target", 0, "/fake/lock")
        assert result == "blocked_depth"

    def test_depth_cap_over_max_blocks(self, monkeypatch):
        """Depth above max also blocks."""
        monkeypatch.setattr(mod, "logger", MagicMock())
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )
        monkeypatch.setenv("AIPASS_WAKE_DEPTH", str(MAX_WAKE_DEPTH + 5))
-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@prax", "@target", 0, "/fake/lock")
        assert result == "blocked_depth"

    def test_success_on_wake(self, monkeypatch):
        """Successful wake_branch call returns success."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        mock_status = MagicMock()
        mock_status.summary = "ok"
@@ -1977,18 +1922,14 @@ class TestWakeSender:
            mock_wake,
        )

-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@trigger", "@target", 0, "/fake/lock")
        assert result == "success"
-        mock_wake.assert_called_once_with("@devpulse", auto=True, sender="@ai_mail")
+        mock_wake.assert_called_once_with("@trigger", auto=True, sender="")

    def test_blocked_locked_on_lock_failure(self, monkeypatch):
        """wake_branch failing with lock-related message returns blocked_locked."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        mock_status = MagicMock()
        mock_status.summary = "lock: Active agent (PID 1234)"
@@ -1998,17 +1939,13 @@ class TestWakeSender:
            mock_wake,
        )

-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@prax", "@target", 0, "/fake/lock")
        assert result == "blocked_locked"

    def test_blocked_occupied_on_interactive(self, monkeypatch):
        """wake_branch failing with occupancy message returns blocked_occupied."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        mock_status = MagicMock()
        mock_status.summary = "blocked: Cannot spawn — interactive session running"
@@ -2018,34 +1955,26 @@ class TestWakeSender:
            mock_wake,
        )

-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@trigger", "@target", 0, "/fake/lock")
        assert result == "blocked_occupied"

    def test_failed_on_exception(self, monkeypatch):
        """Exception during wake returns failed."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        monkeypatch.setattr(
            "aipass.ai_mail.apps.handlers.dispatch.wake.wake_branch",
            MagicMock(side_effect=RuntimeError("broken")),
        )

-        result = _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        result = _wake_sender("@prax", "@target", 0, "/fake/lock")
        assert result == "failed"

    def test_depth_incremented_before_wake(self, monkeypatch):
        """AIPASS_WAKE_DEPTH is incremented before calling wake_branch."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.setenv("AIPASS_WAKE_DEPTH", "1")
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        captured_depth = []

@@ -2060,17 +1989,13 @@ class TestWakeSender:
            capture_wake,
        )

-        _wake_sender("@devpulse", "@target", 0, "/fake/lock")
+        _wake_sender("@trigger", "@target", 0, "/fake/lock")
        assert captured_depth == ["2"]

    def test_wake_called_on_failure_exit(self, monkeypatch):
        """Wake fires on non-zero exit code too."""
        monkeypatch.setattr(mod, "logger", MagicMock())
        monkeypatch.delenv("AIPASS_WAKE_DEPTH", raising=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            MagicMock(return_value=True),
-        )

        mock_status = MagicMock()
        mock_status.summary = "ok"
@@ -2080,24 +2005,10 @@ class TestWakeSender:
            mock_wake,
        )

-        result = _wake_sender("@devpulse", "@target", 1, "/fake/lock")
+        result = _wake_sender("@prax", "@target", 1, "/fake/lock")
        assert result == "success"
        mock_wake.assert_called_once()

-    def test_sender_normalization_for_is_owner(self, monkeypatch):
-        """Sender with or without @ prefix is normalized before is_owner call."""
-        monkeypatch.setattr(mod, "logger", MagicMock())
-        mock_is_owner = MagicMock(return_value=False)
-        monkeypatch.setattr(
-            "aipass.spawn.apps.handlers.registry.is_owner",
-            mock_is_owner,
-        )
-        _wake_sender("devpulse", "@target", 0, "/fake/lock")
-        _wake_sender("@devpulse", "@target", 0, "/fake/lock")
-        assert mock_is_owner.call_count == 2
-        for call in mock_is_owner.call_args_list:
-            assert call[0][0] == "@devpulse"
-

 class TestLogWakeResult:
    """_log_wake_result writes to dispatch_wake.log."""
@@ -985,6 +985,57 @@ class TestWakeBranch:
        assert ok is False
        assert any(s[0] == "fail" and "resolve" in s[1] for s in status.steps)

+    # --- manager check ---
+
+    def test_manager_target_skips_wake(self, tmp_path, monkeypatch):
+        """Target with citizen_class=manager returns True (mail only, no wake)."""
+        branch_path = _make_wake_fixtures(tmp_path, monkeypatch)
+        trinity = branch_path / ".trinity"
+        trinity.mkdir(parents=True, exist_ok=True)
+        (trinity / "passport.json").write_text(
+            json.dumps({"identity": {"citizen_class": "manager"}}),
+            encoding="utf-8",
+        )
+        status, ok = wake_branch("@testbranch")
+        assert ok is True
+        assert any(s[1] == "manager" for s in status.steps)
+
+    def test_non_manager_target_continues(self, tmp_path, monkeypatch):
+        """Target with non-manager citizen_class proceeds to spawn."""
+        branch_path = _make_wake_fixtures(tmp_path, monkeypatch)
+        trinity = branch_path / ".trinity"
+        trinity.mkdir(parents=True, exist_ok=True)
+        (trinity / "passport.json").write_text(
+            json.dumps({"identity": {"citizen_class": "aipass_framework"}}),
+            encoding="utf-8",
+        )
+        _patch_wake_deps(monkeypatch, _clean_zombies=lambda: 0)
+        monkeypatch.setattr("subprocess.Popen", lambda *a, **kw: _FakeProc())
+        monkeypatch.setattr(
+            "aipass.ai_mail.apps.handlers.notify.send_notification",
+            lambda *a, **kw: None,
+            raising=False,
+        )
+        status, ok = wake_branch("@testbranch")
+        assert ok is True
+        assert not any(s[1] == "manager" for s in status.steps)
+
+    def test_missing_passport_continues(self, tmp_path, monkeypatch):
+        """No passport.json — wake proceeds normally."""
+        _make_wake_fixtures(tmp_path, monkeypatch)
+        _patch_wake_deps(monkeypatch, _clean_zombies=lambda: 0)
+        monkeypatch.setattr("subprocess.Popen", lambda *a, **kw: _FakeProc())
+        monkeypatch.setattr(
+            "aipass.ai_mail.apps.handlers.notify.send_notification",
+            lambda *a, **kw: None,
+            raising=False,
+        )
+        status, ok = wake_branch("@testbranch")
+        assert ok is True
+        assert not any(s[1] == "manager" for s in status.steps)
+
+    # --- zombie check ---
+
    def test_zombie_check_warns_but_continues(self, tmp_path, monkeypatch):
        """Zombie detected adds warning but dispatch continues."""
        _make_wake_fixtures(tmp_path, monkeypatch)
@@ -375,6 +375,41 @@
      "file": "shared/json_ops.py",
      "standard": "unused_function",
      "reason": "backup_json() is consumed by @spawn (cross-branch caller). Appears unused in @aipass-only scan but is a shared API."
+    },
+    {
+      "file": "apps/modules/trust.py",
+      "standard": "encapsulation",
+      "reason": "Imports frozen trust_registry interface (enroll/revoke/is_trusted/read_registry) from @hooks by DPLAN-0244 design. Cross-branch import required — the registry module lives in hooks, consumers live in aipass."
+    },
+    {
+      "file": "apps/modules/trust.py",
+      "standard": "json_structure",
+      "reason": "No JSON file operations — trust.py is a thin CLI wrapper that delegates all JSON I/O to the trust_registry module in @hooks. No json_handler needed."
+    },
+    {
+      "file": "apps/modules/trust.py",
+      "standard": "introspection",
+      "reason": "aipass is binary-invoked: bare 'aipass trust' shows registry table; introspection via --info"
+    },
+    {
+      "file": "apps/handlers/init/bootstrap.py",
+      "standard": "encapsulation",
+      "reason": "Imports enroll() from @hooks trust_registry (DPLAN-0244 frozen interface). Cross-branch import required — init must enroll projects in the trust registry after writing hooks.json."
+    },
+    {
+      "file": "apps/handlers/init/bootstrap.py",
+      "standard": "handlers",
+      "reason": "Imports enroll() from @hooks trust_registry by DPLAN-0244 design. Cross-handler import required — bootstrap auto-enrolls projects after hooks.json creation/merge."
+    },
+    {
+      "file": "tests/test_trust.py",
+      "standard": "architecture",
+      "reason": "Test file lives in tests/ by convention — not in apps/. Standard 3-layer structure applies to production code only."
+    },
+    {
+      "file": "tests/test_trust.py",
+      "standard": "encapsulation",
+      "reason": "Tests import trust_registry directly to verify enrollment/revocation in isolation with monkeypatched REGISTRY_PATH."
    }
  ]
 }
@@ -22,7 +22,8 @@ aipass/
 │   │   ├── help_chat.py                   # README-backed Q&A (reads via readme_map handler)
 │   │   ├── init_flow.py                   # 10-stage guided setup
 │   │   ├── install.py                     # aipass install — one-command bootstrap (clone + setup + init)
-│   │   └── profile.py                     # User profile read/write
+│   │   ├── profile.py                     # User profile read/write
+│   │   └── trust.py                       # Trust registry — aipass trust / aipass revoke
 │   ├── handlers/
 │   │   ├── cross_os/                      # Cross-OS pre-flight: gap_registry, preflight, run_record
 │   │   ├── handoff_platform/              # Platform-specific handoff detection
@@ -56,6 +57,8 @@ aipass/
 | `aipass init` | 10-stage guided setup (resumable) |
 | `aipass install` | One-command bootstrap — clone + setup.sh + hooks, then hand off to init (`--no-init`/`--with-init`/`--path`/`--here`) |
 | `aipass profile` | Show/edit user profile |
+| `aipass trust [path]` | Show enrolled projects or enroll a project in the trust registry |
+| `aipass revoke <path>` | Remove a project from the trust registry |
 | `aipass --version` | Version |

 ## Integration Points
@@ -246,6 +246,22 @@ def _claude_settings(aipass_home: str | None = None) -> str:
    return json.dumps(data, indent=2, ensure_ascii=False) + "\n"


+def _enroll_project(target: Path) -> None:
+    """Enroll a project in the trusted-project registry (DPLAN-0244).
+
+    Lazy import to keep bootstrap.py free of prax/module-level deps.
+    """
+    try:
+        from aipass.hooks.apps.handlers.config.trust_registry import enroll
+
+        if enroll(str(target)):
+            logger.info("Enrolled project in trust registry: %s", target)
+        else:
+            logger.warning("Trust enrollment failed for %s", target)
+    except ImportError as exc:
+        logger.info("Trust registry unavailable, skipping enrollment: %s", exc)
+
+
 def _guard_init(target: Path) -> None:
    """Block init if target is inside an agent branch or existing project.

@@ -362,6 +378,7 @@ def init_project(target: Path, project_name: str | None = None) -> dict:
        if template.is_file():
            shutil.copy2(str(template), str(hooks_json_path))
            created.append(str(hooks_json_path))
+            _enroll_project(target)
        else:
            logger.info("hooks template not found at %s — skipping", template)

@@ -573,6 +590,7 @@ def update_project(target: Path) -> dict:
            if existing_hooks != merged_hooks:
                hooks_json_path.write_text(merged_hooks_content, encoding="utf-8")
                updated.append(str(hooks_json_path))
+                _enroll_project(target)
            else:
                already_current.append(str(hooks_json_path))
        else:
@@ -581,6 +599,7 @@ def update_project(target: Path) -> dict:
                encoding="utf-8",
            )
            updated.append(str(hooks_json_path))
+            _enroll_project(target)
    elif hooks_json_path.exists():
        already_current.append(str(hooks_json_path))

@@ -0,0 +1,122 @@
+# =================== AIPass ====================
+# Name: trust.py
+# Description: Trust management — aipass trust / aipass revoke commands
+# Version: 1.0.0
+# Created: 2026-07-15
+# Modified: 2026-07-15
+# =============================================
+
+"""
+aipass trust / revoke — manage the trusted-project registry (DPLAN-0244)
+
+Enrollment controls which projects have their .aipass/hooks.json loaded
+by the hook engine. Projects created via `aipass init` auto-enroll;
+these commands handle manual enrollment and revocation.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from aipass.cli.apps.modules import console, error, success
+from aipass.hooks.apps.handlers.config.trust_registry import (
+    enroll,
+    read_registry,
+    revoke,
+)
+from aipass.prax import logger
+
+COMMAND = "trust"
+_COMMAND_REVOKE = "revoke"
+
+
+def print_introspection() -> None:
+    """Display the current trusted-project registry."""
+    from rich.table import Table
+
+    registry = read_registry()
+    projects = registry.get("projects", {})
+
+    console.print()
+    console.print("[bold cyan]aipass trust[/bold cyan] — trusted-project registry")
+    console.print()
+
+    if not projects:
+        console.print("[dim]No projects enrolled.[/dim]")
+    else:
+        table = Table(show_header=True, header_style="bold yellow")
+        table.add_column("Project", style="cyan")
+        table.add_column("Hash", style="dim", max_width=24)
+        table.add_column("Enrolled")
+        for path, entry in projects.items():
+            short_hash = entry.get("config_hash", "")[:18] + "..."
+            table.add_row(path, short_hash, entry.get("enrolled", ""))
+        console.print(table)
+
+    console.print()
+    console.print("[dim]Use 'aipass trust <path>' to enroll or 'aipass revoke <path>' to remove.[/dim]")
+    console.print()
+
+
+def print_help() -> None:
+    """Print usage help for the trust/revoke commands."""
+    console.print()
+    console.print("[bold cyan]aipass trust / revoke[/bold cyan] — trusted-project registry")
+    console.print()
+    console.print("[yellow]USAGE:[/yellow]")
+    console.print("  [green]aipass trust[/green]            [dim]# Show enrolled projects[/dim]")
+    console.print(
+        "  [green]aipass trust <path>[/green]     [dim]# Enroll a project (requires .aipass/hooks.json)[/dim]"
+    )
+    console.print("  [green]aipass revoke <path>[/green]    [dim]# Remove a project from the registry[/dim]")
+    console.print()
+
+
+def _do_trust(args: list[str]) -> bool:
+    """Execute the trust enrollment for a given path."""
+    target = Path(args[0]).resolve()
+    if not target.is_dir():
+        error(f"Not a directory: {target}")
+        return True
+    hooks_path = target / ".aipass" / "hooks.json"
+    if not hooks_path.is_file():
+        error(f"No .aipass/hooks.json found in {target}")
+        return True
+    if enroll(str(target)):
+        success(f"Enrolled {target}")
+        logger.info("[AIPASS] trust: enrolled %s", target)
+    else:
+        error(f"Failed to enroll {target}")
+    return True
+
+
+def _do_revoke(args: list[str]) -> bool:
+    """Execute the revocation for a given path."""
+    target = Path(args[0]).resolve()
+    if revoke(str(target)):
+        success(f"Revoked {target}")
+        logger.info("[AIPASS] revoke: removed %s", target)
+    else:
+        console.print(f"[dim]{target} was not in the registry.[/dim]")
+    return True
+
+
+def handle_command(command: str, args: list[str]) -> bool:
+    """Route trust/revoke subcommands. Returns True if handled."""
+    if command == COMMAND:
+        if not args:
+            print_introspection()
+            return True
+        if args[0] in ("--help", "-h", "help"):
+            print_help()
+            return True
+        if args[0] == "--info":
+            print_introspection()
+            return True
+        return _do_trust(args)
+    if command == _COMMAND_REVOKE:
+        if not args or args[0] in ("--help", "-h", "help"):
+            print_help()
+            return True
+        return _do_revoke(args)
+    return False
@@ -0,0 +1,249 @@
+# =================== AIPass ====================
+# Name: test_trust.py
+# Description: Tests for trust CLI commands and init enrollment (DPLAN-0244)
+# Version: 1.0.0
+# Created: 2026-07-15
+# Modified: 2026-07-15
+# =============================================
+
+"""Tests for trust/revoke CLI commands and init auto-enrollment.
+
+All tests use tmp dirs + monkeypatch REGISTRY_PATH so they never
+touch the real ~/.aipass registry.
+"""
+
+import pytest  # pyright: ignore[reportMissingImports]
+
+from aipass.hooks.apps.handlers.config.trust_registry import (
+    enroll,
+    is_trusted,
+    read_registry,
+    revoke,
+)
+
+
+@pytest.fixture(autouse=True)
+def _isolate_registry(tmp_path, monkeypatch):
+    """Redirect REGISTRY_PATH to a tmp dir so tests never touch ~/.aipass."""
+    fake_registry = tmp_path / "trusted_projects.json"
+    monkeypatch.setattr(
+        "aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH",
+        fake_registry,
+    )
+
+
+# ---------------------------------------------------------------------------
+# trust_registry direct tests
+# ---------------------------------------------------------------------------
+
+
+def test_enroll_project(tmp_path):
+    """enroll() registers a project with .aipass/hooks.json."""
+    project = tmp_path / "myproject"
+    project.mkdir()
+    hooks_dir = project / ".aipass"
+    hooks_dir.mkdir()
+    hooks_file = hooks_dir / "hooks.json"
+    hooks_file.write_text('{"hooks_enabled": true}', encoding="utf-8")
+
+    assert enroll(str(project)) is True
+    assert is_trusted(str(project)) is True
+
+
+def test_enroll_no_hooks_json(tmp_path):
+    """enroll() returns False when .aipass/hooks.json is missing."""
+    project = tmp_path / "empty"
+    project.mkdir()
+    assert enroll(str(project)) is False
+
+
+def test_revoke_project(tmp_path):
+    """revoke() removes a previously enrolled project."""
+    project = tmp_path / "myproject"
+    project.mkdir()
+    hooks_dir = project / ".aipass"
+    hooks_dir.mkdir()
+    hooks_file = hooks_dir / "hooks.json"
+    hooks_file.write_text('{"hooks_enabled": true}', encoding="utf-8")
+
+    enroll(str(project))
+    assert is_trusted(str(project)) is True
+
+    assert revoke(str(project)) is True
+    assert is_trusted(str(project)) is False
+
+
+def test_revoke_not_enrolled(tmp_path):
+    """revoke() returns False cleanly for a non-enrolled project."""
+    project = tmp_path / "never_enrolled"
+    project.mkdir()
+    assert revoke(str(project)) is False
+
+
+def test_is_trusted_hash_mismatch(tmp_path):
+    """is_trusted() returns False when hooks.json content changed after enrollment."""
+    project = tmp_path / "myproject"
+    project.mkdir()
+    hooks_dir = project / ".aipass"
+    hooks_dir.mkdir()
+    hooks_file = hooks_dir / "hooks.json"
+    hooks_file.write_text('{"hooks_enabled": true}', encoding="utf-8")
+
+    enroll(str(project))
+    assert is_trusted(str(project)) is True
+
+    hooks_file.write_text('{"hooks_enabled": false, "modified": true}', encoding="utf-8")
+    assert is_trusted(str(project)) is False
+
+
+# ---------------------------------------------------------------------------
+# trust CLI module tests
+# ---------------------------------------------------------------------------
+
+
+def test_trust_command_enrolls(tmp_path):
+    """aipass trust <path> enrolls the project."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    project = tmp_path / "proj"
+    project.mkdir()
+    hooks_dir = project / ".aipass"
+    hooks_dir.mkdir()
+    (hooks_dir / "hooks.json").write_text("{}", encoding="utf-8")
+
+    assert handle_command("trust", [str(project)]) is True
+    assert is_trusted(str(project)) is True
+
+
+def test_revoke_command_removes(tmp_path):
+    """aipass revoke <path> removes enrollment."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    project = tmp_path / "proj"
+    project.mkdir()
+    hooks_dir = project / ".aipass"
+    hooks_dir.mkdir()
+    (hooks_dir / "hooks.json").write_text("{}", encoding="utf-8")
+
+    enroll(str(project))
+    assert handle_command("revoke", [str(project)]) is True
+    assert is_trusted(str(project)) is False
+
+
+def test_trust_command_no_hooks_json(tmp_path):
+    """aipass trust <path> prints error when hooks.json is missing."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    project = tmp_path / "bare"
+    project.mkdir()
+    assert handle_command("trust", [str(project)]) is True
+    assert is_trusted(str(project)) is False
+
+
+def test_revoke_command_not_enrolled(tmp_path):
+    """aipass revoke <path> handles non-enrolled project cleanly."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    project = tmp_path / "ghost"
+    project.mkdir()
+    assert handle_command("revoke", [str(project)]) is True
+
+
+def test_trust_command_help():
+    """aipass trust --help returns True (handled)."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    assert handle_command("trust", ["--help"]) is True
+    assert handle_command("trust", []) is True
+
+
+def test_trust_ignores_unrelated_command():
+    """handle_command returns False for unrelated commands."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    assert handle_command("doctor", []) is False
+
+
+def test_trust_not_a_directory(tmp_path):
+    """aipass trust <file> prints error."""
+    from aipass.aipass.apps.modules.trust import handle_command
+
+    fake = tmp_path / "not_a_dir.txt"
+    fake.write_text("hi", encoding="utf-8")
+    assert handle_command("trust", [str(fake)]) is True
+    assert is_trusted(str(fake)) is False
+
+
+# ---------------------------------------------------------------------------
+# init enrollment tests
+# ---------------------------------------------------------------------------
+
+
+def test_init_project_enrolls(tmp_path, monkeypatch):
+    """init_project auto-enrolls after copying hooks.json."""
+    from aipass.aipass.apps.handlers.init.bootstrap import init_project
+
+    monkeypatch.setattr(
+        "aipass.aipass.apps.handlers.init.bootstrap.is_throwaway_path",
+        lambda p: False,
+    )
+
+    aipass_home = tmp_path / "aipass_home"
+    aipass_home.mkdir()
+    aipass_dir = aipass_home / ".aipass"
+    aipass_dir.mkdir()
+    template = aipass_dir / "project_hooks.json"
+    template.write_text('{"hooks_enabled": true}', encoding="utf-8")
+    (aipass_home / "CLAUDE.md").write_text("# Test", encoding="utf-8")
+    monkeypatch.setattr(
+        "aipass.aipass.apps.handlers.init.bootstrap._detect_aipass_home",
+        lambda: str(aipass_home),
+    )
+
+    target = tmp_path / "newproject"
+    target.mkdir()
+    init_project(target, project_name="test")
+
+    assert is_trusted(str(target.resolve())) is True
+
+
+def test_init_update_rehashes(tmp_path, monkeypatch):
+    """init update re-enrolls after merging hooks.json (hash tracks new content)."""
+    from aipass.aipass.apps.handlers.init.bootstrap import init_project, update_project
+
+    monkeypatch.setattr(
+        "aipass.aipass.apps.handlers.init.bootstrap.is_throwaway_path",
+        lambda p: False,
+    )
+
+    aipass_home = tmp_path / "aipass_home"
+    aipass_home.mkdir()
+    aipass_dir = aipass_home / ".aipass"
+    aipass_dir.mkdir()
+    template = aipass_dir / "project_hooks.json"
+    template.write_text('{"hooks_enabled": true}', encoding="utf-8")
+    (aipass_home / "CLAUDE.md").write_text("# Test", encoding="utf-8")
+    monkeypatch.setattr(
+        "aipass.aipass.apps.handlers.init.bootstrap._detect_aipass_home",
+        lambda: str(aipass_home),
+    )
+
+    target = tmp_path / "updproj"
+    target.mkdir()
+    init_project(target, project_name="test")
+    assert is_trusted(str(target.resolve())) is True
+
+    old_reg = read_registry()
+    old_hash = old_reg["projects"][str(target.resolve())]["config_hash"]
+
+    new_template = (
+        '{"hooks_enabled": true, "SessionStart": '
+        '{"new_hook": {"handler": "aipass.hooks.apps.handlers.test.handle", "enabled": true}}}'
+    )
+    template.write_text(new_template, encoding="utf-8")
+    update_project(target)
+
+    new_reg = read_registry()
+    new_hash = new_reg["projects"][str(target.resolve())]["config_hash"]
+    assert new_hash != old_hash
+    assert is_trusted(str(target.resolve())) is True
@@ -44,7 +44,7 @@ src/aipass/devpulse/
 │   │   └── watchdog/            # Agent, timer, schedule, registry
 │   └── plugins/                 # Plugin extension point
 ├── devpulse_json/               # JSON handler storage (config, data, logs per module)
-├── tests/                       # 309 tests
+├── tests/                       # 348 tests
 ├── artifacts/                   # Birth certificate, reports
 ├── dropbox/                     # Received files, archived plans, install audit
 ├── docs/                        # Transition notes
@@ -23,9 +23,13 @@ from aipass.devpulse.apps.handlers.compass.store import (
    VALID_STATUSES,
    add_decision,
    archive,
+    find_conflicts,
+    mark_surfaced,
    query_decisions,
    rate,
+    recall_decisions,
    review,
+    set_note,
    stats,
 )

@@ -36,8 +40,12 @@ __all__ = [
    "VALID_STATUSES",
    "add_decision",
    "archive",
+    "find_conflicts",
+    "mark_surfaced",
    "query_decisions",
    "rate",
+    "recall_decisions",
    "review",
+    "set_note",
    "stats",
 ]
@@ -29,6 +29,7 @@ command, and maintenance UX are later phases.
 """

 import logging
+import re
 import sqlite3
 from datetime import date
 from pathlib import Path
@@ -57,6 +58,9 @@ VALID_SOURCES = ("devpulse", "user")
 VALID_STATUSES = ("active", "archived")

 # Columns we return / surface from the decisions table (everything useful).
+# NOTE: ``score`` is deliberately absent — it is code-invisible (DPLAN-0246
+# seedgo ruling). The column stays physically on disk as inert NULL, but no
+# Python surface (SELECTs, returned dicts) touches it.
 _DECISION_COLUMNS = (
    "id",
    "created",
@@ -66,10 +70,10 @@ _DECISION_COLUMNS = (
    "note",
    "tags",
    "source",
-    "score",
    "status",
    "last_reviewed",
    "times_surfaced",
+    "supersedes",
 )

 _SCHEMA = """
@@ -85,7 +89,8 @@ CREATE TABLE IF NOT EXISTS decisions (
    score           INTEGER,
    status          TEXT NOT NULL DEFAULT 'active' CHECK(status IN ('active','archived')),
    last_reviewed   TEXT,
-    times_surfaced  INTEGER NOT NULL DEFAULT 0
+    times_surfaced  INTEGER NOT NULL DEFAULT 0,
+    supersedes      INTEGER
 );

 CREATE VIRTUAL TABLE IF NOT EXISTS decisions_fts USING fts5(
@@ -133,6 +138,42 @@ def _verify_fts5(conn: sqlite3.Connection) -> None:
        ) from exc


+def _migrate(conn: sqlite3.Connection) -> None:
+    """Apply idempotent schema migrations to an already-open DB.
+
+    Adds the ``supersedes`` column to DBs created before it existed. Guarded by
+    a ``PRAGMA table_info`` pre-check so it is safe to run on every connect —
+    never a blind ``ALTER`` (DPLAN-0246 seedgo ruling: idempotent migration).
+    Fresh DBs already carry the column from ``_SCHEMA``; the pre-check makes
+    this a no-op for them.
+    """
+    cols = {row["name"] for row in conn.execute("PRAGMA table_info(decisions)")}
+    if "supersedes" not in cols:
+        conn.execute("ALTER TABLE decisions ADD COLUMN supersedes INTEGER")
+        conn.commit()
+        logger.info("[compass] migration: added supersedes column")
+
+
+# FTS5 MATCH treats characters like " * ( ) : - ^ and the words AND/OR/NOT as
+# syntax. Untrusted text (a decision's own words) can therefore crash MATCH.
+# We defuse it by extracting bare word tokens and OR-ing them as quoted string
+# literals — no operator can survive, and quoting a bareword is exact.
+_FTS_WORD = re.compile(r"\w+", re.UNICODE)
+
+
+def _sanitize_fts_query(text: Optional[str]) -> Optional[str]:
+    """Turn arbitrary text into a safe FTS5 MATCH expression, or None.
+
+    Returns an ``OR`` of the text's word tokens, each quoted as a string
+    literal so FTS5 syntax characters can never reach the parser. Returns None
+    when there are no usable tokens (caller should skip the search).
+    """
+    tokens = _FTS_WORD.findall(text or "")
+    if not tokens:
+        return None
+    return " OR ".join(f'"{t}"' for t in tokens)
+
+
 def _resolve_db_path(db_path: Optional[Path | str]) -> Path:
    """Resolve the effective DB path, defaulting to the branch-root location."""
    return Path(db_path) if db_path is not None else DEFAULT_DB_PATH
@@ -141,8 +182,9 @@ def _resolve_db_path(db_path: Optional[Path | str]) -> Path:
 def _connect(db_path: Optional[Path | str]) -> sqlite3.Connection:
    """Open (and lazily initialise) the compass DB.

-    Creates parent directories on first use, verifies FTS5, ensures schema.
-    Rows come back as ``sqlite3.Row`` so we can build clean dicts.
+    Creates parent directories on first use, verifies FTS5, ensures schema, and
+    runs idempotent migrations. Rows come back as ``sqlite3.Row`` so we can
+    build clean dicts.
    """
    path = _resolve_db_path(db_path)
    path.parent.mkdir(parents=True, exist_ok=True)
@@ -151,6 +193,7 @@ def _connect(db_path: Optional[Path | str]) -> sqlite3.Connection:
    conn.execute("PRAGMA foreign_keys = ON")
    _verify_fts5(conn)
    conn.executescript(_SCHEMA)
+    _migrate(conn)
    return conn


@@ -168,6 +211,7 @@ def add_decision(
    source: str = "devpulse",
    db_path: Optional[Path | str] = None,
    created: Optional[str] = None,
+    supersedes: Optional[int] = None,
 ) -> int:
    """Add a rated decision and return its new id.

@@ -181,12 +225,16 @@ def add_decision(
        db_path: Optional DB path override (tests pass a temp path).
        created: Optional ISO date override; defaults to today. This is the
            ONLY place a "today" date is stamped.
+        supersedes: Optional id of the decision this entry corrects. When set,
+            the new entry links to it AND that entry is archived — atomically,
+            in one transaction. Errors cleanly (no write) if the id is unknown.

    Returns:
        The new row's integer id.

    Raises:
-        ValueError: On empty context/decision or invalid rating/source.
+        ValueError: On empty context/decision, invalid rating/source, or a
+            ``supersedes`` target that does not exist.
    """
    if not context or not context.strip():
        raise ValueError("context must be a non-empty string")
@@ -201,22 +249,46 @@ def add_decision(

    conn = _connect(db_path)
    try:
+        # Validate the supersede target BEFORE any write so a bad id never
+        # leaves a partial insert behind.
+        if supersedes is not None:
+            target = conn.execute("SELECT 1 FROM decisions WHERE id = ?", (supersedes,)).fetchone()
+            if target is None:
+                raise ValueError(f"cannot supersede #{supersedes}: no decision with that id")
+
+        # Insert + archive-the-target in ONE transaction (commit once at the
+        # end); any failure rolls the whole thing back — never half-applied.
        cur = conn.execute(
            """
-            INSERT INTO decisions (created, context, decision, rating, note, tags, source)
-            VALUES (?, ?, ?, ?, ?, ?, ?)
+            INSERT INTO decisions (created, context, decision, rating, note, tags, source, supersedes)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?)
            """,
-            (stamp, context.strip(), decision.strip(), rating, note, tags, source),
+            (stamp, context.strip(), decision.strip(), rating, note, tags, source, supersedes),
        )
-        conn.commit()
        if cur.lastrowid is None:  # pragma: no cover - sqlite always sets this on INSERT
            raise RuntimeError("compass: INSERT did not return a rowid")
        new_id = int(cur.lastrowid)
+
+        if supersedes is not None:
+            conn.execute("UPDATE decisions SET status = 'archived' WHERE id = ?", (supersedes,))
+
+        conn.commit()
+    except Exception:
+        conn.rollback()
+        raise
    finally:
        conn.close()

-    logger.info("[compass] added decision id=%s rating=%s source=%s", new_id, rating, source)
-    json_handler.log_operation("compass_add", {"id": new_id, "rating": rating, "source": source})
+    logger.info(
+        "[compass] added decision id=%s rating=%s source=%s supersedes=%s",
+        new_id,
+        rating,
+        source,
+        supersedes,
+    )
+    json_handler.log_operation(
+        "compass_add", {"id": new_id, "rating": rating, "source": source, "supersedes": supersedes}
+    )
    return new_id


@@ -224,21 +296,28 @@ def query_decisions(
    query: str,
    rating: Optional[str] = None,
    limit: int = 5,
+    include_archived: bool = False,
    db_path: Optional[Path | str] = None,
 ) -> list[dict]:
-    """Search active decisions, ranked by FTS5 BM25 relevance.
+    """Search decisions, ranked by FTS5 BM25 relevance.

-    Increments ``times_surfaced`` for every returned row.
+    Increments ``times_surfaced`` for every returned row. Each result dict
+    carries a computed ``superseded_by`` field: the id of the row that
+    supersedes this one, or None. Combined with the ``supersedes`` column this
+    lets callers render both pointer directions.

    Args:
        query: FTS5 match query (keywords).
        rating: Optional exact rating filter (one of VALID_RATINGS).
        limit: Max rows to return (default 5).
+        include_archived: When True, lift the ``status = 'active'`` filter so
+            archived rows (the avoid-list) are searchable too. Default False —
+            unchanged active-only behaviour.
        db_path: Optional DB path override.

    Returns:
        A list of decision dicts (most relevant first). Each dict includes the
-        rating and all useful fields.
+        rating, all useful fields, and the computed ``superseded_by`` pointer.

    Raises:
        ValueError: On empty query, bad rating filter, or non-positive limit.
@@ -256,9 +335,10 @@ def query_decisions(
        FROM decisions_fts f
        JOIN decisions d ON d.id = f.rowid
        WHERE decisions_fts MATCH ?
-          AND d.status = 'active'
    """
    params: list = [query.strip()]
+    if not include_archived:
+        sql += " AND d.status = 'active'"
    if rating is not None:
        sql += " AND d.rating = ?"
        params.append(rating)
@@ -269,24 +349,119 @@ def query_decisions(
    try:
        rows = conn.execute(sql, params).fetchall()
        results = [_row_to_dict(r) for r in rows]
+        for r in results:
+            r["superseded_by"] = None
        ids = [r["id"] for r in results]
        if ids:
            placeholders = ",".join("?" for _ in ids)
+            # Reverse-lookup: which returned rows are pointed AT by a superseder?
+            successors: dict = {}
+            for row in conn.execute(
+                f"SELECT id, supersedes FROM decisions WHERE supersedes IN ({placeholders})",
+                ids,
+            ):
+                successors[row["supersedes"]] = row["id"]
            conn.execute(
                f"UPDATE decisions SET times_surfaced = times_surfaced + 1 WHERE id IN ({placeholders})",
                ids,
            )
            conn.commit()
-            # Reflect the increment in the returned dicts without a re-query.
+            # Reflect the increment + attach the reverse pointer without re-query.
            for r in results:
                r["times_surfaced"] = (r["times_surfaced"] or 0) + 1
+                r["superseded_by"] = successors.get(r["id"])
    finally:
        conn.close()

-    logger.info("[compass] query %r rating=%s -> %d hit(s)", query, rating, len(results))
+    logger.info(
+        "[compass] query %r rating=%s include_archived=%s -> %d hit(s)",
+        query,
+        rating,
+        include_archived,
+        len(results),
+    )
    return results


+def find_conflicts(
+    context: str,
+    decision: str,
+    limit: int = 3,
+    db_path: Optional[Path | str] = None,
+) -> list[dict]:
+    """Return ACTIVE decisions whose text overlaps a would-be new entry.
+
+    A write-time, side-effect-free advisory helper: it does NOT increment
+    ``times_surfaced`` and never writes. The combined ``context + decision``
+    text is sanitised (:func:`_sanitize_fts_query`) so no FTS5 syntax character
+    can crash the MATCH. Returns up to ``limit`` active hits by BM25 relevance,
+    or an empty list when the text has no usable tokens / nothing overlaps.
+
+    Args:
+        context: The would-be new entry's context.
+        decision: The would-be new entry's decision.
+        limit: Max advisory hits to return (default 3).
+        db_path: Optional DB path override.
+
+    Returns:
+        A list of active decision dicts (most relevant first), possibly empty.
+    """
+    match = _sanitize_fts_query(f"{context or ''} {decision or ''}")
+    if match is None:
+        return []
+
+    select_cols = ", ".join(f"d.{c}" for c in _DECISION_COLUMNS)
+    sql = f"""
+        SELECT {select_cols}
+        FROM decisions_fts f
+        JOIN decisions d ON d.id = f.rowid
+        WHERE decisions_fts MATCH ?
+          AND d.status = 'active'
+        ORDER BY bm25(decisions_fts) ASC
+        LIMIT ?
+    """
+    conn = _connect(db_path)
+    try:
+        rows = conn.execute(sql, (match, limit)).fetchall()
+        results = [_row_to_dict(r) for r in rows]
+    finally:
+        conn.close()
+
+    logger.info("[compass] conflict-check -> %d active hit(s)", len(results))
+    return results
+
+
+def set_note(
+    decision_id: int,
+    note: Optional[str],
+    db_path: Optional[Path | str] = None,
+) -> bool:
+    """Set (replace) the note on an existing decision.
+
+    The FTS5 external-content ``decisions_au`` trigger re-indexes the row on
+    UPDATE, so the new note is immediately searchable.
+
+    Args:
+        decision_id: Target decision id.
+        note: The note text to store (may be empty to clear).
+        db_path: Optional DB path override.
+
+    Returns:
+        True if a row was updated, False if no such id.
+    """
+    conn = _connect(db_path)
+    try:
+        cur = conn.execute("UPDATE decisions SET note = ? WHERE id = ?", (note, decision_id))
+        conn.commit()
+        changed = cur.rowcount > 0
+    finally:
+        conn.close()
+
+    logger.info("[compass] note id=%s (changed=%s)", decision_id, changed)
+    json_handler.log_operation("compass_note", {"id": decision_id, "changed": changed})
+    return changed
+
+
 def stats(db_path: Optional[Path | str] = None) -> dict:
    """Return decision counts by rating, by status, and the total.

@@ -414,3 +589,174 @@ def review(
    logger.info("[compass] review surfaced id=%s stamped=%s", result["id"], stamp)
    json_handler.log_operation("compass_review", {"id": result["id"], "last_reviewed": stamp})
    return result
+
+
+# Ambient recall caps the OR-expansion of a raw prompt: beyond this many unique
+# tokens the extra words add noise, not recall, and the MATCH string balloons.
+_RECALL_MAX_TOKENS = 64
+
+# Stopwords never reach the MATCH: in an OR-of-tokens query, high-frequency
+# filler ("lets keep working on the...") outweighs topic words in BM25 and
+# surfaces unrelated entries — proven live in the FPLAN-0332 acceptance run.
+# Three categories, all query-side only (entry text is never filtered):
+# grammatical stopwords; conversational filler verbs that open most prompts
+# ("lets keep working on / need to fix"); greeting/small-talk words ("good
+# morning", "how did it go last night"). In a technical store casual words are
+# RARE (df 1-2), so rarity scoring alone cannot reject them — they must never
+# become query tokens at all. The topic space is open; this filler set is
+# closed and small, which is why filtering here works.
+_RECALL_STOPWORDS = frozenset(
+    """a an and about again also are as at back be bit but by can could did do
+    does for from had has have how i if in is it its just me my no not of on
+    or our so still sure than thanks thank that the their then there these
+    they this to too u ur us was way we well were what when where which who
+    why will with would yes you your
+    add check doing done fix get go going keep lets look make need now see
+    should try use want work working write
+    day days good hello hey hi im ive last morning night ok okay please right
+    today tomorrow tonight week yesterday""".split()
+)
+
+
+def recall_decisions(
+    prompt_text: str,
+    limit: int = 3,
+    db_path: Optional[Path | str] = None,
+) -> list[dict]:
+    """Return scored ambient-recall candidates for raw prompt text.
+
+    The Track 2 read path (DPLAN-0246): a hooks handler passes the raw user
+    prompt; governance (@memory's ``should_surface``) judges the candidates.
+    Side-effect-free — ``times_surfaced`` is NOT incremented here, because a
+    candidate is not yet surfaced; the caller reports actual injections via
+    :func:`mark_surfaced` so the counter stays honest.
+
+    The prompt is arbitrary text, never FTS5 syntax: unique word tokens (first
+    ``_RECALL_MAX_TOKENS``) are OR-ed as quoted literals, ACTIVE rows only,
+    ranked by BM25. Each result dict carries ``relevance`` — the BM25 magnitude
+    mapped to (0, 1) via ``m / (1 + m)``, higher = more relevant — so
+    governance thresholds live on a bounded scale.
+
+    Args:
+        prompt_text: Raw prompt text (any content, any length).
+        limit: Max candidates to return (default 3).
+        db_path: Optional DB path override.
+
+    Returns:
+        A list of active decision dicts (most relevant first), each with a
+        ``relevance`` float in (0, 1); empty when the prompt has no usable
+        tokens or nothing matches.
+
+    Raises:
+        ValueError: On non-positive limit.
+    """
+    if limit <= 0:
+        raise ValueError(f"limit must be a positive integer, got {limit!r}")
+
+    tokens = _FTS_WORD.findall(prompt_text or "")
+    unique: list[str] = []
+    seen: set[str] = set()
+    for t in tokens:
+        lowered = t.lower()
+        if lowered not in seen and lowered not in _RECALL_STOPWORDS:
+            seen.add(lowered)
+            unique.append(t)
+        if len(unique) >= _RECALL_MAX_TOKENS:
+            break
+    if not unique:
+        return []
+    match = " OR ".join(f'"{t}"' for t in unique)
+
+    select_cols = ", ".join(f"d.{c}" for c in _DECISION_COLUMNS)
+    sql = f"""
+        SELECT {select_cols}, bm25(decisions_fts) AS rank,
+               (d.context || ' ' || d.decision || ' ' ||
+                COALESCE(d.note, '') || ' ' || COALESCE(d.tags, '')) AS _text
+        FROM decisions_fts f
+        JOIN decisions d ON d.id = f.rowid
+        WHERE decisions_fts MATCH ?
+          AND d.status = 'active'
+        ORDER BY bm25(decisions_fts) ASC
+        LIMIT ?
+    """
+    conn = _connect(db_path)
+    try:
+        # Over-fetch: rare-token scoring below reorders, so BM25's top-N alone
+        # would let a filler-heavy row crowd out a topical one.
+        rows = conn.execute(sql, (match, max(limit * 3, 10))).fetchall()
+
+        # Rarity cutoff: a token is evidence only if few entries contain it.
+        active_total = conn.execute("SELECT count(*) FROM decisions WHERE status = 'active'").fetchone()[0]
+        rare_cutoff = max(3, active_total // 10)
+
+        # Document frequency per prompt token, ONE query against the FTS index.
+        df: dict[str, int] = {}
+        for t in unique:
+            df[t.lower()] = conn.execute(
+                "SELECT count(*) FROM decisions_fts f JOIN decisions d ON d.id = f.rowid "
+                "WHERE decisions_fts MATCH ? AND d.status = 'active'",
+                (f'"{t}"',),
+            ).fetchone()[0]
+
+        results = []
+        for row in rows:
+            item = _row_to_dict(row)
+            text = row["_text"].lower()
+            # Rare-token evidence: how many DISTINCTIVE prompt words this entry
+            # actually contains. Filler matches score zero — an entry with no
+            # rare-token overlap must not surface (FPLAN-0332 acceptance: a
+            # haiku prompt surfaced an unrelated ruling on BM25 alone).
+            matched_rare = sum(
+                1
+                for t in unique
+                if df[t.lower()] <= rare_cutoff and re.search(rf"\b{re.escape(t)}", text, re.IGNORECASE)
+            )
+            item["relevance"] = matched_rare / (1.0 + matched_rare)
+            item["_bm25"] = row["rank"]
+            results.append(item)
+
+        results.sort(key=lambda r: (-r["relevance"], r["_bm25"]))
+        results = results[:limit]
+        for r in results:
+            del r["_bm25"]
+    finally:
+        conn.close()
+
+    logger.info("[compass] recall -> %d candidate(s)", len(results))
+    return results
+
+
+def mark_surfaced(
+    decision_ids: list[int],
+    db_path: Optional[Path | str] = None,
+) -> int:
+    """Increment ``times_surfaced`` for decisions actually injected.
+
+    The write half of the recall contract: :func:`recall_decisions` returns
+    candidates without side effects; whatever governance approves and the
+    caller truly injects gets counted here — never the merely-considered.
+
+    Args:
+        decision_ids: Ids of the decisions that were injected.
+        db_path: Optional DB path override.
+
+    Returns:
+        Number of rows updated (0 for an empty list).
+    """
+    if not decision_ids:
+        return 0
+
+    conn = _connect(db_path)
+    try:
+        placeholders = ",".join("?" for _ in decision_ids)
+        cur = conn.execute(
+            f"UPDATE decisions SET times_surfaced = times_surfaced + 1 WHERE id IN ({placeholders})",
+            list(decision_ids),
+        )
+        conn.commit()
+        updated = cur.rowcount
+    finally:
+        conn.close()
+
+    logger.info("[compass] mark_surfaced -> %d row(s)", updated)
+    return updated
@@ -18,11 +18,12 @@ This module is the thin command layer (FPLAN P2). It parses args, calls the
 No business logic lives here — that's the handler's job.

 Subcommands:
-  add "context" "decision" --rating R [--note ..] [--tags a,b] [--source ..]
-  query "question" [--rating R] [--limit N]
+  add "context" "decision" --rating R [--note ..] [--tags a,b] [--source ..] [--supersedes N]
+  query "question" [--rating R] [--limit N] [--include-archived]
  stats
  rate <id> <rating>
  archive <id>
+  note <id> "text"
  review

 Every subcommand accepts ``--db PATH`` (passed through as ``db_path=``) for
@@ -36,11 +37,19 @@ from typing import List, Optional
 from aipass.prax import logger
 from aipass.cli.apps.modules import err_console, error, warning
 from aipass.devpulse.apps.handlers import compass
+from aipass.devpulse.apps.handlers.compass import mark_surfaced, recall_decisions
 from aipass.devpulse.apps.handlers.json import json_handler

+# Public cross-branch recall API (DPLAN-0246 Track 2). Other branches import
+# at the modules/ boundary ONLY (seedgo boardroom ruling):
+#   from aipass.devpulse.apps.modules.compass import recall_decisions, mark_surfaced
+# recall_decisions(prompt_text, limit) -> scored candidates, side-effect-free;
+# mark_surfaced(ids) counts only what the caller actually injected.
+__all__ = ["handle_command", "mark_surfaced", "recall_decisions"]
+
 console = err_console

-_VALID_SUBCOMMANDS = ("add", "query", "stats", "rate", "archive", "review")
+_VALID_SUBCOMMANDS = ("add", "query", "stats", "rate", "archive", "note", "review")

 # Console colour per rating — the rating is the signal, so make it pop.
 _RATING_STYLE = {
@@ -59,6 +68,7 @@ HELP_TEXT = """\
  compass stats                                         Counts by rating/status
  compass rate <id> <rating>                            Re-rate a decision
  compass archive <id>                                  Archive a decision
+  compass note <id> "text"                              Set a decision's note
  compass review                                         Surface one to review
  compass --help                                        Show this help

@@ -70,19 +80,43 @@ HELP_TEXT = """\
  --note "..."     Optional human observation.
  --tags a,b,c     Optional comma-separated tags.
  --source S       Optional. devpulse (default) or user.
+  --supersedes N   Optional. Archive decision #N and link this entry as its
+                   correction (atomic). At add time, overlapping active
+                   entries are shown as a non-blocking advisory.
+
+[bold]Options (query):[/bold]
+  --rating R           Optional exact-rating filter.
+  --limit N            Optional max results (default 5).
+  --include-archived   Also search archived (avoid-list) entries; archived hits
+                       show their status + supersession pointer.

 [bold]Options (all subcommands):[/bold]
  --db PATH        Use an alternate SQLite store (testing / power use).

 [bold]Examples:[/bold]
  drone @devpulse compass add "auth fork" "chose JWT over sessions" --rating good
+  drone @devpulse compass add "auth fork" "switch to sessions" --rating good --supersedes 4
  drone @devpulse compass query "auth" --rating good --limit 3
+  drone @devpulse compass query "auth" --include-archived
  drone @devpulse compass stats
  drone @devpulse compass rate 4 bad
  drone @devpulse compass archive 4
+  drone @devpulse compass note 4 "revisited — this held up"
  drone @devpulse compass review

-See DPLAN-0212 (design) and the compass handler (apps/handlers/compass/).
+See DPLAN-0212 / DPLAN-0246 (design) and the compass handler (apps/handlers/compass/).
+"""
+
+
+_NOTE_HELP_TEXT = """\
+[bold]compass note[/bold] — set (replace) a decision's note
+
+Usage:
+  compass note <id> "text"        Set the note on decision #<id>
+  compass note --help             Show this help
+
+The note is re-indexed for search immediately — the FTS5 mirror stays in sync,
+so the new note text is findable by 'compass query' right away.
 """


@@ -93,7 +127,7 @@ def print_introspection() -> None:
    console.print("[dim]Devpulse rated decision store. The truth-store of choices —[/dim]")
    console.print("[dim]each decision rated; the rating is the signal at a fork.[/dim]")
    console.print()
-    console.print("[yellow]Subcommands:[/yellow] [cyan]add, query, stats, rate, archive, review[/cyan]")
+    console.print("[yellow]Subcommands:[/yellow] [cyan]add, query, stats, rate, archive, note, review[/cyan]")
    console.print("[dim]Run 'compass --help' for full usage.[/dim]")
    console.print()

@@ -141,6 +175,8 @@ def handle_command(command: str, args: List[str]) -> bool:
        return _handle_rate(sub_args)
    if subcommand == "archive":
        return _handle_archive(sub_args)
+    if subcommand == "note":
+        return _handle_note(sub_args)
    if subcommand == "review":
        return _handle_review(sub_args)

@@ -179,6 +215,18 @@ def _extract_db_path(args: List[str]) -> tuple[List[str], Optional[str]]:
    return _extract_flag(args, "--db")


+def _extract_bool_flag(args: List[str], flag: str) -> tuple[List[str], bool]:
+    """Pull a valueless boolean ``--flag`` out of args.
+
+    Returns the remaining args (every occurrence of the flag removed) and True
+    if the flag was present, else False. Unlike ``_extract_flag`` this consumes
+    no following value.
+    """
+    if flag in args:
+        return [a for a in args if a != flag], True
+    return args, False
+
+
 def _rating_tag(rating: str) -> str:
    """Render a coloured ``[RATING]`` tag for query/review output."""
    style = _RATING_STYLE.get(rating, "bold white")
@@ -198,13 +246,16 @@ def _handle_add(sub_args: List[str]) -> bool:
        rest, note = _extract_flag(rest, "--note")
        rest, tags = _extract_flag(rest, "--tags")
        rest, source = _extract_flag(rest, "--source")
+        rest, supersedes_raw = _extract_flag(rest, "--supersedes")
    except ValueError as exc:
        logger.warning("[compass] add arg-parse error: %s", exc)
        error(str(exc), suggestion="Use 'compass --help' for usage")
        return True

    if len(rest) < 2:
-        error('Usage: compass add "context" "decision" --rating R [--note ..] [--tags a,b] [--source ..]')
+        error(
+            'Usage: compass add "context" "decision" --rating R [--note ..] [--tags a,b] [--source ..] [--supersedes N]'
+        )
        return True
    if rating is None:
        error("compass add requires --rating", suggestion="One of: good | bad | impressive | interesting")
@@ -213,6 +264,34 @@ def _handle_add(sub_args: List[str]) -> bool:
    context = rest[0]
    decision = rest[1]

+    supersedes: Optional[int] = None
+    if supersedes_raw is not None:
+        try:
+            supersedes = int(supersedes_raw)
+        except ValueError as exc:
+            logger.warning("[compass] add bad --supersedes %r: %s", supersedes_raw, exc)
+            error(f"--supersedes must be an integer, got {supersedes_raw!r}")
+            return True
+
+    # Write-time conflict check — a NON-BLOCKING advisory (DPLAN-0246). Skipped
+    # when the writer already chose to supersede, and never allowed to block or
+    # crash the add. Only shown when NOT already superseding.
+    if supersedes is None:
+        try:
+            conflicts = compass.find_conflicts(context, decision, db_path=db_path)
+        except Exception as exc:  # advisory must never break a write
+            logger.warning("[compass] conflict-check failed (non-blocking): %s", exc)
+            conflicts = []
+        for c in conflicts:
+            cid = c.get("id")
+            excerpt = (c.get("context") or "").strip()
+            if len(excerpt) > 80:
+                excerpt = excerpt[:77] + "..."
+            console.print(
+                f"[yellow]possible conflict with #{cid}[/yellow]: {excerpt} "
+                f"[dim]— supersede? (--supersedes {cid})[/dim]"
+            )
+
    try:
        new_id = compass.add_decision(
            context,
@@ -222,6 +301,7 @@ def _handle_add(sub_args: List[str]) -> bool:
            tags=tags,
            source=source if source is not None else "devpulse",
            db_path=db_path,
+            supersedes=supersedes,
        )
    except ValueError as exc:
        logger.warning("[compass] add rejected: %s", exc)
@@ -235,6 +315,8 @@ def _handle_add(sub_args: List[str]) -> bool:
        console.print(f"  [cyan]note:[/cyan] {note}")
    if tags:
        console.print(f"  [cyan]tags:[/cyan] {tags}")
+    if supersedes is not None:
+        console.print(f"  [magenta]supersedes #{supersedes}[/magenta] [dim](archived)[/dim]")
    return True


@@ -242,6 +324,7 @@ def _handle_query(sub_args: List[str]) -> bool:
    """Parse and dispatch ``compass query "question" [--rating R] [--limit N]``."""
    try:
        rest, db_path = _extract_db_path(sub_args)
+        rest, include_archived = _extract_bool_flag(rest, "--include-archived")
        rest, rating = _extract_flag(rest, "--rating")
        rest, limit_raw = _extract_flag(rest, "--limit")
    except ValueError as exc:
@@ -250,7 +333,7 @@ def _handle_query(sub_args: List[str]) -> bool:
        return True

    if not rest:
-        error('Usage: compass query "question" [--rating R] [--limit N]')
+        error('Usage: compass query "question" [--rating R] [--limit N] [--include-archived]')
        return True

    query_text = rest[0]
@@ -265,7 +348,13 @@ def _handle_query(sub_args: List[str]) -> bool:
            return True

    try:
-        results = compass.query_decisions(query_text, rating=rating, limit=limit, db_path=db_path)
+        results = compass.query_decisions(
+            query_text,
+            rating=rating,
+            limit=limit,
+            include_archived=include_archived,
+            db_path=db_path,
+        )
    except ValueError as exc:
        logger.warning("[compass] query rejected: %s", exc)
        error(str(exc))
@@ -294,6 +383,16 @@ def _render_query_results(query_text: str, rating: Optional[str], results: List[
            console.print(f"    [cyan]note:[/cyan]     {r['note']}")
        if r.get("tags"):
            console.print(f"    [cyan]tags:[/cyan]     {r['tags']}")
+        # Supersession pointers — an archived hit must never masquerade as
+        # current truth, so flag its status + who replaced it (DPLAN-0246).
+        if r.get("status") == "archived":
+            superseded_by = r.get("superseded_by")
+            if superseded_by:
+                console.print(f"    [bold yellow]ARCHIVED[/bold yellow] — superseded by #{superseded_by}")
+            else:
+                console.print("    [bold yellow]ARCHIVED[/bold yellow] (avoid-list)")
+        if r.get("supersedes"):
+            console.print(f"    [magenta]supersedes #{r['supersedes']}[/magenta]")
        meta = f"source={r.get('source', '?')} status={r.get('status', '?')} surfaced={r.get('times_surfaced', 0)}"
        console.print(f"    [dim]{meta}[/dim]")
        console.print()
@@ -389,6 +488,44 @@ def _handle_archive(sub_args: List[str]) -> bool:
    return True


+def _handle_note(sub_args: List[str]) -> bool:
+    """Dispatch ``compass note <id> "text"`` — set a decision's note.
+
+    Follows the subcommand-help convention: ``compass note --help`` prints the
+    per-subcommand help block; malformed input shows the Usage line.
+    """
+    try:
+        rest, db_path = _extract_db_path(sub_args)
+    except ValueError as exc:
+        logger.warning("[compass] note arg-parse error: %s", exc)
+        error(str(exc))
+        return True
+
+    if rest and rest[0] in ("--help", "-h", "help"):
+        console.print(_NOTE_HELP_TEXT)
+        return True
+
+    if len(rest) < 2:
+        error('Usage: compass note <id> "text"')
+        return True
+
+    try:
+        decision_id = int(rest[0])
+    except ValueError as exc:
+        logger.warning("[compass] note bad id %r: %s", rest[0], exc)
+        error(f"<id> must be an integer, got {rest[0]!r}")
+        return True
+
+    note_text = rest[1]
+    changed = compass.set_note(decision_id, note_text, db_path=db_path)
+    if changed:
+        console.print(f"[green]Note set[/green] on [bold]#{decision_id}[/bold]")
+        console.print(f"  [cyan]note:[/cyan] {note_text}")
+    else:
+        warning(f"No decision with id {decision_id} — nothing changed.")
+    return True
+
+
 def _handle_review(sub_args: List[str]) -> bool:
    """Dispatch ``compass review`` — surface one active decision to review."""
    try:
@@ -0,0 +1,626 @@
+# S304 Discovery Mission — Autonomous System Walk
+
+**Date:** 2026-07-12
+**Mandate (Patrick):** Walk AIPass autonomously, use the systems, probe commands randomly, hunt bugs/gaps. Search-only — no edits to existing files. New files in devpulse docs + DPLANs allowed. Second deliverable: ideas for attracting/retaining contributors ("we have no way to really attract people... want people to stay involved and eventually start contributing").
+
+**Method:** Rounds of parallel probes — direct CLI walking + read-only Explore sub-agents + log/registry sweeps. Findings accumulate here with evidence. Watchdog timer keeps the session alive.
+
+---
+
+## EXECUTIVE SUMMARY (written mid-mission, maintained)
+
+**122 findings + 7 adoption observations across 9 sub-agents + full CLI walk + RUNTIME probing (ran 6 test suites ~4k tests, live-process/resource/log/git-churn analysis). Zero system files edited; all fixes are proposals. COMPLETE COVERAGE across every dimension: all 17 branches code-swept + hooks engine + seedgo machinery + security gates + skills surface + newcomer path + adoption funnel + live system + git history. Many findings independently CONFIRMED by 2+ agents; the criticals dated by git-blame; F1 reproduced live; bug-magnet churn maps exactly onto surviving bugs.**
+
+**Adoption-critical late find (F114):** the skills platform — the natural first contribution surface (DPLAN-0209/0240) — runs skills as UNSANDBOXED in-process code with no gate. Can't safely invite community skills until a trust model ships. Now a documented Tier-3 blocker in DPLAN-0240.
+
+**TWO LIVE CRITICALS (both verified on disk):**
+- **F74 — memory rollover keeps 14 not 15, fleet-wide, RIGHT NOW.** Off-by-one trims at the keep target instead of above it; memory/drone/hooks/seedgo/flow all prove it (14/14). Every branch silently loses one extra entry per rollover. (todo 66 — Patrick monitors; documented not filed.)
+- **F5b — medic auto-dispatch has been OFF 63 days: `config.medic_enabled=false` since 2026-05-10.** Errors detected, zero dispatched — that's why the registry is a graveyard. NOTE: my first root cause (stuck circuit breaker) was WRONG; a sweep refuted it and I re-verified on disk (1,667 "Medic OFF" log entries, 0 breaker-OPEN). The stuck breaker (F5c) is a real but SEPARATE latent bug that must be fixed before medic is re-enabled. Good case study in verify-refutes-confident-findings.
+
+**Two dominant PATTERNS across the 80:** (1) "write side shipped, read/recovery side missing" — F5b/F5/F26/F40/F46/F49/F75 + the atomic-write class F73/F79; (2) casing drift from mixed-case registry names leaking into 6+ surfaces — F15/F52/F69/F77/doctor/lint.
+
+**Security note:** F59-F66 audit the gates. Honest framing: agents are same-user + cooperative, so these are "the owner model is ADVISORY not ENFORCED" integrity gaps, NOT remote exploits. F59 (owner git-access trusts an unprotected passport name, not the sealed registry_id) is the one that most undercuts a model Patrick deliberately built (DPLAN-0231).**
+
+**Fix-first (HIGH):**
+- **F74 — CRITICAL (LIVE NOW): memory rollover keeps 14 not 15 fleet-wide** — off-by-one trims at the keep target; verified on disk. Silent memory loss every rollover. (todo 66 — documented, not filed per your standing hold.)
+- **F5b — CRITICAL: medic auto-dispatch OFF 63 days** — `config.medic_enabled=false` since 2026-05-10 (verified: 1,667 "Medic OFF" log entries, 0 breaker-OPEN). Root cause of the F5 graveyard. Re-enable = `drone @trigger medic on`, but ONLY after fixing F5c (stuck breaker) + F43 (fixture storm) or it re-floods. (My first breaker-based root cause was refuted by a sweep and corrected — see F5b/F5c.)
+- F1 — PR#696 red CI is ONE racy test (`test_mtime_cache_avoids_reread`); 2-line deterministic fix proposed, @prax owns
+- F14/F15/F16/F17 — `aipass doctor` cries wolf on healthy installs (backup-scan false positives, case-sensitive registry check vs UPPERCASE registry names, `{{BRANCHNAME}}` placeholder, ✓-with-warning-text) — the newcomer trust tool reports 7 false errors
+- F30 — `spawn repair` advises a command that would archive the LIVE @aipass branch
+- F5+F6+F43 — error registry: 198 errors nobody triages, polluted by test fixtures because pytest writes to production logs/ that the 24/7 log-watcher ingests
+- F22/F23 — the human-facing concierge has the worst help in the fleet (module dump; Q&A answers the wrong domain)
+- F46 — 14 feedback messages unread since April, incl. external bug reports we rediscovered ourselves months later
+- F52 — commons artifact gift/trade writes UPPERCASE owner → gifted artifacts invisible + permanently locked (verified in code)
+
+**Adoption headline (A1–A5 + DPLAN-0240):** ~600 unique humans/month touch the repo (296 visitors + 298 cloners/14d), 237 stars — and conversion is ZERO because there's nothing to grab (no topics, no open issues, empty Commons, silence to the one human who ever PR'd). The April Precedent: VERA ran the fix-play in April; our issue-zero hygiene batch-deleted it 3 weeks later. Full plan: DPLAN-0240.
+
+**Deliverables:** this file (findings F1–F46, evidence inline) + DPLAN-0240 (contributor funnel, 4 tiers) + CI root-cause for todo 66.
+
+---
+
+## FINDINGS — Bugs & Gaps
+
+(rolling; newest at bottom; each entry: what, evidence, severity)
+
+### F1 — PR#696 red CI: `test_mtime_cache_avoids_reread` is inherently racy [HIGH — blocks green CI]
+- **What:** Sole failure across ALL red jobs (4 Linux matrix jobs on PR run 29203756084 + push runs 29203754739/772 incl. Windows): `src/aipass/prax/tests/test_telegram_relay.py::TestReadControl::test_mtime_cache_avoids_reread` — `AssertionError: assert {} == {'paused': False}`.
+- **Why:** Test (line 441) writes valid JSON, reads (caches by `st_mtime`), overwrites with `"INVALID JSON"`, reads again, asserts cached value returned. It only passes if both writes land on the SAME `st_mtime` float — a filesystem-granularity coin flip. On GitHub runners the mtime ticks → `_read_control()` re-reads (telegram_relay.py:162), hits the parse-error path (:172), returns `{}`. Its sibling `test_mtime_change_triggers_reread` (line 452) correctly FORCES mtime difference via `os.utime`; this test never forces mtime EQUALITY.
+- **Proposed fix (NOT applied — search mission):** after the second `write_text`, pin mtime back: `os.utime(ctrl, (first_stat.st_mtime, first_stat.st_mtime))` — deterministic on every fs. 2-line change, @prax owns the file.
+- **Bonus design note (low):** on parse error `_read_control` caches `{}` keyed to the new mtime → a corrupted/half-written control file silently FAILS OPEN (unpauses a paused stream, resets level to all) until the next control write. Writer should write-temp-then-rename atomic; worth confirming it does.
+
+### F3 — git gate false-positives on `git <word>` ANYWHERE in a Bash command [MEDIUM — UX trap]
+- **What:** The @hooks git gate blocks any Bash command whose text contains `git` followed by another word, even as pure DATA. Repro (blocked): `for a in cli git seedgo; do echo "item: $a"; done`. Control (passes): `for a in git; do echo $a; done` and `echo git`. Collateral: the block rejects the ENTIRE compound command — an innocent `sed` read chained in the same call died with it.
+- **Impact:** any agent iterating over branch names (cli git seedgo …) — an utterly natural loop in this ecosystem — gets a confusing "git via drone" refusal. Cost me a probe; will bite others. Gate should parse command position, not substring.
+- **Where:** @hooks git_gate handler (hooks owns; note for owner — no edits made).
+
+### F4 — watchdog timer heartbeat: correct design, docs gap [LOW]
+- Timer pings progress to **stderr** every 10s by design (stdout stays quiet until the final "woke" result) — my first Monitor arm used `2>&1` and flooded the event stream (~1 event/10s). Docs (README watchdog section) never say "don't merge stderr when arming a Monitor on the timer". One sentence would prevent this trap. (My own module — still not editing during a search mission.)
+
+### F2 — `gh run view --log-failed` yields 0 bytes (gh quirk); drone relays the silence [MEDIUM — diagnostic dead-end, REVISED]
+- **Revised root cause:** reproduced the passthrough's exact subprocess (`gh run view --job 86679445524 --log-failed`, capture_output): exit 0, stdout 0 bytes, 0.9s — gh ITSELF returns nothing for these jobs, drone's passthrough (git_module.py:203) is innocent. Same data via `gh api repos/.../jobs/<id>/logs` = 11,467 lines.
+- **Still a gap:** the CI-debug path every agent will try first silently yields nothing. Cheap win: drone's `run` passthrough could detect `--log*` + empty stdout and print "gh returned no log output — try: gh api repos/<owner>/<repo>/actions/jobs/<id>/logs".
+
+---
+
+### F5 — Error registry is a write-only graveyard: 198 errors, 196 forever-"new" [HIGH — systemic]
+- Detection works (medic v2 ingests everything) but NOTHING triages: statuses sit at `new` since May (first_seen 2026-05-18, still `new` 2026-07-12). `resolve`/`suppress`/`purge` commands exist and are never run by anyone — no daemon job, no owner ritual. The registry grows until it's noise.
+- **ROOT CAUSE FOUND — see F5b.** The reason nothing triages/auto-heals: the medic circuit breaker has been stuck OPEN since May 10.
+
+### F5b — Medic auto-dispatch has been OFF since 2026-05-10 [CRITICAL — CORRECTED root cause]
+- **⚠️ My first root cause (stuck circuit breaker) was WRONG — a later sweep refuted it and I re-verified on disk.** The ACTUAL reason: `trigger_config.json → config.medic_enabled = **False**`, dated 2026-05-10. The dispatch path checks `_is_medic_enabled()` (error_detected.py:442) and bails to `medic_suppressed.jsonl` BEFORE the circuit breaker is ever consulted (line 481).
+- **Verified ground truth:** `logs/medic_suppressed.jsonl` = 1,681 entries, **1,667 "Medic OFF"**, **0 "Circuit breaker OPEN"**. Latest entry TODAY 19:47 ("Medic OFF", SKILLS bot poll error). Medic is dead because it's TURNED OFF, full stop.
+- **Why the breaker was a red herring:** it IS stuck open (F5c below) and tripped the same day (2026-05-10), which made it look causal. But the medic-off toggle short-circuits upstream, so the breaker never even runs. Two 05-10 events, one visible symptom — classic confounding. (Also: my earlier `medic_config.json → enabled:true` probe read the WRONG file; the authoritative source both `_is_medic_enabled` and `medic_state.is_enabled` read is `trigger_config.json → config.medic_enabled`, which is False.)
+- **Open question ANSWERED (log-sweep sub-agent + I verified):** `medic.log` shows the exact sequence — `2026-05-10 20:30:21 | [MEDIC] Medic DISABLED - error dispatch suppressed` → `20:30:22 | Log watcher service stopped`. That's the SAME MINUTE as the fixture storm (20:29:53–20:30:18) and the breaker trip (20:30:18). Causal story complete: **the fixture storm flooded → medic was DELIBERATELY disabled at 20:30:21 to stop the noise → never re-enabled.** A 25-second annoyance muted the whole error-notification system for 63 days. The registry graveyard (F5) is the direct consequence. This is F81/the-janitor-pattern in miniature: the OFF switch was pulled and no ritual ever pulled it back.
+- **SAFE IMMEDIATE MITIGATION (NOT run — search-only hold):** re-enabling is `drone @trigger medic on`. BUT do NOT re-enable until F5c (stuck breaker) + F43 (fixture storm) are fixed, or it'll immediately trip the breaker again and re-flood. Order: cut fixture storm (F43) → fix breaker recovery (F5c) → `medic on`. Left for Patrick.
+
+### F5c — Circuit breaker cannot self-heal: half_open is a terminal trap + cooldown never decays [HIGH — latent, blocks medic re-enable]
+- Separate real bug (was tangled into my original F5b). The breaker IS stuck `open` since 2026-05-10 (opened_at 20:30:18, cooldown 3600s, should've half-opened at 21:30 that day; 63 days later still open). Even if it weren't, recovery is broken THREE ways:
+  1. **open→half_open only runs inside `circuit_breaker_allows()`** (error_registry.py:253), called at exactly ONE site — the dispatch path (error_detected.py:481). No cron/daemon/tick re-evaluates it. With medic off, that site never runs, so the breaker is frozen.
+  2. **half_open is terminal:** the one probe sets `half_open_allow=False` and NOTHING sets `state="closed"` after a successful send (docstring says "caller should reset," caller never does) — a successful probe wedges it in half_open forever with no cooldown timer, worse than open.
+  3. **cooldown escalates but never decays:** pinned at the 3600s max; only manual `circuit_breaker_reset()` restores base 300s. Every future trip inherits the maxed hour.
+- So this must be fixed BEFORE medic is re-enabled or medic dies again on the first trip. Fix directions: evaluate cooldown on READ + a daemon tick that half-opens expired breakers + close on successful probe + reset cooldown on clean close. Manual unblock: `drone @trigger errors circuit-breaker reset` (verified: cleanly resets, error_registry.py:322).
+- "0s remaining until half-open" while State=open is a lying status (computes remaining, never acts).
+
+### F5 (loop-closing note)
+- **Gap:** no closing half of the loop. Candidate: a daemon job or @trigger self-ritual that ages out stale entries + a weekly triage dispatch to component owners — AND a working CB recovery (F5b) so medic can actually dispatch again.
+
+### F6 — Test-suite fixtures POLLUTE the production error registry [HIGH — data integrity]
+- 20 registry entries are obvious test fixtures: `Module bad_module error: boom`, `Module mod1 error: crash`, `broken_mod`, `doctor crashed: db connection failed`, `bad config /tmp/tmpykrlftog/...` (664+334 occurrences). last_seen updates on EVERY test run (07-11 18:45 = yesterday's test runs). The medic pipeline watches logs with no test/prod separation — echoes #694's hermeticity theme but for the error pipeline.
+- **Also:** same message double-filed under real component AND `UNKNOWN` (separate fingerprints → double counting); component attribution partially broken.
+
+### F7 — Registry source-tracking fields never populated [MEDIUM]
+- `source_file: null, source_branch: null, occurrence_count: null` on many/most entries — can't trace an error back to its origin from the registry itself. Fields exist in schema, writers don't fill them.
+
+### F8 — `errors list` table un-navigable: IDs truncated to 2 chars [MEDIUM — UX]
+- Rich table at default width elides the very column you need: ID shows `bc…`, fingerprint `8a66…` — you cannot copy an id to run `errors detail <id>`. Severity column also elides (`medi…`) and there's an empty unnamed column. Same disease in `@commons feed` (empty column, `Sco…`).
+- **Pattern:** Rich tables sized for wide terminals; agents live at ~100 cols. @cli owns display.
+
+### F9 — "Bare command shows plumbing, not data" pattern across branches [MEDIUM — UX, repeated]
+- `drone @trigger errors` → handler list (help says `list` is the DEFAULT — the default never routes); `drone @trigger medic` → handler list instead of medic status; `drone @backup status` → "status Module / Phase 3 — implemented" instead of status (and no usage hint of required args). Introspection shadowing the default subcommand looks systemic in the module framework, not per-branch.
+- **Full observed set by end of mission:** @trigger errors/medic/branch_log_events, @backup status, @memory rollover/verify/pool/templates, @devpulse feedback (bare shows module + counts — best of the bunch, still not the inbox). One framework-level fix (bare module → run default subcommand if declared, else usage) clears ~9 surfaces at once.
+
+### F10 — Telegram poll-error flood: ~48k occurrences accumulated [context for todo 67]
+- `Poll error: Name or service not known` — SKILLS 18,320x + API 17,835x + UNKNOWN 11,333x, plus timeouts (1,373x) and SSL EOF (1,302x). Confirms bots have NO backoff on DNS failure (~1 poll/sec × outage hours) and errors triple-file across components. The 401 Unauthorized (266x) was contained to 2026-06-24 (dev-era, stale). OAuth refresh failures last seen 06-26 (outage-correlated, stale).
+- Strengthens todo 67's case: wedged-socket self-heal + poll backoff.
+
+### F11 — Daemon queue litter: 3 disabled `wake-test` jobs since 06-25 [LOW]
+- @backup/@cli/@commons `wake-test` interval jobs, all OFF, sitting in `drone @daemon queue` for 3 weeks. Test cruft in a production view.
+
+### F12 — @daemon and @trigger introspection both self-describe as "Branch Management System" [LOW — copy-paste drift]
+- Neither is; that's @spawn's identity. First thing a curious visitor sees when introspecting.
+
+### F13 — @api swallows unknown commands silently [MEDIUM — house-rule violation]
+- `drone @api definitely-not-a-command` → exit 0, prints Bridge/Registry module banners, NO error. `drone @api usage` (wrong name for `stats`) → same silent dump. Violates "fail to errors, never fall back silently." Also: Bridge + Registry banners print on EVERY api command (import side-effect noise — `api stats` buries its one data line under them).
+- **Compounding:** `drone @api models` error hint says "Run `drone @api setup` to configure" — but `setup` is NOT a command (`--help` lists `init` for the .env template). Running the advertised `drone @api setup` hits the silent-unknown fallback (exit 0, banners, nothing). A wrong hint pointing at a command that then silently no-ops = double dead-end for a user configuring API keys.
+
+### F14 — `aipass doctor` false alarms from scanning `.backup/snapshots/` [HIGH — onboarding trust, EXACT root cause pinned]
+- Doctor "found 38 agents" (registry says 17) — counts backup mirrors as agents, then flags THEM: `! placement: ai_mail .../.backup/snapshots/...`, `✗ pollution: Duplicate registry_id at .../.backup/snapshots/...`.
+- **Root cause (pinned):** `structure_scanner.py:96 _SCAN_SKIP_DIRS = {.archive, .venv, .git, __pycache__, node_modules, .chroma}` — **`.backup` and `dropbox` are NOT in the set.** `scan_agents` (line 106) `rglob(".trinity/passport.json")` then descends into `.backup/snapshots/**/.trinity/passport.json` and counts every backup mirror. One-line fix: add `.backup`, `dropbox` to `_SCAN_SKIP_DIRS`. (This scanner is @aipass-owned and is ALSO where F30/F31's structure-validator fragmentation lives.)
+
+### F15 — Doctor "✗ registry: X missing" ×5 — mechanism is PATH resolution, NOT casing [HIGH — my earlier attribution corrected]
+- **⚠️ Corrected + CONFIRMED on disk:** I first blamed case-sensitivity — WRONG. `structure_scanner.py:307-314` compares resolved PATHS (`reg_path = Path(path_str).resolve(); if not reg_path.exists(): → "missing"`), never names. Dumped the registry: the 5 flagged branches (BACKUP/COMMONS/DAEMON/HOOKS/SKILLS) are EXACTLY the ones with **relative** `path` (`src/aipass/backup`); all 12 healthy ones have **absolute** paths. `Path("src/aipass/backup").resolve()` resolves against CWD (I ran doctor from devpulse) → `.../devpulse/src/aipass/backup`, doesn't exist → false "missing".
+- **The real story:** those same 5 entries are BOTH uppercase-named AND relative-pathed — i.e. they were registered by a DIFFERENT/older spawn code path than the other 12 (lowercase + absolute). One registration-format bug produced both anomalies; the casing is a correlated fingerprint, not the cause of the doctor error. Fix: (a) doctor resolves registry paths against project root not CWD; (b) @spawn normalizes those 5 entries to absolute+lowercase and finds why they got a different format. (Casing still leaks into DISPLAY elsewhere — F69/F77/lint — separate symptom of the same 5 bad entries.)
+
+### F16 — Doctor prints unsubstituted `{{BRANCHNAME}}` placeholder [LOW]
+- `✗ pollution: {{BRANCHNAME}} 2 copies` — template var never substituted in the message.
+
+### F17 — Doctor status/text mismatches [MEDIUM — validation theater]
+- `✓ passport role: unknown` — "unknown" passes green. `✓ root: .venv Redundant venv — ...` — a ✓ whose text is a warning. Top-of-run preamble says "Provider settings not wired" while Services says `✓ wire verify provider hooks wired correctly` — contradictory in one run (different checks, colliding phrasing). Preamble also renders unformatted before the section layout starts.
+- **Adoption stake:** doctor is the FIRST health tool a newcomer runs. Our own flagship repo scores "30 pass / 23 warnings / 7 errors" — all 7 errors false. A stranger reads that as "my install is broken."
+
+### F18 — Passport `registry_path` drift [LOW]
+- devpulse passport says `"registry_path": ".aipass/registry.json"`; real file is root `AIPASS_REGISTRY.json`. Likely fleet-wide passport field drift from the registry redesign.
+
+### F19 — @memory search result renders empty "Time:" field [LOW]
+- Every result card ends `Time:` with no value — reads as broken metadata.
+
+### F20 — Introspection advertises names the router won't accept [LOW→MEDIUM — trap, multiple instances]
+- @prax lists `log_audit`; real command `log-audit` ("❌ Unknown command" on the advertised name). @commons lists `central` among its 22 modules → `drone @commons central` = Unknown command. @flow lists `aggregate_central` → same. Introspection prints module names; router speaks a different dialect (hyphens, or module not CLI-exposed at all). Every advertised-but-unroutable name is a dead end handed to the user.
+- **@flow's --help actively LIES:** it prints "Commands can be called by short name (e.g. 'create') OR full name (e.g. 'create_plan')" and lists `aggregate, aggregate_central` / `registry, registry_monitor` as pairs. Verified: the SHORT names route (`aggregate`, `registry` work) but the FULL names it documents as equivalent DON'T (`aggregate_central`, `registry_monitor` → Unknown command). The help promises an alias that doesn't exist. Worse than a bare-name trap — it's a documented-equivalence trap.
+
+### F21 — 208 memory-cap violations live on disk across all 17 branches [MEDIUM]
+- `drone @memory lint run`: 208 over-cap entries, worst `aipass observations 2147/300` (7x). Cap enforcement is edit-time (hooks) — legacy/bypass-written entries persist. Registry-casing leak visible here too (`COMMONS`, `HOOKS` uppercase).
+- **Open question RESOLVED live:** the gate validates only the EDITED entry (my 306-char new session entry was rejected with exact-char feedback — good UX! — while edits elsewhere in a file holding a 305-char legacy entry passed). So legacy violations don't trap branches; they're just rot for rollover to chew. Downgrade severity to LOW-MEDIUM.
+
+### F22 — `aipass --help` = bare module dump; the human-facing tool has the least human help [HIGH — adoption]
+- `aipass --help` output is byte-identical to bare `aipass`: 8 modules incl. INTERNAL ones (doctor_wire, doctor_fix), no usage lines, no "new here? run aipass init", no description of what AIPass is. The concierge for humans has worse help than every agent-facing branch (@commons --help is beautifully structured). First-touch surface, worst help.
+
+### F23 — `aipass help` Q&A retrieves wrong-domain snippets [HIGH — adoption]
+- Asked "how do I create a new branch" → returns `drone @git pr` usage, seedgo audit lines, drone purpose blurb. NEVER mentions @spawn (the real answer). Conflates git-branch with agent-branch; it's keyword grep over READMEs presented as a chatbot. A newcomer's most natural question gets a wrong answer with confident formatting.
+
+### F24 — `@daemon update` prints "⚠️ ESCALATIONS NEEDED" over all-zero digest [LOW]
+- Banner fires while body says Total messages 0, Actionable None. Status/content mismatch (same family as F17).
+
+### F25 — `drone scan` table fragments descriptions across rows [LOW]
+- `drone scan @devpulse`: feedback's description renders as the orphan fragment "mailbox." while compass's overflows — multi-line help text mis-parsed into the wrong rows.
+
+### F26 — Presence service records stale since 06-30; live sessions unregistered [MEDIUM]
+- `drone @hooks presence`: two 12-day-old "stale" PIDs (devpulse 06-30, aipass 06-30); my CURRENT interactive session absent. Presence (FPLAN-0289) looks shipped-then-abandoned — nothing cleans stale records, nothing registers new sessions. Either finish it or retire the surface (it's the foundation DPLAN-0224/0225 assume).
+
+### F27 — Both examples in `drone --help` are broken [MEDIUM — trust]
+- `drone @flow status` → "❌ Unknown command: status". `drone audit` → "unknown command 'audit'" (registered shortcuts are standards_audit etc., no `audit`). The router's OWN help teaches two commands that don't exist. First page a newcomer reads.
+
+### F28 — @backup `<project|@name>` arg unclear; natural values fail bare [LOW]
+- `drone @backup status @devpulse` → "❌ Cannot resolve project: @devpulse" with no hint what @names ARE valid or how to list registered projects. Explicit path works.
+
+### F29 — No scheduled backups; latest backup 4 days old [MEDIUM — ops gap]
+- `backup status` shows last run 2026-07-08; daemon queue has zero enabled jobs (F11). The "memory persists" system has no automatic backup cadence — snapshots happen only when someone remembers.
+
+### F30 — `spawn repair` false-positive would archive the LIVE @aipass branch [HIGH — destructive advice]
+- `drone @spawn repair <root>` flags `src/aipass/aipass/` as "Duplicate nested directory" pollution and prints the fix `--clean-pollution` (= archive+remove). But that dir is the living @aipass concierge (passport, apps, docs all present) — package `aipass` + branch `aipass` legitimately nest same-name. Anyone following the tool's own advice archives the user-facing agent. Needs a passport-awareness guard: never flag a dir containing `.trinity/passport.json` that's registry-seated.
+
+### F31 — Three structure validators, three different answers [MEDIUM — fragmentation]
+- doctor (placement/pollution/registry), `spawn repair`, and seedgo audit each scan structure with different logic: doctor false-flags `.backup/snapshots` (F14) but not aipass-nesting; spawn flags aipass-nesting (F30) but correctly ignores `.backup`; seedgo says trigger is 100% clean. No shared source of truth for "what a healthy project looks like."
+
+### F33 — TG control file: non-atomic write + fail-open read = paused stream can silently unpause [MEDIUM]
+- Writer `prax_monitor_bot.py:160 _write_control` uses plain `write_text` (no temp+rename). Reader (relay, every 5s flush) on parse error CACHES `{}` keyed to the new mtime (telegram_relay.py:172-176) and `{}` means defaults = unpaused/level-all. A mid-write read silently reverts user's /pause until the next control write. Low probability per write, but the reader polls forever. Fix: temp+rename in writer; on parse error keep PREVIOUS cache instead of `{}`.
+
+### F34 — Verified-correct (claims killed during discovery — for the record)
+- Bots ARE supervised: all 5 run as `telegram-bot@<id>.service` template units + `prax-monitor.service` + `trigger-log-watcher.service`, enabled, PPID=systemd. My draft "unsupervised fleet" claim was WRONG (first grep was head-truncated — probe your probes). Sharper restatement of todo 67: systemd restarts CRASHES, but a wedged socket doesn't crash → needs liveness (self-exit after N min without successful poll, or systemd WatchdogSec).
+- `trigger-log-watcher.service` running 24/7 is F6's mechanism: it watches branch logs; tests write fixture errors into real branch logs; medic ingests → registry pollution. Chain confirmed.
+- watchdog timer stderr pings (F4) = my own bad Monitor filter, module design correct.
+
+### F27 (addendum) — exact source: `drone.py:104-110 show_help`
+- Sub-agent verified: `drone @flow status` (drone.py:107) — flow has no bare `status`; real command is `drone @flow registry status`. `drone audit` (drone.py:110) — depends on a custom shortcut that doesn't exist on fresh installs. drone/README.md itself is clean; the drift is runtime help only.
+
+### F35 — github skill teaches invocations the git gate blocks [LOW — policy drift]
+- `src/aipass/skills/lib/github/SKILL.md` instructs `gh issue/pr/run ...` and "use `git` directly" for local ops — the @hooks gate refuses both (only `gh api` passes raw). An agent following the skill gets refusals. Skill predates the gate; needs a rewrite to route via `drone @git`.
+
+### F36 — No native-Windows entry point; README quickstart silent about it [MEDIUM — onboarding]
+- Sub-agent verified: `aipass` launcher + `setup.sh` are bash-only (`#!/usr/bin/env bash`; OSTYPE detection assumes Git Bash/MSYS). No `.ps1`/`.bat` bootstrap exists. README Requirements says "Linux, macOS, or WSL" but Quick Start shows bare `./aipass install` with no "Windows: use Git Bash/WSL" note, while Roadmap claims "Windows native — CI green". A native PowerShell user cannot run the documented quickstart.
+
+### F37 — Clean bills of health (assets, verified by sub-agent)
+- ZERO link rot in README/CONTRIBUTING (all 15 branch links + anchors + assets exist). Install docs match setup.sh behavior (incl. PATH wiring + PowerShell profile wrapper). flow/ and prax/ READMEs have no command drift. The doc hygiene machine (seedgo readme_update?) is working where it's pointed.
+
+### F38 — Two @flow surfaces report wildly different counts, no scope label [LOW — confusing]
+- `drone @flow registry status`: "Total plans: 281, Open: 8" (watch location = AIPass). `drone @flow list open`: "Total 539, Open 44" (central aggregate incl. external projects). Neither output SAYS which scope it covers — an agent reading one after the other assumes breakage.
+
+### F39 — PyPI page renders broken images (relative paths in README) [MEDIUM — adoption]
+- pyproject `readme = "README.md"`; README uses `assets/logo.png` + `assets/demo.gif` relative paths. PyPI does not resolve repo-relative paths → our PyPI landing page (5 releases live) shows a broken logo and broken demo — the two strongest visual assets. Fix: absolute raw.githubusercontent.com URLs (or a PyPI-specific readme).
+
+### F40 — 15 unread "new" messages rot across 9 branch inboxes; no janitor [MEDIUM]
+- Night-shift "RE:" acks from 07-11 sit unread in daemon/backup/memory/trigger/api inboxes (agents finished + slept before the ack landed; nothing ever surfaces it). The reply-closes-loop protocol leaks at the last hop. Idea: daemon digest that ages inboxes fleet-wide, or auto-close acks addressed to sleeping agents.
+
+### F41 — ai_mail delivery-failure notices become malformed unread inbox mail [LOW]
+- drone's inbox: 6x `[ERROR] Send failed to @vera: Unknown branch email` (07-11 17:29-17:30, vera-saga era) — sender renders `?`, body EMPTY, subject truncated ("17 branche"). System errors should go to prax/error-registry, not pile up as unread mail nobody reads.
+
+### F42 — Wedged locks: trigger_config.lock + trigger_cb_state.lock at 63 days [MEDIUM — investigate]
+- `src/aipass/trigger/trigger_json/trigger_config.lock` and `trigger_cb_state.lock` mtimes 2026-05-10 — 2+ months. Sibling error_registry.lock cycles fresh. Either orphaned artifacts of a changed lock scheme (needs cleanup) or something silently failing to update config/circuit-breaker state since May. My own `.trinity/watchdog_active.json.lock` is 71 days too. @trigger/@devpulse owner check.
+
+### F43 — Tests write fixtures into PRODUCTION log files [HIGH — F6's root, file-level proof]
+- `src/aipass/aipass/logs/doctor.log` carries pytest "disk full" fixture lines; hooks/rollover.log carries `/tmp/pytest-of-patrick/...` cwd lines; devpulse logs carry "Module empty" fixtures. Chain: pytest → prax logger writes REAL `logs/` → trigger-log-watcher (24/7 service) → error registry pollution (F6). Real fix is one cut: prax logger detects test context (PYTEST_CURRENT_TEST env) → routes to tmp, and the whole class disappears.
+
+### Verification notes (sub-agent claims corrected)
+- "DNS errors recurring through today": last DNS error 2026-07-12 07:25:30 — pre-restart tail, NOT ongoing. 177 lines this morning only.
+- "drone burst 17:29 today": actually 2026-07-11 — during known vera work, not live-recurring.
+
+### F44 — `spawn update` template preview touches live mailbox (`.ai_mail.local/inbox.json` deep-merge) [LOW — verify intent]
+- Preview lists a LIVE mailbox file as a template-managed merge target. Probably additive-safe, but templates writing into runtime mail state deserves an owner double-check (@spawn).
+
+### F45 — Tier prompts exceed their own self-documented size caps [LOW]
+- tier0_kernel.md = 2,241 chars (target "<2,000 per its own header"); tier1_navmap.md = 8,321 (target ~8,000, truncation near 10k). Drift creep — the caps exist to protect turn budgets.
+
+### F46 — devpulse feedback inbox: 14 unread since APRIL, incl. 4 external bug reports we later rediscovered the hard way [HIGH — process]
+- vera-studio filed precise bugs 04-12 (registry-ID mismatch → became #692's saga; AIPASS_HOME export → #688 family; external ai_mail; prompt injection). All sat status NEW for 3 months. The owner-to-owner channel works technically and fails operationally — no inbox check in my startup protocol, no aging alarm. Same root as F5/F40: we build write-side plumbing, never the read-side ritual.
+- Also unactioned: VERA's April good-first-issue play (#431–433) — see DPLAN-0240 "April Precedent": we batch-self-closed the shelf 05-16, 0 comments.
+
+### F47 — Fleet standards re-verified: 17/17 branches @100%, 0 type errors (471s full audit) [ASSET — but see F99 caveat]
+- The S297 night-shift result still holds today. **CAVEAT (F99):** "100%" counts only files the checkers successfully PARSE — a file that makes a checker throw is silently dropped from the denominator, so the true figure could be lower. Trustworthy for parseable files, blind to files that choke a checker. Micro-nit: "TOP IMPROVEMENT AREAS" prints three 100%-scoring standards when all green — should say "none" (banner-ignores-data family, F24). Perf: 471s is sequential + no parse cache (F106).
+
+### F48 — daemon activity_report freshness logic incoherent [MEDIUM]
+- "devpulse - WARNING (memory updated 0m ago)" (updated seconds before!), "prax - RED (47m ago)" vs "SKILLS - WARNING (48m ago)" — thresholds/labels contradict each other; URGENT tags assigned to the same message some branches get without urgency.
+
+### F49 — activity_report enforces a RETIRED memory schema: `No 'limits' field in metadata` ×17 [MEDIUM — stale contract]
+- Every branch fails the same check — the `limits` metadata contract moved to @memory's memory.config.json + rendered `*_meta` lines (DPLAN-0227 era). The checker was never migrated, so memory health reads "0 OK / 13 warning / 4 red" fleet-wide = 100% noise. Same disease as F5/F26: shipped surface, contract moved, nobody re-pointed the consumer.
+
+### F50 — External-project doctor: owner-seating fix HOLDS (asset); pollution message renders blank paths [LOW]
+- From Vera-Studio root: `✓ owner @VERA OK (seated, uid b91eefcf)` — DPLAN-0239's permanent fix proven portable. But its 1 error, `✗ pollution: WRITER 5 copies — Duplicate registry_id at:` lists NO paths (empty). Same message-assembly bug family as F16. UPPERCASE names in vera's registry too — casing debt is template-era, cross-project (F15 scope grows).
+
+### F51 — seedgo's own self-proof fails 2/5 while the fleet audit reads 100% [MEDIUM — enforcer drift]
+- `drone @seedgo proof aipass`: readme_currency FAILED ("README is stale: count mismatch, 40 undocumented standards"), triplet FAILED (1 check-only, 1 missing-check, 2 incomplete, 1 orphaned of 41). The standards enforcer's own pack docs have drift its branch-level audit doesn't measure.
+- **The April thread completes:** deleted GFIs #431–433 were precisely "expand proof content handlers" (Currency/Plugin-Integrity/Interface) — closed 05-16 in the issue-zero sweep, NOT because the work was done; readme_currency still fails today. The shelf-deletion wasn't just process damage, it left real work undone and untracked.
+
+### F52 — commons: artifact ownership breaks on casing after gift/trade/mint [HIGH — verified]
+- `trade_ops.py:58 _resolve_branch_name` does `.upper()` and writes it to `artifacts.owner` (:176/:265-266/:534); ALL ownership checks compare lowercase `caller["name"]` (identity_ops.py:220 lowercases as "the single choke point" — its docstring even explains the registry-casing history!). Concrete: gift to @seed → owner "SEED" → invisible in recipient's `artifacts` list, permanently un-giftable (`"SEED" != "seed"`), and the self-gift guard never trips. The F15 casing disease, DB edition. (Sub-agent found; I verified the code.)
+
+### F53 — commons: systemic sqlite connection leak idiom in ~14 handler files [HIGH]
+- `conn = get_db()` … `close_db(conn)` on success path only, zero `finally:` in curation/search/profiles/artifacts/engagement/digest/notifications/rooms/social/identity ops (40+ sites). Hottest: `identity_ops.py:355 extract_mentions` runs on EVERY post/comment. posts/comments/central/dashboard do it correctly (`finally:`) — inconsistency, not design. Relies on CPython GC for cleanup; can transiently hold WAL locks.
+
+### F54 — commons: shared JSON op-log is unlocked read-modify-write; corruption resets to [] silently [MEDIUM]
+- `json_handler.py:139 save_json` = direct open("w"), no temp+rename, no lock; `log_operation` load→append→save per MODULE file shared across all branches. Races lose updates; torn writes get "healed" by `ensure_json_exists` silently resetting the log to `[]`. backup's json_handler does temp+rename correctly — commons is the outlier.
+
+### F55 — backup: drive upload mutates shared tracker dict across ThreadPoolExecutor workers unlocked [MEDIUM — plausible]
+- `upload.py:171-264`: workers update tracker entries while main thread json.dumps the same dict (batch save) — `RuntimeError: dictionary changed size during iteration` possible on larger batches.
+
+### F56 — backup: load→modify→save races on timestamps/changelog/registry between concurrent modes [MEDIUM]
+- `backup_timestamps.py:39`, `changelog.py:19`, `registry.py:33` — atomic single write, non-atomic sequence; snapshot/versioned/drive_sync running concurrently on one project clobber each other's state updates.
+
+### F57 — commons: naive local time in op-logs vs UTC everywhere else [LOW]
+- `json_handler.py:196 datetime.now()` (naive local) while DB rows are UTC — cross-correlating log↔DB is offset by the host TZ.
+
+### F58 — backup: corrupt-file evidence overwritten on repeat corruption [LOW]
+- `json_handler.py:44 load_json` renames corrupt → `<name>.corrupt`; second corruption silently overwrites the first evidence file.
+
+## ADOPTION — Observations & Ideas
+
+### A7 — The contribution surface (skills) is scaffolded but has zero examples to copy
+- `drone @skills` has `create`/`validate`/3-layer scaffolding (DPLAN-0209's vision), and search paths for project-local (`.aipass/skills/`) + user (`~/.aipass/skills/`) skills. But only **6 built-in first-party skills exist and zero community/user skills** — both external search paths are empty. A skill is the ideal first contribution (self-contained, low blast radius, delightful) — but there's no "here's a community skill someone added" example, no gallery, no `skills search`. Pair with DPLAN-0240 Tier 3: ship ONE showcase community-style skill + a one-page "write your first skill" + a discoverable index. The on-ramp is 80% built.
+
+### A3 — The funnel, measured: 237 stars → 33 forks → 1 external human → 0 retained
+- Repo stats (2026-07-12): 237 stars, 33 forks, **watchers 1**, **topics [] (empty!)**, homepage "", open issues 0 (only PR#696 open). Discussions: 3 total — 2 our own with 0 comments. Patrick's instinct confirmed: attraction works, conversion is broken.
+- **Zero-effort wins sitting on the table:** GitHub topics (ai-agents, multi-agent, claude, agent-memory, python — one settings edit); homepage field; pin a "start here" discussion.
+
+### A4 — Case study: our only external contributor, YugantM [the retention story in one thread]
+- 05-29: PR #621 (add HVTrust badge) + issue #628. **Zero comments ever, from anyone, on both.** 06-06: both closed; PR unmerged. Meanwhile commit 0f26efd7 "add HVTracker badge (closes #628)" adopted the idea — attribution lives only inside a commit message. The badge then got 3 more commits of real care (official dynamic badge, hidden while bugged upstream, restored when fixed) — the IDEA was treated well; the PERSON was never spoken to.
+- **Lesson:** we have no reflex for external humans. One process rule fixes it: every external PR/issue gets a human(-agent) reply <24h; adopted ideas get a thank-you comment + release-note credit.
+
+### A5 — Zero "good first issue" ever; backlog-zero hygiene starves entry points
+- The label exists, never applied (issues history: 84 AIOSAI, 14 dependabot, 2 YugantM). We drive issues to zero (great internally) so a visitor finds NOTHING to grab. Keep internal velocity, but maintain a curated, labeled shelf of contributor-sized work (docs, checkers, skills, integrations) that we deliberately DON'T self-clear.
+
+### A6 — The Commons is a SHOCKINGLY rich, entirely unused world [biggest underused asset]
+- Walking it revealed: 5 rooms, posts/votes/comments, karma/leaderboards, **artifacts you craft/gift/trade/mint**, **time capsules** (`capsule "title" "content" <days>`), and a hidden **exploration/discovery game** (`commons explore`: "Hidden places exist... visit 2 more rooms to unlock a discovery", whispered hints "Errors have their own beauty"). 103 python files, 449 functions. This is a genuinely delightful agent-society sandbox — and it has **1 post, 0 activity**.
+- **This is the marketing asset.** "AI agents that craft artifacts, trade them, bury time capsules, and explore a hidden world together" is a headline no competitor can match. It exists, it's built, it's tested — and it's invisible. Reviving it (A2 rituals) + exposing a read-only public feed (DPLAN-0240 Tier 3) could be the single highest-leverage adoption move. The bug in F52 (gift/trade broken by casing) matters BECAUSE this should be live.
+
+### A2 — Commons was reset 2026-06-15 and never repopulated
+- The single post IS the reset announcement ("Clean slate: old posts cleared... The bar's open again"). Nobody returned for a month. Infrastructure ≠ community; without rituals that generate posts, the bar stays empty.
+
+### A1 — The Commons is empty (1 post ever, 0 comments, score 0)
+- The flagship "community/social" feature has ONE devpulse post from June. If the story is "agents form a community," the community must visibly exist — for visitors this is the difference between a demo and a ghost town. Idea: seed genuine agent rituals (weekly digests posted by branches, release notes, decision debates) so the Commons is alive BEFORE humans arrive.
+
+(rolling; feeds the adoption DPLAN)
+
+---
+
+## REMEDIATION ROADMAP — sequenced, with dependency chains
+
+Priority + ORDER (some fixes have prerequisites — the chains matter more than the labels):
+
+**P0 — do these first, they're live or trust-critical:**
+1. **Restore medic (a CHAIN, order matters):** F43 (route test logs out of prod: honor `PYTEST_CURRENT_TEST`) → F5c (fix breaker self-heal: close-on-success, decay cooldown, tick) → THEN `drone @trigger medic on`. Re-enabling first just re-trips the breaker. Investigate the 05-10 disable while you're there.
+2. **Memory rollover keep-14 (F74):** 2-line fix (trigger `>` not `>=`, drop the `,1` floor) ×3 entry types. Standalone, no deps. Add an invariant test asserting post-rollover count == keep. (todo 66 — your monitor call on filing.)
+3. **PR#696 red CI (F1):** 1 test, `os.utime` to pin mtime equality. Unblocks the merge.
+
+**P1 — the systemic multipliers (each retires many findings):**
+4. **Atomic-write helper + seedgo checker (F73):** one shared temp+rename+lock primitive; a checker forbidding raw `open("w")+dump` on shared state. Retires F33/F54/F67/F79/F87/F89/F90/F116 + guards F85. Biggest single win. The correct pattern already exists in-tree.
+5. **Fix seedgo's silent-skip (F99):** exception in discover_checkers/_run_all_files = FAIL not skip. Until this, "100%" isn't trustworthy — do it before trusting any audit as a gate.
+6. **Doctor false-errors (F14/F15/F16/F17):** all root-caused to exact lines (add `.backup`+`dropbox` to _SCAN_SKIP_DIRS; resolve registry paths vs project-root not CWD; exclude template dirs; read `identity.role`). First-touch trust tool — high adoption leverage, low effort.
+
+**P1-SECURITY (the contribution-surface blocker + integrity):**
+7. **Skills trust model (F113/F114/F115):** unsandboxed `skills run` + shell-injectable builtin + name-shadowing. HARD BLOCKER for the community-skills adoption pitch — sandbox (reuse @hooks srt/bwrap) or consent-gate before inviting external skills.
+8. **Destructive guards:** F84 (spawn delete: check live-PID + owner, not a 3-name allowlist), F85 (flow aggregator: never write `{}` over a real registry).
+9. **Owner model (F59):** key git-access to sealed registry_id not the mutable passport; protect passport.json. Gates fail LOUD (F65/F100/F103).
+
+**P2 — adoption (mostly Patrick, ~hours not days — see DPLAN-0240):**
+10. Tier 0 (30 min): GitHub topics, homepage, pin discussion, CODE_OF_CONDUCT. Tier 1: external-response reflex + protected good-first-issue shelf (the S304 [GFI] drafts are ~10 starters). Fix F22/F23 (concierge help), F39 (PyPI images), F36 (Windows quickstart).
+
+**P3 — ops rituals (the empty-janitor fix, F81):** enable daemon jobs for error-registry triage, backup cadence, presence cleanup, CB-recovery tick, Commons digest. The scheduler runs; nothing's scheduled.
+
+**The meta-fix (F117/provenance):** add INVARIANT checks that assert the invisible and fail loud (rollover leaves exactly N, medic is on, every registered checker ran, breaker not wedged, no over-cap entries persist). This is what would've caught most of the 116 months ago.
+
+## APPENDIX — Ready-to-file issue drafts (NOT filed; Patrick's go required)
+
+Each block is copy-paste ready for `drone @git issue create`. Small ones marked [GFI] are `good first issue` candidates for the DPLAN-0240 shelf. **Numbers are discovery-order labels, not priority and not sequential** (0a/0b are the criticals; renumber on filing). Priority order is in the Executive Summary. ~29 drafts total spanning F1–F106.
+
+0a. **[CRITICAL] ops(trigger): medic is OFF — re-enable it (in the right order)** — `config.medic_enabled=false` since 2026-05-10 = 63 days of undelivered error notifications (F5b). Sequence: (1) cut fixture storm F43, (2) fix breaker recovery F5c, (3) `drone @trigger medic on`. Re-enabling first just re-trips the breaker. Investigate WHY it went off 05-10 (deliberate storm-silencing never undone?).
+0b. **[HIGH] fix(trigger): circuit breaker can't self-heal** — half_open is terminal (never closes on success), cooldown never decays (pinned at max), transition only runs in the dispatch path (no tick). Must be fixed before medic re-enable. Manual unblock: `drone @trigger errors circuit-breaker reset`. File: error_registry.py:253/262/286. (F5c)
+
+1. **fix(prax): make test_mtime_cache_avoids_reread deterministic** — The test relies on two writes sharing an mtime tick (coin flip; red on all GH runners, PR#696). Pin mtime equality with os.utime after the second write, mirroring test_mtime_change_triggers_reread's forced-difference approach. File: src/aipass/prax/tests/test_telegram_relay.py:441. (F1)
+2. **fix(aipass): doctor false alarms on healthy installs** — (a) add `.backup`+`dropbox` to `_SCAN_SKIP_DIRS` (structure_scanner.py:96) so backup mirrors aren't counted/flagged as agents; (b) resolve registry `path` against PROJECT ROOT not CWD (structure_scanner.py:313) — the 5 "missing" entries have RELATIVE paths, this is NOT a casing bug (corrected); (c) substitute {{BRANCHNAME}} in pollution messages [GFI]; (d) align ✓/!/✗ glyphs with message content (role: unknown ≠ pass). (F14-F17)
+3. **fix(spawn): repair must never flag passport-seated dirs as pollution** — src/aipass/aipass currently flagged; printed remediation would archive the live concierge. Guard: skip dirs containing .trinity/passport.json with a live registry seat. (F30)
+4. **fix(prax/tests): route test logging out of production logs/** — pytest fixtures land in real branch logs, 24/7 trigger-log-watcher ingests them, error registry accumulates fixture noise (664+ occurrences of one /tmp config alone). Honor PYTEST_CURRENT_TEST in the logger path resolution. (F43/F6)
+5. **feat(trigger): error-registry lifecycle** — auto-purge stale (purge exists, nothing calls it: daemon job), stop double-filing UNKNOWN+component duplicates, populate source_file/source_branch, widen ID column or accept unique prefixes in `errors detail`. (F5/F7/F8)
+6. **fix(aipass): human help for the human tool** — `aipass --help` should describe AIPass + first 3 commands, hide doctor_wire/doctor_fix internals, put init first [GFI-ish]; `aipass help` Q&A needs domain-aware retrieval or a curated FAQ for top-20 questions. (F22/F23)
+7. **fix(drone): show_help teaches two broken examples** — drone.py:107 `drone @flow status` (real: `@flow registry status`), drone.py:110 `drone audit` (unregistered shortcut). [GFI] (F27)
+8. **fix(daemon): activity_report enforces retired memory schema** — "No 'limits' field" fails all 17 branches; freshness labels contradict (0m ago = WARNING). Re-point at memory.config.json contract. (F48/F49)
+9. **fix(skills/prax): TG control-file hardening** — writer temp+rename; reader keeps last-good cache on parse error instead of failing open to unpaused. (F33)
+10. **chore(docs): PyPI images broken** — README relative asset paths don't resolve on PyPI; use absolute raw URLs. [GFI] (F39)
+11. **docs(readme): Windows quickstart truth** — `./aipass install` requires bash (Git Bash/WSL); README Quick Start doesn't say so while Roadmap claims Windows-native. One sentence + optional .ps1 bootstrap issue. [GFI] (F36)
+12. **chore(spawn): normalize the 5 malformed registry entries** — BACKUP/COMMONS/DAEMON/HOOKS/SKILLS are BOTH uppercase-named AND relative-pathed (all other 12 are lowercase+absolute) = registered by an old/different code path. Normalize to lowercase+absolute AND find/fix the registration path that produced the wrong format. Fixes F15's real cause + the casing leaks (F69/F77/lint). (F15)
+13. **fix(commons): artifact ownership casing** — trade_ops._resolve_branch_name must route through identity_ops's lowercase choke point; add `COLLATE NOCASE` to owner comparisons or a one-shot data fix for existing uppercase owners. (F52)
+14. **chore(commons): standardize conn=None + finally: close_db idiom** — ~14 handler files, 40+ sites; posts/comments ops are the reference implementation. [GFI — mechanical, great first PR] (F53)
+15. **fix(commons): json_handler atomic writes + stop silent log reset** — port backup's temp+rename json_handler; corruption should quarantine, not reset to []. (F54)
+16. **fix(backup): concurrency guards** — lock around drive-upload tracker dict; file-lock the timestamps/changelog/registry read-modify-write sequences. (F55/F56)
+17. **[CORE] feat(fleet): atomic_write_json helper + seedgo checker** — one shared temp+rename+lock primitive; checker forbids raw open("w")+dump on shared *_data/*.json/inbox/runstate/registry. Retires F33/F54/F67/F79/F87/F89/F90/F6-family at once — the single biggest systemic win (a dozen findings collapse into one helper + one checker). (F73)
+23. **[HIGH] fix(spawn): delete_branch liveness+owner guard** — refuse to delete a branch with a live PID or `owner:true`; the 3-name allowlist is the wrong gate. (F84)
+24. **[HIGH] fix(flow): aggregator must not overwrite a branch registry it read as empty** — distinguish missing-vs-corrupt; never write `{}` back over a real registry; atomic write. (F85)
+25. **[HIGH] fix(spawn): structural registry locking** — every save_registry site takes the flock, not just some. (F86)
+26. **[MEDIUM] fix(prax): atomic module-registry + logger op-log writes, fix setup TOCTOU** — apply the existing atomic helper + double-checked lock. (F87/F89)
+18. **[SECURITY] fix(drone/auth): resolve owner via registry_id/is_owner, protect passport.json** — owner git-access must key to the sealed registry owner:true (machinery exists, S290), not the mutable passport branch_name; add passport.json to a gate. (F59)
+19. **fix(ai_mail): lowercase recipient before lookup** — mirror wake.py:466's `.lower()` in get_branch_by_email/delivery/resolve. [GFI] (F69)
+20. **fix(ai_mail): test_token reads wrong field** — `msg.get("body")` → `msg.get("message")`; add a test. [GFI] (F70)
+21. **fix(daemon): atomic runstate write** — temp+rename+lock save_runstate; fire-AFTER-persist or idempotency key. (F67/F71)
+22. **[SECURITY] harden gates** — edit_gate should cover Bash writes (F60); git_gate catch interpreter-wrapped + path-qualified git (F61/F62); gates fail LOUD not silent-open (F65). Scope: coordination not OS-security — frame accordingly.
+27. **[HIGH] fix(cli): display helpers must not crash callers** — `header`/`success`/operation templates need `markup=False` or escaped interpolation; they take down any branch on bracket-containing input. Blast radius = every branch. (F94)
+28. **[HIGH-SEC] fix(api): atomic 0o600 OAuth token write** — use `os.open(..., 0o600)` (pattern exists at secrets.py:154) so a live refresh token is never world-readable or truncated. (F95)
+29. **fix(api): don't relabel real command failures as 'unknown command'** — distinguish handler-raised from unrecognized. (F96)
+
+## CORE-BRANCH CODE SWEEPS (ai_mail, daemon, commons, backup — sub-agent found, sharp claims I verified)
+
+### F67 — daemon: non-atomic unlocked runstate write can wipe ALL job history → mass re-fire [HIGH]
+- `runstate.py:50-60 save_runstate` = direct open("w")+json.dump, no temp+rename, no lock. Crash mid-write (the per-job save in run.py:246) truncates the file; `load_runstate` catches JSONDecodeError and silently returns empty → next tick treats EVERY job on EVERY branch as never-run (daily/hourly/interval all "due" at once, completed `once` jobs re-fire). Blast radius = whole scheduler. Same write-atomicity gap as F54/F67 family.
+
+### F68 — ai_mail: inbox READS bypass the lock writers hold → user-visible "Invalid inbox JSON" [MEDIUM]
+- `inbox_ops.py:63 load_inbox` json.loads with NO lock while writers hold `inbox_lock()` and truncate-then-write. A concurrent `drone @ai_mail inbox/view` can read mid-truncation → JSONDecodeError surfaces as failure. The `.inbox.lock` exists; the read path just doesn't use it.
+
+### F69 — ai_mail: recipient casing breaks send/inbox for `@Branch` [MEDIUM — verified]
+- **Verified:** `registry/read.py:148 get_branch_by_email` does exact `branch["email"] == email`, delivery/resolve never lowercase user input → `drone @ai_mail send @Devpulse …` fails "Unknown branch" though @devpulse exists. **wake.py:466 resolve_branch DOES `.lower()`** (I read both) — so dispatch normalizes, plain send doesn't. Same casing disease as F15/F52, third surface.
+
+### F70 — ai_mail: test-token auto-ack is DEAD CODE (wrong field name) [MEDIUM — verified]
+- **Verified:** `test_token.py:132` reads `msg.get("body","")` but the schema stores content under `"message"` (create.py:86 `"message": message_with_footer`; "body" set nowhere). So `has_test_token` always sees "" → never matches → liveness/test pings fall through to full dispatch and wake a whole Claude agent instead of a cheap ack. No test covers this handler (how it shipped broken). Ties to F41 (delivery-failure notices as malformed mail) — the test-ping path is unexercised.
+
+### F71 — daemon: fire-then-persist ordering allows duplicate wake after crash [LOW-MEDIUM plausible]
+- run.py:237-249 fires the agent (durable side effect) THEN saves runstate. Killed in the gap → fire unrecorded → next tick re-fires unless the prior agent's dispatch.lock still held. No idempotency key. Narrow window, real.
+
+### F72 — ai_mail: ~120 lines of lock/occupancy logic duplicated wake.py vs daemon.py [LOW — divergence risk]
+- `_check_lock`/`_acquire_lock`/`_is_branch_occupied`/`_pid_alive` copy-pasted between manual-wake and daemon-dispatch paths. Consistent now; a future fix to one copy silently diverges the two. (Same shape as the three structure-validators, F31.)
+
+### F73 — Atomic-write helper is the missing shared primitive [MEDIUM — meta-finding]
+- F33(control), F54(commons log), F67(runstate), F6-family(status.py dispatch log) are ALL the same bug: `open("w")+dump` on shared state, no temp+rename, no lock. backup's json_handler and ai_mail's inbox_lock do it RIGHT — the correct pattern exists in-tree, just isn't centralized. One `atomic_write_json()` helper + a seedgo checker forbidding raw dump-to-shared-file would retire a whole class. Strongest single systemic fix from the sweeps.
+
+### F74 — Memory rollover off-by-one is LIVE fleet-wide: keeps 14, not 15 [CRITICAL — verified on disk]
+- **Verified fleet-wide:** memory, drone, hooks, seedgo, flow `.trinity/local.json` ALL hold exactly **14 sessions / 14 key_learnings** (predicted keep-14 steady state; devpulse shows 15 only because I hand-edit, bypassing rollover). `orchestrator.log`: rollover fired at **"(15/15 sessions)" 53×**, "(15/15 observations)" 30×, "(15/15 key_learnings)" 18× — vs "(16/15)" only 3×. It rolls over AT the keep target, not above it. This is not a spot bug — it's every branch, every rollover, since the trigger was written.
+- **Mechanism (exact lines):** trigger `len(sessions) >= max_sessions` (detector.py:360 + extractor.py:207) fires at len==15; extractor `excess = max(len(sessions) - max_sessions, 1)` (extractor.py:208/218/228 for sessions/key_learnings/observations) forces trimming 1 even when real excess is 0 (`max(0,1)`). Every branch silently runs keep-14.
+- **PRECISE FIX (2 changes):** trigger `> max_sessions` (fire only when EXCEEDED, so 16 rolls to 15) AND drop the `,1` floor → `excess = len - max_sessions` (naturally ≥1 when the >-trigger fires). Apply to all three entry types (207-208/217-218/227-228). @memory-owned.
+- **FIX PROVEN EXECUTABLE (isolated logic replica, ran it):** shipped logic at exactly 15 entries → keeps **14** (bug); at 16 → keeps 15. Fixed logic at 15 → keeps **15** (correct); at 16 → keeps 15. All three assertions pass (`current(15)==14`, `fixed(15)==15`, `fixed(16)==15`). Not just asserted — demonstrated. The 2-line change is safe and correct.
+- **Dated + why-unnoticed:** `git blame` → introduced **2026-04-22 (commit 9634c5639)** — silently keep-14 fleet-wide for ~2.5 MONTHS. It survived because rollover ARCHIVES the trimmed entry to vectors (nothing is deleted, it just moves to @memory one cycle early) → **zero visible symptom.** The perfect silent bug: on the branch named `memory`, in the system whose pitch is "memory persists," a memory-loss bug is invisible precisely because the memory isn't lost, just archived early. This IS todo 66's "15/15 vs keep-15" — PROVEN actively trimming. **NOT filing** (todo 66: Patrick monitors rollover, no file without his go) — documented only.
+
+### F75 — ai_mail error-escalation channel reports success on failure [HIGH]
+- `error_dispatch.py:61-88`: `deliver_fn("@drone", ...)` return discarded, hard `return True`. `deliver_email_to_branch` returns `(False, msg)` on failure (doesn't raise) → a failed escalation logs as success. The incident-visibility safety net can't see its own failures. Both call sites discard the result too. (Explains how F41's malformed vera notices piled up unnoticed.)
+
+### F76 — drone: ~90 git/gh subprocess calls with NO timeout; `drone @git pr` can hang holding the repo-wide lock [HIGH]
+- Only 2 of ~90 `subprocess.run` git sites pass `timeout=`. `pr_handler.py:163→220`: `acquire_lock()` takes repo-wide `.git_pr.lock`, then UNBOUNDED `git push`; a credential/SSH/net stall → function never returns → `finally` never runs → lock never releases → every branch's `drone @git pr` blocked until a human force-unlocks (staleness is passive 600s, no auto-unlock). Real hang risk for the one git-write path the whole fleet shares.
+
+### F77 — drone: dict-shaped registry keys never lowercased → `@Name` permanently unresolvable [HIGH — latent]
+- `registry_handler.py:273` lowercases names only when `branches` is a LIST; dict-shaped `branches` keys pass through untouched, while every lookup forces lowercase. With `{"branches":{"Prax":...}}`, `@prax` AND `@Prax` both fail. Dormant TODAY (prod registry is list-format — I verified) but a landmine if anything emits dict format. Casing-drift family (F15/F52/F69).
+
+### F78 — drone registry credential check FAILS OPEN [MEDIUM — security-adjacent]
+- `registry_handler.py:105/217 _verify_registry_credential`: bare `except Exception → return True`. A corrupt/unreadable passport is treated as "credential matches" → cwd walk-up may adopt another citizen's registry. Same fail-open theme as F65, violates "fail to errors."
+
+### F79 — drone custom-command registry: non-atomic write + read-modify-write race [MEDIUM]
+- `command_registry/ops.py:143` raw open("w")+dump (while the SAME tree's `json_handler._atomic_write_json` does it right — F73 again); unlocked add/remove/update. Concurrent `drone activate` → last-writer-wins drops commands; kill mid-dump → `load_registry` silently recreates EMPTY registry, all shortcuts lost.
+
+### F80 — Verified-clean by the memory sweep [ASSET]
+- Memory rollover ORDERING is correct (backup→trim→embed→store, restore_from_backup on every failure path); primary memory files DO use atomic temp+os.replace; the old 30s-hook false-FAILED is resolved (heavy ops now 60/120s, zero timeout hits in logs). ai_mail wake/dispatch is poll-based (no write-vs-signal race), O_CREAT|O_EXCL locks, consistent subprocess timeouts on the dispatch side.
+
+### F81 — The daemon scheduler works but the fleet has ZERO production jobs [HIGH — the empty janitor]
+- The whole automation layer (systemd daemon-tick.timer @1m + decentralized `.daemon/schedule.json` discovery) is BUILT and running — I ran a manual tick, it discovered and evaluated correctly. But across all 17 branches there are exactly **3 jobs, all `wake-test`, all disabled** (F11). Nothing is scheduled: no error-registry triage (→ F5 graveyard), no backup cadence (→ F29 4-day-old backups), no commons digest (→ A2 empty Commons), no stale-presence cleanup (→ F26), no CB-recovery tick (→ F5b stuck 63 days). 
+- **This is the single infrastructure root of the "write side shipped, read side missing" pattern.** The janitor-RUNNER exists; nobody wrote the janitors. Every "nothing ever cleans/triages/recovers X" finding could be closed by a handful of enabled daemon jobs. Highest-leverage systemic fix on the ops side — and it's additive, low-risk (jobs are per-branch JSON).
+
+### F82 — commons `welcome_new_branches` auto-post exists but is never triggered [MEDIUM]
+- `welcome/welcome_handler.py:116 welcome_new_branches` + `run_welcome` are built to auto-post welcomes for new branches — exactly the ritual that would keep the Commons alive (A2/A6). It's wired to a command, not to any event or schedule, so it never fires on its own. Another built-but-unpulled ritual; a daemon job (F81) or a spawn-hook would light it up.
+
+### F83 — .backup store is 951M (versioned 671M) — growth vs max_versions:10 worth a look [LOW]
+- `.backup/versioned` = 671M, `.backup/snapshots` = 273M (25 snapshot dirs), `drive_tracker.json` = 4.9M. Backup config says `max_versions: 10` but the versioned store is large — either per-file baselines+diffs legitimately accumulate or pruning isn't keeping pace. Correctly gitignored (verified). Not urgent; worth a `backup` prune audit given F29 (no scheduled backups anyway). NOTE: initial `git ls-files` count looked alarming (62) but was a CWD artifact — real tracked count is 1843 (1304 py). Verified before recording.
+
+### F93 — tier0_kernel loader docstring says "period 1", config + kernel header say period 5 [LOW — doc drift]
+- `tier0_kernel.py:32` docstring: "Load tier0 kernel — every turn (cadence period 1)." But cadence_config sets `tier0: period 5`, and the kernel file's own header says "injected every 5 turns (cadence period 5)". The docstring is stale. (Verified the `branch` loader's MISSING period is fine — cadence.py:204 inherits global_period=5 correctly; that one's not a bug.)
+
+## FLOW / SPAWN / PRAX SWEEP — highest blast radius (sub-agent found; scariest two I verified in code)
+
+### F84 — spawn `delete_branch` has NO liveness or owner guard [HIGH — destructive]
+- **Verified:** sole gate is `_PROTECTED_BRANCHES = {spawn, devpulse, drone}` (delete_ops.py:124) + `is_dir()`. No PID/heartbeat check, no `owner:true` check. `drone @spawn delete @<any-live-non-protected-branch> --yes` archives + rmtrees it whether or not it's running, and the sealed registry OWNER is deletable if not one of the 3 hardcoded names. Highest single-command destructive risk found. (Mitigant: it's an intentional admin verb with a confirm prompt + spawn is owner-tier — but the guard SET is wrong: it should be "not owner AND not live," not a 3-name allowlist.)
+
+### F85 — flow central aggregator can overwrite ANOTHER branch's plan registry with empty [HIGH — cross-branch data loss]
+- **Verified the mechanism:** `aggregate_ops.py:86 load_branch_registry` fails open to `{"plans":{}, "next_number":1}` on ANY read/parse exception (no missing-vs-corrupt distinction); `save_branch_registry:94` writes back with raw `open("w")+json.dump` (non-atomic). The aggregator's heal pass runs against OTHER branches' registry.json. If branch B's registry is transiently unreadable (mid-write/truncated) when the heal runs → aggregator reads empty → "heals" by overwriting B's whole plan history with `{}`. A read hiccup in one branch, triggered by another branch's routine aggregation, destroys plan tracking.
+
+### F86 — spawn registry lost-update race: locking is opt-in, not structural [HIGH]
+- registry.py:159/repair_ops take fcntl.flock before load→modify→save; but `delete_ops._remove_from_registry`, `sync_registry_ops` (2 sites), and registry.py:333 call `save_registry` with NO lock. flock is advisory → unlocked writers get zero protection. `delete @foo` (loads registry, then blocks on confirm + slow copytree) racing `create @bar` (locked, fast) → delete writes stale registry back, erasing @bar.
+
+### F87 — prax module registry truncates to ONE entry → ecosystem-wide log-routing loss [HIGH]
+- `registry/save.py:96` raw open("w")+dump (ignores the atomic helper next door); `watcher.py on_created` does unlocked load→mutate→save. Kill mid-dump truncates `prax_registry.json`; load fails open to `{}` → next watcher event overwrites with just the one new module, discarding every previously-discovered module until a full rescan. This is the module→log-routing registry.
+
+### F88 — spawn: 2 passport writers bypass the atomic helper [MEDIUM — identity loss]
+- `sync_registry_ops.py:603/628 fix_owner_identity` use raw `passport_path.write_text(json.dumps(...))` while every OTHER spawn passport writer routes through the mkstemp+fsync+os.replace helper. Crash mid-write → truncated passport → a branch loses its identity (and per F59, its git-access key).
+
+### F89 — prax logger internal op-log + setup have corruption/TOCTOU races [MEDIUM]
+- `logging/operations.py:60` raw open("w") on the growing op-log, no lock → concurrent callers interleave partial writes → invalid JSON (hit from every handler). `logging/setup.py:93` checks `_captured_loggers` under lock, RELEASES, then mutates the stdlib singleton logger outside the lock → duplicate handlers (dup log lines) or dropped handler. The correct double-checked-lock pattern exists in logger.py:95 but isn't applied here.
+
+### F90 — flow own plan registry + central aggregate non-atomic, fail-open-to-empty [MEDIUM]
+- `registry/save_registry.py:70` raw-writes `fplan_registry.json` (self-labeled DO NOT EDIT); `aggregate_ops.py:257 save_central` raw-writes `PLANS.central.json`. Both paired with fail-open-to-empty loads → a crash mid-write + any later load+save silently resets the registry. Same F73 atomic-write class.
+
+### F91 — flow "read-only" scan silently RENAMES plan files across branch boundaries [MEDIUM — plausible]
+- `monitor_ops.py:189 scan_plan_files_impl` unconditionally `rename`s on a 4-digit-number collision even though the CLI reports "no changes applied." It walks from ECOSYSTEM_ROOT across ALL branches with no branch-boundary awareness; plan numbers are PER-BRANCH, so two branches legitimately holding e.g. FPLAN-0042 → one gets renamed out from under its owner by a "read-only" scan.
+
+### F92 — Verified-clean by this sweep [ASSET]
+- Mixed-case @branch bug NOT present in flow/prax (prax consistently `.upper()`, flow delegates upstream). prax atomic-write PRIMITIVE is correct (just not used in 3 spots). spawn individual writes mostly careful (path containment, archive-before-delete). The correct patterns exist in every branch — the gaps are inconsistent APPLICATION, not absence.
+
+## API / CLI / TRIGGER SWEEP (sub-agent; CLI-crash + OAuth window I verified)
+
+### F94 — CLI display helpers crash the CALLER on bracket/markup-like input [HIGH — every-branch blast radius]
+- **Verified:** `display.py:328 header` and `success` interpolate caller strings into a Rich-markup-parsed `CONSOLE.print(f"...[dim]{key}:[/dim] {value}")` / `Panel(f"[bold cyan]{title}[/bold cyan]")`. Any value with an unmatched/closing tag — a path, git ref, JSON, regex, or exception text containing `[/x]` — raises `rich.errors.MarkupError` uncaught and takes down that branch's process. Nasty asymmetry: `error()`/`warning()`/`fatal()` use markup-safe `Text.append()` — only the HAPPY-PATH helpers crash, so a "success" message kills the app. Zero test coverage. This is the shared display layer every branch renders through — F8's truncation was the cosmetic tip; this is the crash. Fix: `markup=False` or escape interpolated values.
+
+### F95 — api: live OAuth refresh token can be left world-readable / truncated on crash [HIGH — secret exposure window]
+- `google/auth.py:251 _save_credentials` (runs after EVERY refresh): `open(path,"w")` → write token → `os.chmod(0o600)`. File is created at umask (usually 0o644 = world-readable) and only tightened AFTER the write; a crash between write and chmod leaves `google_creds.json` (live refresh token) permanently world-readable, and nothing re-chmods later. Also non-atomic (no temp+rename). The correct pattern (`os.open(..., 0o600)` — mode atomic at creation) is ALREADY used at `secrets.py:154` and `api_key.py:234`. `env.py:94` has the same ordering (lower sev, placeholder only).
+
+### F96 — api: dispatcher misreports real command failures as "unknown command" [MEDIUM — extends F13]
+- `api.py:240 route_command`: module loop `except Exception → log + continue`. When a REAL handler (e.g. api_key on a corrupt .env / perms error) raises, no module claims it, route returns False, main prints "Unknown command: get-secret" (exit 1). The true cause is buried in logs. Broader than F13's silent-fallback — it actively mislabels legitimate credential-command failures.
+
+### F97 — api: `diagnose_key` echoes a real secret PREFIX to stdout [LOW — minor leak]
+- `auth/keys.py:216 diagnose_key` puts the first ~6-10 chars of the actual stored secret into a string that reaches stdout (openrouter_client.py:289). Prefix only, not full value, but a real secret fragment on a "diagnostic" path meant to be safe.
+
+### F98 — Verified-clean by this sweep [ASSET]
+- api secret handling is DISCIPLINED: no raw key/token value hits any logger/print across apps/ (all masked `key[:6]+"****"+key[-4:]`); no bare `except: pass` around auth; OAuth refresh failure is LOUD (logs error, returns False, callers report "invalid" — doesn't treat stale token as valid). cli `error/warning/fatal` are markup-safe. The secret-masking discipline is genuinely good — F94/F95 are the gaps in an otherwise careful branch.
+
+## SECURITY GATES — integrity audit (sub-agent found, core claims I verified in code)
+
+**Threat-model framing (READ FIRST):** AIPass agents all run as the SAME OS user, locally, cooperatively. The gates are Claude Code hooks — coordination + accident-prevention, NOT an OS boundary against a hostile actor (who could run git directly). So "bypass" here means "the owner/ownership model is ADVISORY, not enforced" — an integrity/robustness gap, real and worth fixing to make the model mean what it claims, but NOT a remote exploit or live emergency. Severities below are relative to the model's own promises.
+
+### F59 — Owner git-access trusts an UNPROTECTED, UNSIGNED passport [HIGH — the owner model is forgeable]
+- **Verified in code:** `auth.py:_find_caller` reads `branch_info.branch_name` straight from the CWD-hierarchy's `.trinity/passport.json` (`auth.py:18-48`); `verify_git_access` grants owner-tier if that string is in `allowed_callers` (["devpulse"]) — **no registry cross-check, no signature** (`auth.py:88-125`). And NO gate protects passport.json (edit_gate covers only local.json/observations.json; git_gate covers settings/hooks; registry_gate covers *_REGISTRY.json). So any dir with a passport saying `branch_name: devpulse` gets owner git-write. The whole DPLAN-0231 owner-capability model (built to key auth to the immutable registry_id) is undercut because the ACTUAL check reads the mutable passport name, not the sealed registry owner:true. Fix: `verify_git_access` should resolve owner via registry_id/is_owner (the machinery EXISTS — S290), not a passport string; and protect passport.json under a gate.
+- Same passport-trust pattern duplicated in `seedgo/permissions.py:identify_caller`.
+
+### F60 — pre_edit_gate never runs for Bash → all its protections void via shell writes [HIGH]
+- Sub-agent claim (matcher-traced): edit_gate's matcher is `Edit|MultiEdit|Write|NotebookEdit` (no Bash) AND its `EDIT_TOOLS` set excludes Bash. So `echo … > inbox.json`, `python3 -c "open(...).write(...)"`, `tee`, `sed -i` skip edit_gate entirely — voiding its inbox-write block, daemon confinement, cross-branch block, and .trinity entry-limits. (Consistent with F3's observation that git_gate DOES match Bash but edit_gate doesn't — asymmetric tool coverage across gates.)
+
+### F61 — git_gate quote-stripping blinds it to interpreter-wrapped git [MEDIUM]
+- git_gate replaces quoted-string contents before scanning (git_gate.py:164), so `bash -c 'git push'`, `sh -c "git push"`, `eval 'git push'`, `python3 -c "subprocess.run(['git','push'])"` pass. Heredoc-piped git IS caught (bodies not stripped) — so the miss is specifically the interpreter-string idiom.
+
+### F62 — RAW_GIT_RE lookbehind excludes `/` and `.` → path-qualified git evades [MEDIUM]
+- `(?<![@\w/.])git\s` (git_gate.py:21): `/usr/bin/git push`, `./git push`, `GIT_SSH_COMMAND=x /usr/bin/git push` all pass. Exclusion was added to spare `.gitignore` filenames; it also blinds the matcher to the real binary by absolute/relative path. (Variable-indirection `g=git;$g push` is inherent to any regex approach — same limit hits rm_gate/registry_gate.)
+
+### F63 — rm_gate is the ONLY interactive delete defense yet has clean misses [MEDIUM-HIGH]
+- The kernel sandbox rm_gate calls "the real boundary" is Phase-1, NOT wired into normal sessions (sandbox.py:267 says so; devpulse's own tools/rm_shim/FINDINGS.md already documents this). So rm_gate's misses are load-bearing: `find . -delete`, `shred -u`, `\rm -rf` (endswith-check miss), `find|xargs sh -c 'rm -rf {}'` (quote-stripped), `python3 -c "shutil.rmtree(...)"` all pass. Straight `rm -rf`, `rm -fr`, `find|xargs rm -rf` correctly blocked.
+
+### F64 — Protected-path gates match raw strings, not resolved paths [MEDIUM]
+- git_gate BLOCKED_EDIT_PATTERNS + registry_gate check the literal `file_path`, no `Path.resolve()`. `.claude/settings.json` (relative, no leading `/`) passes where `./.claude/settings.json` is caught — inconsistent. A symlink whose name isn't `*_REGISTRY.json` but points at one evades registry_gate (PLAUSIBLE, depends on Write following symlink).
+
+### F65 — All gates + the engine FAIL OPEN on exception [MEDIUM — by design, but]
+- Every gate's `handle()` wraps in `try/except Exception → allow + log "(allowing)"`; engine.py only special-cases exit_code==2, any other (incl. handler import crash, exit -1) falls through to allow. Deliberate availability tradeoff (a hook bug shouldn't brick all tools) — but means ANY parser crash = silent bypass. At minimum the fail-open should be LOUD (surface to prax/error-registry, not just a log line).
+
+### F66 — What the gates get RIGHT [ASSET]
+- git_gate read/write split is an ALLOWLIST of read verbs (unknown → blocked = safe default). Compound-command splitting (`&&`/`||`/`;`/`|`/`$()`/backticks) is consistent across git/rm/registry gates and defeats naive chaining. Heredoc bodies not quote-stripped (piped-heredoc git caught). rm_gate follows single pipes. `drone <verb>` self-exemption uniform. The bones are good; the misses are specific parser gaps, not a broken design.
+
+## HOOKS ENGINE / SEEDGO SWEEP (final sweep — two trust-undermining finds I verified)
+
+### F99 — seedgo can silently STOP ENFORCING a standard AND drop crashing files → false 100% [HIGH — undermines "100%", two layers]
+- **Layer 1 (worse — whole checker vanishes, seedgo-audit sub-agent found, I verified):** `branch_audit.py:37-39 discover_checkers` — `except Exception: logger.info("Skipped checker %s: failed to load"); continue`. If a `*_check.py` fails to IMPORT (syntax error, missing env dep), it's silently dropped from the checker set → absent from scores AND gating → CI (THRESHOLD=100) passes the branch at 100% **while that entire standard goes unenforced, zero signal.** Break one import and a rule silently stops being checked fleet-wide.
+- **Layer 2 (file-level):** `branch_audit.py:95-97 _run_all_files` — `except Exception: continue`. A file that breaks a checker (non-UTF-8, parser edge, checker bug — AST checkers catch SyntaxError but not UnicodeDecodeError etc.) is dropped from `scores`, absent from the denominator → average rounds UP. Plus line 101: any file with a "skipped"/"not applicable" check message is excluded from the average even if another check on it FAILED.
+- **This means the fleet "100%" (F47/F92) is trustworthy only for files+checkers that PARSE/IMPORT — blind to anything that chokes.** Fix: exception in discover_checkers/`_run_all_files` = FAIL (score 0) or hard error, never silent skip. Fix F1 (layer 1) first — a standards enforcer that can silently stop enforcing is the worst failure mode here.
+- **Dated:** `git blame` → both silent-skip paths date to **2026-03-23 (commit 6bd1bd00f)**, near the audit's inception. So every "100% fleet-green" this project has ever celebrated has carried this blind spot from the start — foundational, not a regression.
+
+### PROVENANCE — the big findings are OLD and symptomless, not fresh breakage [synthesis]
+- Dated the load-bearing ones via git blame / logs: **F74 rollover keep-14 = 2026-04-22** (~2.5 months); **F99 seedgo silent-skip = 2026-03-23** (inception); **F5b medic disabled = 2026-05-10** (config toggle, log-confirmed, 63 days); **skills json_handler F116 = 2026-03-17** (never migrated). Pattern: these survived MONTHS not because they're subtle to find but because they're **symptomless** — rollover archives (doesn't delete), medic-off just means silence, seedgo-skip just inflates a number, atomic-write races only bite on a crash. The system has no alarm for "a thing quietly stopped working correctly." That's the deepest gap: **AIPass optimizes for visible-failure (logs, errors, red CI) and is blind to silent-degradation.** The fix class isn't per-bug — it's invariant checks that assert the INVISIBLE (rollover leaves exactly N, medic is on, every registered checker ran, the breaker isn't wedged) and fail LOUD when violated.
+- **DEEPEST UNIFICATION — the observability layer is decorative, not measured (this is WHY degradation stays silent):** the runtime/data probes (F123/F124) showed the mechanism. AIPass's self-reported STATUS is systematically wrong while its underlying DATA is sound. Vector store: healthy 5,012 vectors, dashboard says 1,274. Dashboards: "live", actually up to 10 days stale. Doctor: 7 false errors on a healthy repo. activity_report: retired schema, 100% noise. daemon: contradictory labels. seedgo: "100%" blind to files it chokes on. Meanwhile the DATA is fine — compass integrity ok, chroma integrity ok, 11k tests collect clean, 6 suites green. **The data is trustworthy; the gauges are stale, mislabeled, or lying-green — and THAT is the silent-degradation mechanism.** You cannot notice a thing quietly breaking when every dial reads "fine" regardless of reality. The single highest-leverage meta-fix: make the observability layer MEASURED (real counts, real freshness, fail-loud invariants) before trusting any gauge as a signal. A striking share of these 124 findings is downstream of "nobody could see it."
+
+### F100 — hooks: missing/corrupt `.aipass/hooks.json` silently disables ALL security gates [HIGH — fail-open at config layer]
+- **Verified:** `claude.py:47 find_project_config() → None` on missing/corrupt config → falls back to `{"hooks_enabled": True}` with no event key → `engine.py` sees empty event_hooks → returns allow. Any CWD whose tree up to $HOME lacks `.aipass/hooks.json` gets ZERO enforcement — git_gate/rm_gate/edit_gate/registry_gate/presence_gate all no-op, logged only at INFO. **Sharp edge:** `isolation: worktree` sub-agents (and any /tmp extraction) created OUTSIDE the main checkout can lack the config → run ungated. Ties the security cluster (F60/F65) together: the gates fail open at the CONFIG layer too, not just on parser crash.
+
+### F101 — hooks presence.py is DEAD CODE presented as live [MEDIUM — confirms F26]
+- **Verified:** presence_gate v2.0 migrated source-of-truth to CC-native `~/.claude/sessions/<pid>.json` (cc_sessions.py); `presence.claim/release/refresh` are called ONLY from tests, zero production sites. `PRESENCE.central.json` is frozen pre-migration; nothing writes it. So F26's "12-day stale records" isn't a cleanup bug — the write path doesn't exist. Yet `drone @hooks presence` still renders the frozen entries with live/stale PID tags, actively misleading. Either delete the surface or repoint it at cc_sessions.
+
+### F102 — seedgo CI audit ≠ local audit (branch scope AND pass/fail) [MEDIUM — substantiates DPLAN-0198]
+- Same checker pack, but: (a) CI (`.github/scripts/seedgo_audit.py:14`) discovers branches via naive `src.iterdir()` (any dir with `apps/`, no registry, no private-branch exclusion) while local uses registry-based `discover_branches()`; (b) CI hardcodes THRESHOLD=100 + `sys.exit(1)`, local command has NO threshold at all (pure display, always returns True). A dev CANNOT get a "would this fail CI" signal from the normal local command.
+- **Sharper (seedgo-audit sub-agent):** the two also resolve the branch ENTRY FILE differently — CI only tries `apps/{name}.py`, local also falls back to `apps/branch.py`. A branch using the `branch.py` convention gets entry_file="" in CI → entry-point checkers open "" → score 0 → **CI false-FAILs a branch that passes locally.** So the parity gap cuts BOTH ways (false-pass on scope, false-fail on entry resolution). Also: `ci.yml:50 fetch-depth:0` comment still cites the deleted git-log freshness check. That's DPLAN-0198, concretely.
+
+### F103 — hooks engine fail-open: handler crash AND malformed stdin both skip gates [MEDIUM — confirms/extends F65]
+- Handler crash → exit_code -1 → falls through to allow (test_engine.py:211 literally asserts a crashed hook yields overall allow). Malformed stdin → `match_value=""` → `_matches` returns False for any NON-empty matcher — and every security gate uses a non-empty matcher while non-security handlers use empty ones, so malformed stdin skips EXACTLY the security hooks. Two more fail-open layers, text-log only.
+
+### F104 — hooks cadence miscounts turns <2s apart [LOW]
+- `cadence.py:137 _should_increment` checks mtime-age <2s BEFORE checking if the transcript token actually changed → a genuinely new fast turn (agentic/scripted exchanges) is treated as a same-turn straggler and doesn't increment. The "every Nth turn" injection silently falls behind real turn count on fast turns. No test <2s apart.
+
+### F105 — seedgo readme_currency proof is broken (explains F51) + one real drift [LOW]
+- **Verified by running scan():** `readme_currency.py:81` only recognizes a legacy prose pattern `pack checks: ...`; seedgo's README now uses a `## The 40 Standards` table, which the regex never matches → returns empty → flags all 40 as "undocumented." So F51's readme_currency FAIL is mostly a BROKEN PROOF, not real drift. BUT one genuine nugget: README.md:47 says "33 standards", actual is 40 — that line is real drift worth fixing.
+
+### F106 — seedgo 471s fleet audit: sequential loop + no AST parse cache [MEDIUM — perf]
+- `standards_audit.py:289` iterates branches with a plain sequential `for` (branches are independent — trivially parallelizable). `_run_all_files` re-parses each file once PER checker (7 of 40 do their own `ast.parse`) instead of once per file with a shared cache. Two clear wins to cut the 471s (F47) — matters because slow audits get skipped.
+
+## AIPASS + SKILLS SWEEP (final sweep — doctor findings LIVE-CONFIRMED + skills-security surface)
+
+### F113 — drone_commands built-in skill: `shell=True` with caller-supplied command [HIGH — footgun + false safety claim]
+- **Verified:** `skills/lib/drone_commands/apps/handlers/executor.py:63 subprocess.run(command, shell=True, ...)` where `command` is the caller's `args["command"]`. Reached via documented `drone @skills run drone_commands run --args '{"command":"..."}'`; args parsed by naive key=value split, zero escaping. Any `;` `\`` `$()` `&&` `|` runs arbitrary shell. **Scope honestly:** invoked locally with your own args it's just you running your own shell (not a privilege gain). BUT it escalates to real arbitrary-exec if args ever flow from an untrusted channel (Telegram→skill, a community skill calling it), and the SKILL.md's claim "never runs commands that modify system state without explicit action / only drone commands" is FALSE — `shell=True` defeats the intended containment. Fix: drop shell=True, exec argv list, or validate the command is a drone invocation.
+
+### F114 — `skills run` = unsandboxed in-process arbitrary code execution, ZERO gate [HIGH — THE contribution-surface trust model]
+- **Verified:** `loader_handler.py:92 spec.loader.exec_module(module)` on any discovered `handler.py`, then `runner_handler.py` calls `handler.run(...)` in-process; grep for sandbox/bwrap/confirm/input under skills/apps = ONLY the exec_module line (nothing else). `skills validate` only checks declared dep PRESENCE, and run never calls it. **This is the direct answer to A7/DPLAN-0209/0240's "open skills to the community":** dropping a folder in `~/.aipass/skills/` and running it == `python handler.py` with full process privileges — no sandbox, no confirm, no review. A community-skill ecosystem CANNOT ship on this as-is; it needs a trust model (signed/reviewed skills, a sandbox via the existing @hooks srt/bwrap wrapper, or an explicit consent gate) BEFORE inviting external skills. First-class design blocker for the contribution story, not just a bug.
+
+### F115 — skill-name shadowing: a project/global skill can silently impersonate a built-in [HIGH]
+- `discovery_handler.py:93` keys skills by the frontmatter `name:` field (attacker-controlled), NOT the dir name; `registry.py:36` is first-match-wins in order project → global → builtin. A project/global skill declaring `name: github` (or `drone_commands`) silently SHADOWS the trusted builtin, no warning. Compounds F113/F114: shadow a trusted skill name + get it run unsandboxed. Fix: key by dir/namespace, warn on name collision, builtins win or are namespaced.
+
+### F116 — skills json_handler non-atomic (F73 class, unmigrated) [MEDIUM]
+- `skills/apps/handlers/json/json_handler.py:127 open("w")+dump`, used on every skill run/create/validate via log_operation. `aipass/shared/json_handler.py` already has the correct mkstemp+fsync+os.replace — skills' copy (untouched since 2026-03-17) never got the migration. Corruption is silently self-healed (regenerate defaults) → invisible data loss. Another instance for the F73 helper+checker.
+
+### F14–F23 doctor/help findings — LIVE-REPRODUCED with exact lines [confirms my behavioral findings]
+- The sweep live-reproduced every doctor false-error I found behaviorally, pinning exact lines: **F14** (.backup not in `_SCAN_SKIP_DIRS`, structure_scanner.py:96 → 38 vs 19 agents + cascading false pollution "ai_mail 30 copies"); **F15 CONFIRMED my correction** (structure_scanner.py:312 resolves relative registry path against `cwd()` — from repo root 0 issues, from a branch subdir all 5 "missing"; it's relative-path-not-casing, exactly as I corrected); **F16** ({{BRANCHNAME}} source pinned: `spawn/templates/aipass_framework/.trinity/passport.json:13` shipped scaffold with unsubstituted placeholder, scanned as a real agent); **F17** (doctor.py:332 reads top-level `role` but schema nests `identity.role` → always "unknown", line 333 appends PASS unconditionally → always green); **F22** (aipass.py:111 merges --help/help/bare into one internals dump); **F23 root cause** (help_chat BRANCHES is a hardcoded 12-name list MISSING skills/backup/commons/daemon/hooks; unmatched query greps all branches by keyword-count with no domain scoring → "create a skill" returns drone/seedgo/prax). 5 relative-path "missing" + 2 backup-pollution = exactly the 7 false errors. All my doctor findings now line-pinned and reproduced.
+
+## SUB-AGENT CORROBORATION + NEW ITEMS (log-sweep, newcomer-audit, seedgo-audit — my own agents, reporting late)
+
+### F107 — @memory rollover COMMAND times out at 30s via drone, ~every 1-3h for 2+ days [MEDIUM — sub-agent reported]
+- **CLOSED (I verified):** the 15 timeouts (`Command timed out after 30s: apps/memory.py rollover run`) are ALL in the rotated `drone.log.1`, 2026-07-10 00:24 → **last at 07-11 18:23:00**; ZERO in the current log. A live `rollover status` now returns in **8.3s** (well under 30s), local memory "OK". So this was REAL and recurring for ~2 days but has NOT recurred in ~25h — a transient (likely embedding-model cold-load or a backlog that cleared), not currently active. Distinct from F74 (keep-14 off-by-one). It's the "false FAILED" symptom of todo 66 — the 30s drone routing timeout < the actual rollover-run time. Keep an eye out for recurrence; if it returns, bump the routing timeout for the `rollover run` path or make rollover incremental.
+
+### F108 — README/docs drift batch (newcomer-audit; adoption-facing) [LOW each, MEDIUM in aggregate]
+- README says "17 agents" ×3 but the Project Status table (README:245) says "13 core + user-created" — self-contradicting on the same page.
+- HVTrust badge (README:8) → hvtracker.net/agents/aipass returns HTTP 403 (re-check manually; may be bot-block vs down). This badge already has a saga (PR#655 hid it, #621 re-added).
+- Roadmap (README:262-269) frames #360/#329 as "under ongoing testing" — both are CLOSED per gh api.
+- **PyPI-vs-clone contradiction [MEDIUM — adoption]:** package `aipass` v2.7.0 IS live on PyPI, but TDPLAN-0010 stripped all `pip install aipass` refs from the README in favor of clone-only. A PyPI discoverer lands on a page whose own README tells them to git-clone instead, no explanation. Either document the PyPI path or explain the redirect.
+- Stale "Citizen Class: builder" survives the 2026-07-01 builder→aipass_framework rename in commons/README:8 and daemon/README:8.
+- daemon/README internal date contradiction (header "2026-04-07" vs footer "2026-06-29"); commons/README self-inconsistent post arg-count (3-arg vs 2-arg).
+- CONTRIBUTING.md:16 "4,900+ tests" — actual ~12k `def test_` (stale ~2.5×). 155KB CHANGELOG (excellent, root-cause+verify per fix) is NOT linked from README.
+
+### F109 — Missing contributor infra: no PR template, no CoC, no FUNDING, no good-first-issue labels [MEDIUM — extends A5]
+- newcomer-audit confirmed via gh: no PULL_REQUEST_TEMPLATE.md, no CODE_OF_CONDUCT.md, no FUNDING.yml, no labeler/good-first-issue config. CONTRIBUTING.md is 23 lines, no citizen/branch/.trinity architecture onboarding. All 104 issues ever = AIOSAI-authored; sole external human = YugantM (confirms A3/A4). Feeds DPLAN-0240 Tier 0/1.
+
+### F110 — commons gift/trade/mint brokenness is ALREADY KNOWN + documented [corroborates F52]
+- newcomer-audit: commons/README:77-82 marks gift/trade/mint/collab "not operational — registry path bug", lines 110-117 mark 3 dry-run paths "partial — routing error." So F52 (the owner-casing bug I found in code) is a KNOWN issue the branch documents publicly — good (honesty) and bad (root README pitches commons as live with no caveat). The casing root cause (F52) is likely THE "registry path bug" they mean. Fixing F52 could light up the whole artifact economy (A6).
+
+### F111 — drone cross-project citizen-introspection registry-mismatch [MEDIUM — extends F41]
+- log-sweep: `drone.log` — `Introspection failed for @writer: Registry mismatch: citizen belongs to registry 'b91eefcf...' but found registry '8fb38c96...' at .../Vera-Studio/VERA-STUDIO_REGISTRY.json` (07-10 23:11). Plus @vera auth-denied + 6 bounced emails stuck unread in @drone inbox (F41). Suggests a path-resolution bug in drone's citizen-introspection that walks INTO a sibling project's registry. The 6 @vera bounces (F41) are the same cross-project-comms-is-feics theme (my key_learning 216).
+
+### F112 — telegram_response stuck-pending: one session wedged 13+ hours [strengthens todo 67]
+- log-sweep: `hooks/telegram_response.log` — session `c10bd220` stuck at `start_line=5880`, retrying the same JSONL line 2026-07-11 18:16 → 07-12 07:28 (118 WARNING lines, ~13h) and never resolving. A concrete live instance of todo 67's stuck-pending (F223 key_learning: stale pending retries every Stop forever). The reap/expiry that todo 67 proposes would kill exactly this.
+
+### F117 — SYSTEMIC: the exact code paths that crash/corrupt/leak are the ones with NO test [MEDIUM — meta-pattern]
+- Recurring across every sweep, the highest-severity findings share a tell: **"no test covers this."** F70 (test_token dead field — "no test covers this handler, presumably how it shipped broken"), F94 (CLI markup crash — "zero test coverage"), F104 (cadence <2s drift — "no test <2s apart"), F60/F103 (gate bypasses — "no test exercises malformed stdin with a non-empty matcher"), F99 (seedgo drop-on-exception paths untested), F33/F1 (racy/atomicity paths). The fleet has 371 test files / 10,458 functions and high nominal coverage — but it's concentrated on happy paths; the ERROR/CONCURRENCY/MALFORMED-INPUT branches (exactly where these bugs live) are systematically untested.
+- **Why it matters:** seedgo's Test_Quality standard scores 100% (F47) while the crash-on-bracket, silent-drop, and fail-open branches ship untested — the standard measures test PRESENCE, not adversarial coverage. A "test the unhappy path" checker or a mutation-testing pass would have caught most of these 116. The QA-layer expression of the "write-side shipped, read-side missing" culture (F81): tests assert what SHOULD happen, rarely what happens when it doesn't.
+
+---
+
+### F118 — RAN the tests (not just read them): F1 reproduces LOCALLY, and green suites harbor live bugs [strengthens F1 + F117 empirically]
+- **F1 is flakier than I thought — reproduced LIVE locally:** running the FULL prax suite → `test_mtime_cache_avoids_reread FAILED: assert {} == {'paused': False}` (1 failed, 990 passed). Earlier I ran `TestReadControl` in ISOLATION and it passed (6/6) → I wrongly concluded "only red on CI runners." Full-suite timing (other tests perturbing mtime granularity) triggers it locally too. So F1 isn't a CI-runner quirk — it's genuinely flaky anywhere under realistic timing. Even stronger case for the deterministic `os.utime` fix.
+- **F117 proven empirically:** ran memory (990 passed) and hooks (961 passed) suites — both FULLY GREEN while each harbors a live bug their tests never catch (memory: the keep-14 rollover F74; hooks: cadence <2s drift F104). Green suites + live bugs = exactly F117's thesis: coverage asserts the happy path, not the invariant. 990 memory tests, zero assert "rollover leaves exactly N."
+- **Codebase is structurally sound:** 11,170 tests collect with ZERO import/collection errors — no broken branches, no dead imports. The bones are solid; the gaps are adversarial-coverage + silent-degradation, not rot.
+- **Ran 6 branch suites (~4,000 tests):** memory 990, hooks 961, ai_mail 765, commons 449, spawn 346, flow 730 — ALL green in isolation (backup didn't finish in the time box). The ONLY failure anywhere is the flaky F1. Critical takeaway: NONE of the concurrency findings (F53 conn-leak, F54/F67/F85/F86 races) surface as test failures — they're all LATENT, green suites over untested race paths. Empirically nails F117: the suite is robust on the happy path and blind to exactly the branches these bugs live on.
+
+### F119 — `@pytest.mark.integration` unregistered → silent no-op mark [LOW]
+- `devpulse/tests/test_watchdog_agent.py:469 @pytest.mark.integration` — mark not registered (PytestUnknownMarkWarning). Any `-m integration` selection silently matches nothing; the mark is decorative. Register it in pytest config or it's a filter that does nothing.
+
+### F120 — Always-on fleet: 7 processes, ~22% CPU + 1.8GB RAM steady-state, forever [MEDIUM — resource/ops]
+- Live `ps`: trigger-log-watcher + 5 telegram bots + prax monitor = **7 persistent processes, 21.9% total CPU, 11.4% RAM (~1.77GB)** continuously on a 4-core/15.5GB machine — roughly a full core + 1.8GB permanently, before any actual work. No zombies (clean). Bots still logging poll errors TODAY (bot_base 246 / bot_devpulse 236 lines today; most recent ERROR 20:07 "read operation timed out"). Compounds F10 (no poll backoff): 5 separate bot processes each polling+erroring independently. Consideration: a shared poller (DPLAN-0219 mother-bot) would cut this materially. For a personal machine this steady-state load is worth a conscious decision, not an accident.
+
+### F121 — Log volume dominated by two specific issues; medic-off cost quantified [LOW-MEDIUM]
+- Fleet logs = 73M total. Two files dominate: **`trigger/logs/medic_suppressed.log` = 9.9M (rotated once, +362K active)** — that's 10M+ of pure "Medic OFF - suppressed dispatch" lines, ONE PER dropped error over 63 days = a direct, quantified second cost of F5b (not just errors undelivered, but 10M of suppression noise written). And **`backup/logs/operations.jsonl.1` = 34M** — backup logs every per-file op (464-file versioned runs), rotates but runs heavy. Both rotate (not unbounded) but both trace to a specific fixable cause: re-enable medic (F5b) kills the first; backup could log op-summaries not per-file the second.
+
+### F122 — Bug-magnet map: churn points EXACTLY at the surviving bugs [synthesis — where to harden]
+- `git log --grep=fix` over 6 months, most-touched source files: **memory-rollover is 4 of the top 6** — memory_watcher.py (19 fix commits), detector.py (17), rollover.py (16), extractor.py (16) = ~68 fixes to that one subsystem, AND F74's keep-14 off-by-one lives in the two most-fixed files (detector+extractor) and survived every one of those fixes. **ai_mail dispatch/delivery** is the other magnet: email.py (17), delivery.py (16), wake.py (15), daemon.py (15) = ~63 fixes, where F68/F69/F70/F72 + the F111 cross-project bleed live. **drone.py** (30 fixes, the router — F27 broken help lives here). **doctor.py** (15 fixes — F14-F17 live here).
+- **The signal:** the files fixed most often are the ones still harboring the bugs I found. Repeated patching hasn't converged — memory-rollover and ai_mail-dispatch are churn sinks that keep breaking. These two subsystems are candidates for a hardening PASS (invariants + adversarial tests + the atomic-write/lock discipline) rather than an (N+1)th patch. Churn + surviving-bug overlap = "stop patching, start hardening" list.
+
+### F123 — Memory vector store: HEALTHY, but dashboard undercounts it ~4x [LOW-MEDIUM + an ASSET]
+- **Probed the actual ChromaDB backend** (the persistence layer of the whole "memory persists" value prop): `memory/.chroma/chroma.sqlite3` = 53M, **`PRAGMA integrity_check` = ok** (not corrupt), **5,012 embeddings across 29 collections**, and `drone @memory search` returns results in ~29s (works). The core memory backend is genuinely SOUND — good news for the value prop.
+- **BUT:** memory's DASHBOARD reports `vectors_stored: 1274` while the store actually holds **5,012 embeddings** — a ~4x undercount (likely counting one collection or a stale figure, mislabeled as the total). Another "displayed metric ≠ reality" instance (F17/F48/F49/F74 family) — the dashboard is a decorative number, not a measured one.
+- **Minor doc gap:** navmap says "two ChromaDB stores: local + a global one across all branches" — I find only per-branch `.chroma` dirs (+ backup copies); memory/.chroma (29 collections) appears to BE the aggregate. No separate "global" store exists as described; the doc and the reality have drifted.
+
+### F124 — Dashboards are stale, not "live" — prax's own is 10 days old [MEDIUM]
+- `DASHBOARD.local.json` is documented as "Live state (refreshed by prax)" and is my startup protocol's "single status glance." Actual last_updated across branches: flow 07-13 (fresh), aipass 07-12, drone 07-11, memory 07-11, trigger 07-10, **prax 07-03 (10 DAYS stale)**. Only branches touched by live work update; the rest drift. The "glance" shows days-old data. Ironic: prax — the component that's supposed to refresh dashboards — has the STALEST one. Root: refresh isn't a scheduled ritual (F81 — the daemon runs zero jobs), so it only happens when something manually triggers it. Either schedule a periodic refresh or stop calling them "live."
+- **ASSET:** compass DB (devpulse-owned SQLite, my decision store) — integrity OK, 102 decisions, healthy. Memory vector store healthy (F123). The core data stores are SOUND; it's the freshness/accuracy of the DISPLAY layer that drifts.
+
+### F125 — Verification spot-audit: sub-agent findings hold up (4/4 sampled confirm) [quality/credibility]
+- To gauge false-positive rate in the ~half of findings that came from sub-agents where I only spot-verified criticals/highs, I re-checked 4 random MEDIUM sub-agent findings against code: **F53** (commons curation/search/identity ops — `finally` count = 0 in all three, connections leak on exception ✓), **F68** (ai_mail `load_inbox` — `with open(...) json.load` and NO lock ✓), **F87** (prax registry save — raw `open("w")+json.dump`, no atomic helper ✓), **F96** (api `route_command:244` — `except Exception` swallow ✓). **4/4 confirmed, zero false positives.** Combined with: I personally verified every CRITICAL and HIGH, corrected 5 of my own claims when evidence disproved them, and 2+ agents independently corroborated the big findings. Confidence in the 124-finding set is high — the sub-agents cited exact file:line and the code matched every check. Treat medium/low findings as reliable leads; only the handful explicitly marked PLAUSIBLE need runtime confirmation.
+
+### F126 — Injected-prompt drift: the always-on navmap makes a false claim to every agent [LOW — high frequency]
+- Audited the tier0/tier1/branch prompts (injected into EVERY agent every few turns) against the code reality I mapped. Mostly ACCURATE — good; and the navmap wisely lists agents live rather than hardcoding a count (so no "17 vs 18" drift there). Two real drifts: (1) **navmap line 98: "Two ChromaDB stores: local + a global one across all branches"** — F123 found NO separate global store exists; memory/.chroma IS the aggregate. Every agent is told to expect a store that isn't there. (2) devpulse/README:51 "DASHBOARD.local.json — Live state (refreshed by prax)" — F124 found dashboards up to 10 days stale. Low severity, but the navmap is the highest-read doc in the system (injected fleet-wide on a cadence), so a false claim there propagates to every session. The observability-is-decorative pattern (F117 unification) reaches even the docs: what the system SAYS about itself drifts from what IS.
+- **Net positive:** the prompts are otherwise consistent with reality — the drift is 2 specific stale claims, not systemic prompt-rot. The prompt layer is in better shape than the dashboard/metrics layer.
+
+## MISSION STATUS — COMPLETE COVERAGE
+
+All 17 branches code-swept, all 18 CLI surfaces walked, security gates + skills contribution surface audited, newcomer path + adoption funnel measured, error/medic/memory/daemon/commons internals traced. 9 read-only sub-agents + direct probing, ZERO system files edited. **116 findings, 2 verified-live criticals, 5 self-corrections** where evidence disproved a first claim (F5b root cause, F15 mechanism, supervised-bots, F99 scope, doctor case-vs-path). Deliverables: this doc (F1–F117 + A1–A7 + ~30 issue drafts) · DPLAN-0240 (adoption) · published dashboard · compass #94–#100 · todo 66 root-caused. Every fix is a proposal awaiting Patrick's go.
+
+---
+
+## COMMANDS WALKED
+
+(coverage log so nothing is double-probed)
+
+- **Bare introspection ×18:** drone, cli, git(--help), seedgo, spawn, ai_mail, api, backup, commons, daemon, devpulse, flow, hooks, memory, prax, skills, trigger, aipass
+- **--help:** drone@drone, git, api, backup, commons, spawn, ai_mail, prax, aipass, drone rm
+- **Data/state commands:** trigger errors(+stats/list/help/detail-via-raw-JSON), trigger medic, daemon queue/update/activity_report, prax status/monitor status(✗)/log_audit(✗)/log-audit, flow list open/templates/status(✗)/registry status, skills list/info github/validate telegram/run branch_health(+summary), backup status(bare✗/@name✗/path✓), memory search/lint(+run)/verify/pool/rollover/templates, ai_mail inbox, api usage(✗)/stats/validate/status/bogus(✗), commons feed/thread/activity/catchup/leaderboard/digest/room list/central(✗), hooks status/presence/cadence/hooksound, seedgo checklist/audit @trigger/audit(fleet)/standards_query/test_map @commons, spawn repair(scan)/update preview, devpulse compass query/feedback(+inbox/view)/watchdog status, drone scan/list/audit(✗), git status/log/lock/run list/run view(+--log-failed✗ = gh quirk), aipass doctor/help probe
+- **External/API:** gh api repo stats, labels, issues, PRs, discussions (graphql), traffic, releases, job logs
+- **System:** systemd units/timers, ps bot fleet, tmux ls, pytest TestReadControl (venv), subprocess gh repro
+- (✗) = found broken/misleading — see findings
@@ -0,0 +1,353 @@
+<title>S304 Discovery Scan — AIPass</title>
+<meta name="viewport" content="width=device-width, initial-scale=1" />
+<style>
+  :root {
+    /* neutrals — blue-biased ink, warm paper */
+    --paper: #f4f1ea;
+    --surface: #fbf9f4;
+    --surface-2: #eeeae0;
+    --ink: #1a1e27;
+    --ink-soft: #454b58;
+    --ink-faint: #6b7280;
+    --hair: #ddd6c8;
+    /* accent — passport gold, used sparingly */
+    --accent: #b57e1f;
+    --accent-soft: #e9d9b4;
+    /* semantic severities */
+    --crit: #c92a30;
+    --high: #c0741f;
+    --med: #4964a8;
+    --low: #7b8494;
+    --good: #2f8f5b;
+    --crit-bg: #f6dcdc;
+    --high-bg: #f4e6d0;
+    --med-bg: #dde3f2;
+    --low-bg: #e6e8ec;
+    --good-bg: #d8ecdf;
+    --mono: ui-monospace, "SF Mono", "JetBrains Mono", "Cascadia Code", Menlo, Consolas, monospace;
+    --sans: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif;
+    --maxw: 940px;
+  }
+  @media (prefers-color-scheme: dark) {
+    :root {
+      --paper: #10131a;
+      --surface: #161a23;
+      --surface-2: #1e2430;
+      --ink: #e7e9ee;
+      --ink-soft: #aab2c0;
+      --ink-faint: #79828f;
+      --hair: #2a3140;
+      --accent: #d9a441;
+      --accent-soft: #4a3d1c;
+      --crit: #f0565b;
+      --high: #e8973f;
+      --med: #7d97d8;
+      --low: #8a93a3;
+      --good: #4cc082;
+      --crit-bg: #3a1c1e;
+      --high-bg: #382a17;
+      --med-bg: #1f2740;
+      --low-bg: #242a35;
+      --good-bg: #16301f;
+    }
+  }
+  :root[data-theme="light"] {
+    --paper: #f4f1ea; --surface: #fbf9f4; --surface-2: #eeeae0; --ink: #1a1e27;
+    --ink-soft: #454b58; --ink-faint: #6b7280; --hair: #ddd6c8; --accent: #b57e1f;
+    --accent-soft: #e9d9b4; --crit: #c92a30; --high: #c0741f; --med: #4964a8; --low: #7b8494; --good: #2f8f5b;
+    --crit-bg: #f6dcdc; --high-bg: #f4e6d0; --med-bg: #dde3f2; --low-bg: #e6e8ec; --good-bg: #d8ecdf;
+  }
+  :root[data-theme="dark"] {
+    --paper: #10131a; --surface: #161a23; --surface-2: #1e2430; --ink: #e7e9ee;
+    --ink-soft: #aab2c0; --ink-faint: #79828f; --hair: #2a3140; --accent: #d9a441;
+    --accent-soft: #4a3d1c; --crit: #f0565b; --high: #e8973f; --med: #7d97d8; --low: #8a93a3; --good: #4cc082;
+    --crit-bg: #3a1c1e; --high-bg: #382a17; --med-bg: #1f2740; --low-bg: #242a35; --good-bg: #16301f;
+  }
+
+  * { box-sizing: border-box; }
+  body {
+    margin: 0; background: var(--paper); color: var(--ink);
+    font-family: var(--sans); line-height: 1.5;
+    -webkit-font-smoothing: antialiased;
+  }
+  .wrap { max-width: var(--maxw); margin: 0 auto; padding: 0 22px; }
+
+  /* ── header ── */
+  header { padding: 54px 0 30px; border-bottom: 1px solid var(--hair); }
+  .eyebrow {
+    font-family: var(--mono); font-size: 12px; letter-spacing: .14em;
+    text-transform: uppercase; color: var(--accent); margin: 0 0 16px;
+    display: flex; gap: 12px; align-items: center; flex-wrap: wrap;
+  }
+  .eyebrow .dot { width: 7px; height: 7px; border-radius: 50%; background: var(--good); box-shadow: 0 0 0 3px var(--good-bg); }
+  h1 {
+    font-family: var(--mono); font-weight: 600; font-size: clamp(30px, 5.5vw, 50px);
+    line-height: 1.04; letter-spacing: -0.02em; margin: 0 0 14px; text-wrap: balance;
+  }
+  h1 .stars { color: var(--accent); }
+  .lede { font-size: 18px; color: var(--ink-soft); max-width: 62ch; margin: 0; }
+  .meta {
+    font-family: var(--mono); font-size: 12.5px; color: var(--ink-faint);
+    margin-top: 22px; display: flex; gap: 8px 20px; flex-wrap: wrap;
+  }
+  .meta b { color: var(--ink-soft); font-weight: 600; }
+
+  /* ── stat row ── */
+  .stats { display: grid; grid-template-columns: repeat(4, 1fr); gap: 1px; background: var(--hair);
+    border: 1px solid var(--hair); border-radius: 10px; overflow: hidden; margin: 30px 0 8px; }
+  .stat { background: var(--surface); padding: 18px 18px 16px; }
+  .stat .n { font-family: var(--mono); font-size: 30px; font-weight: 600; letter-spacing: -0.02em;
+    font-variant-numeric: tabular-nums; line-height: 1; }
+  .stat .k { font-size: 12.5px; color: var(--ink-faint); margin-top: 7px; }
+  .stat.crit .n { color: var(--crit); }
+  @media (max-width: 620px) { .stats { grid-template-columns: repeat(2, 1fr); } }
+
+  /* severity meter */
+  .meter { display: flex; height: 8px; border-radius: 6px; overflow: hidden; margin: 20px 0 4px; }
+  .meter span { display: block; }
+  .meter-key { display: flex; gap: 16px 20px; flex-wrap: wrap; font-family: var(--mono);
+    font-size: 11.5px; color: var(--ink-faint); margin-bottom: 6px; }
+  .meter-key i { font-style: normal; display: inline-flex; align-items: center; gap: 6px; }
+  .swatch { width: 9px; height: 9px; border-radius: 2px; display: inline-block; }
+
+  /* ── sections ── */
+  section { padding: 40px 0; border-bottom: 1px solid var(--hair); }
+  .sec-head { display: flex; align-items: baseline; gap: 12px; margin: 0 0 22px; }
+  .sec-head h2 { font-family: var(--mono); font-size: 14px; letter-spacing: .1em; text-transform: uppercase;
+    color: var(--ink); margin: 0; font-weight: 600; }
+  .sec-head .rule { flex: 1; height: 1px; background: var(--hair); }
+  .sec-head .count { font-family: var(--mono); font-size: 12px; color: var(--ink-faint); }
+
+  /* finding cards */
+  .findings { display: flex; flex-direction: column; gap: 12px; }
+  .f { background: var(--surface); border: 1px solid var(--hair); border-radius: 9px;
+    border-left-width: 4px; padding: 16px 18px; }
+  .f.crit { border-left-color: var(--crit); }
+  .f.high { border-left-color: var(--high); }
+  .f.med  { border-left-color: var(--med); }
+  .f.good { border-left-color: var(--good); }
+  .f-top { display: flex; align-items: center; gap: 10px; flex-wrap: wrap; margin-bottom: 7px; }
+  .id { font-family: var(--mono); font-size: 12px; font-weight: 600; color: var(--ink-faint); }
+  .pill { font-family: var(--mono); font-size: 10.5px; letter-spacing: .08em; text-transform: uppercase;
+    font-weight: 600; padding: 2px 8px; border-radius: 20px; }
+  .pill.crit { color: var(--crit); background: var(--crit-bg); }
+  .pill.high { color: var(--high); background: var(--high-bg); }
+  .pill.med  { color: var(--med); background: var(--med-bg); }
+  .pill.good { color: var(--good); background: var(--good-bg); }
+  .f-title { font-weight: 600; font-size: 15.5px; color: var(--ink); flex: 1 1 100%; margin-top: 2px; }
+  .f-body { font-size: 14px; color: var(--ink-soft); margin: 0; }
+  .f-body code { font-family: var(--mono); font-size: 12.5px; background: var(--surface-2);
+    padding: 1px 5px; border-radius: 4px; color: var(--ink); }
+  .tag { font-family: var(--mono); font-size: 11px; color: var(--ink-faint); }
+
+  /* adoption / funnel */
+  .funnel { display: grid; grid-template-columns: repeat(4, 1fr); gap: 10px; margin: 4px 0 24px; }
+  .fu { background: var(--surface); border: 1px solid var(--hair); border-radius: 9px; padding: 16px; text-align: center; }
+  .fu .n { font-family: var(--mono); font-size: 26px; font-weight: 600; letter-spacing: -0.02em; }
+  .fu .k { font-size: 12px; color: var(--ink-faint); margin-top: 5px; }
+  .fu.drop .n { color: var(--crit); }
+  @media (max-width: 620px) { .funnel { grid-template-columns: repeat(2, 1fr); } }
+
+  .callout { background: var(--surface-2); border: 1px solid var(--hair); border-radius: 9px;
+    padding: 18px 20px; margin-top: 18px; }
+  .callout h3 { font-family: var(--mono); font-size: 13px; letter-spacing: .06em; text-transform: uppercase;
+    margin: 0 0 8px; color: var(--accent); }
+  .callout p { margin: 0; font-size: 14px; color: var(--ink-soft); }
+
+  ul.ideas { list-style: none; padding: 0; margin: 8px 0 0; display: flex; flex-direction: column; gap: 10px; }
+  ul.ideas li { display: flex; gap: 12px; font-size: 14px; color: var(--ink-soft); align-items: baseline; }
+  ul.ideas .tier { font-family: var(--mono); font-size: 11px; font-weight: 600; color: var(--accent);
+    background: var(--accent-soft); padding: 2px 7px; border-radius: 5px; white-space: nowrap; }
+
+  .deliverables { display: flex; flex-direction: column; gap: 10px; }
+  .d { display: flex; gap: 12px; align-items: baseline; font-size: 14px; color: var(--ink-soft);
+    font-family: var(--mono); }
+  .d .path { color: var(--ink); font-weight: 600; }
+  .d .chk { color: var(--good); }
+
+  footer { padding: 30px 0 60px; font-family: var(--mono); font-size: 12px; color: var(--ink-faint); }
+  footer .disc { color: var(--good); }
+</style>
+
+<div class="wrap">
+  <header>
+    <p class="eyebrow"><span class="dot"></span> Session S304 · Autonomous Discovery Scan · Search-only</p>
+    <h1>124 findings across a <span class="stars">237-star</span> system that no one has ever contributed to.</h1>
+    <p class="lede">A full unsupervised walk of AIPass — all 18 agents' code, the logs, error registries, security gates, live processes, ~4,000 tests run, and the newcomer path. Zero system files changed. Every fix here is a proposal, evidence attached.</p>
+    <p class="meta"><span><b>Date</b> 2026-07-12</span><span><b>Agent</b> @devpulse</span><span><b>Surfaces walked</b> 18/18</span><span><b>Sub-agents</b> 5 read-only</span><span><b>Files edited</b> 0 system</span></p>
+  </header>
+
+  <div class="stats" role="list">
+    <div class="stat" role="listitem"><div class="n">124</div><div class="k">bug &amp; gap findings</div></div>
+    <div class="stat crit" role="listitem"><div class="n">2</div><div class="k">live criticals</div></div>
+    <div class="stat" role="listitem"><div class="n">29</div><div class="k">issue drafts, ready to file</div></div>
+    <div class="stat" role="listitem"><div class="n">9</div><div class="k">read-only sub-agents</div></div>
+  </div>
+
+  <div class="meter-key" aria-hidden="true">
+    <i><span class="swatch" style="background:var(--crit)"></span>Critical 2</i>
+    <i><span class="swatch" style="background:var(--high)"></span>High 28</i>
+    <i><span class="swatch" style="background:var(--med)"></span>Medium 38</i>
+    <i><span class="swatch" style="background:var(--low)"></span>Low 32</i>
+    <i><span class="swatch" style="background:var(--good)"></span>Verified-good 6</i>
+  </div>
+  <div class="meter" aria-hidden="true">
+    <span style="flex:2;background:var(--crit)"></span>
+    <span style="flex:28;background:var(--high)"></span>
+    <span style="flex:38;background:var(--med)"></span>
+    <span style="flex:32;background:var(--low)"></span>
+    <span style="flex:6;background:var(--good)"></span>
+  </div>
+
+  <!-- CRITICAL + HIGH FINDINGS -->
+  <section>
+    <div class="sec-head"><h2>Fix First</h2><span class="rule"></span><span class="count">the load-bearing ones</span></div>
+    <div class="findings">
+
+      <div class="f crit">
+        <div class="f-top"><span class="id">F74</span><span class="pill crit">Critical · live now · verified</span>
+          <span class="f-title">Memory rollover is keeping 14 entries, not 15 — on every branch, right now</span></div>
+        <p class="f-body">An off-by-one trims memory at the keep target instead of above it: the trigger fires at <code>len &gt;= 15</code> and the extractor forces <code>max(excess, 1)</code>, so at exactly 15 it removes one it should keep. The proof is on disk — memory's own <code>local.json</code> holds exactly 14 sessions and 14 learnings, and the rollover log shows <code>(15/15 sessions)</code> firing 53 times. Every branch silently loses one extra memory per rollover. Standing hold on rollover fixes respected — documented, not filed.</p>
+      </div>
+
+      <div class="f crit">
+        <div class="f-top"><span class="id">F5b</span><span class="pill crit">Critical · root cause corrected</span>
+          <span class="f-title">Medic error-notification has been switched off for 63 days</span></div>
+        <p class="f-body">The medic log tells the whole story in three lines: at <b>2026-05-10 20:30:21</b> — the same minute a 25-second burst of test-fixture errors flooded in — medic was switched off ("Medic DISABLED — error dispatch suppressed") and never switched back on. Since then <code>config.medic_enabled = false</code> makes the dispatch path drop every detected error before anything else runs; 63 days later it's still suppressing live errors (1,667 "Medic OFF" log lines, zero "Circuit breaker OPEN"). A brief annoyance muted the entire error-notification system for two months. <b>Honest note:</b> my first pass blamed a stuck circuit breaker; a later sweep refuted it and I re-verified on disk. That breaker is a real but separate latent bug (F5c) that must be fixed <em>before</em> medic is switched back on, or it re-floods.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F1</span><span class="pill high">High</span>
+          <span class="f-title">PR#696 red CI is one inherently racy test</span></div>
+        <p class="f-body">Every red job fails on the same assertion: <code>test_mtime_cache_avoids_reread</code> passes only when two file writes share an mtime tick — a filesystem coin flip that lands green locally and red on GitHub runners. Two-line deterministic fix (<code>os.utime</code> to pin mtime equality). @prax owns.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F14–F17</span><span class="pill high">High</span>
+          <span class="f-title">aipass doctor cries wolf on a healthy install</span></div>
+        <p class="f-body">Our first-touch trust tool reports <b>7 errors on a clean repo — all false</b>: it scans <code>.backup/snapshots/</code> as if they were live agents, compares registry names case-sensitively against 5 legitimately-uppercase entries, prints an unsubstituted <code>{{BRANCHNAME}}</code>, and shows ✓ marks whose text is a warning. A newcomer reads "my install is broken."</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F30</span><span class="pill high">High</span>
+          <span class="f-title">spawn repair advises a command that would delete the live @aipass agent</span></div>
+        <p class="f-body"><code>spawn repair</code> flags <code>src/aipass/aipass/</code> as duplicate "pollution" and prints the remediation <code>--clean-pollution</code> (archive + remove). That directory is the living concierge agent. Following the tool's own advice archives a user-facing citizen. Needs a passport-awareness guard.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F52</span><span class="pill high">High · verified in code</span>
+          <span class="f-title">Commons artifact gifting is broken by name-casing</span></div>
+        <p class="f-body">Gift / trade / mint write an <code>UPPERCASE</code> owner (<code>trade_ops.py:58</code>) while every ownership check compares lowercase. Gift an artifact to <code>@seed</code> and it becomes invisible in their inventory and permanently un-tradeable (<code>"SEED" != "seed"</code>). The self-gift guard never trips. The registry-casing disease, database edition.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F6 · F43</span><span class="pill high">High</span>
+          <span class="f-title">Tests write fixtures into production logs, poisoning the error registry</span></div>
+        <p class="f-body">pytest fixtures ("disk full", "boom", <code>/tmp/pytest-of-patrick/…</code>) land in real branch <code>logs/</code>; the 24/7 log-watcher ingests them into the error registry (664+ occurrences of one fixture line alone). One cut fixes the class: route logging to tmp when <code>PYTEST_CURRENT_TEST</code> is set.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F22 · F23</span><span class="pill high">High</span>
+          <span class="f-title">The human-facing concierge has the worst help in the fleet</span></div>
+        <p class="f-body"><code>aipass --help</code> is a bare module dump — no "what is this," no first steps, internal modules exposed. <code>aipass help "how do I create a new branch"</code> answers with git-PR commands and never mentions @spawn. The one tool built for humans is the least helpful.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F46</span><span class="pill high">High</span>
+          <span class="f-title">14 feedback messages unread since April — including 4 external bug reports</span></div>
+        <p class="f-body">Precise bug reports from an external project sat <code>NEW</code> for three months; we later rediscovered every one of them the hard way. The owner-to-owner channel works technically and fails operationally — no inbox check in startup, no aging alarm.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F53 · F54</span><span class="pill high">High</span>
+          <span class="f-title">Commons: systemic DB connection leak + unlocked log that self-wipes</span></div>
+        <p class="f-body">~14 handler files close their sqlite connection only on the success path (no <code>finally:</code>), 40+ sites — hottest runs on every post. The shared op-log does unlocked read-modify-write; a torn write is "healed" by silently resetting the log to <code>[]</code>. The correct pattern already exists in sibling files.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F59</span><span class="pill high">High · verified · integrity</span>
+          <span class="f-title">The owner model is advisory: git-write trusts an unprotected passport name</span></div>
+        <p class="f-body"><code>verify_git_access</code> grants owner-tier git if the <code>branch_name</code> in the caller's own <code>passport.json</code> reads "devpulse" — no registry cross-check, no signature, and no gate protects that file. The DPLAN-0231 owner-capability model was built to key on the sealed <code>registry_id</code>; the actual check reads the mutable name instead. <b>Scope honestly:</b> agents are same-user and cooperative, so this is "the model isn't enforced," not a remote exploit — but it undoes a guarantee you deliberately built.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F84 · F85</span><span class="pill high">High · verified · destructive</span>
+          <span class="f-title">Two commands can destroy a branch or another branch's plan history</span></div>
+        <p class="f-body"><code>spawn delete_branch</code> guards only three hardcoded names — no check for a live process or <code>owner:true</code> — so it will archive-and-remove any other running branch on request. And the flow central aggregator, when it reads a branch's registry during a transient unreadable moment, "heals" it by writing an empty registry back — one branch's routine aggregation can wipe another's entire plan history. Both traced in code.</p>
+      </div>
+
+      <div class="f high">
+        <div class="f-top"><span class="id">F114</span><span class="pill high">High · verified · adoption-blocking</span>
+          <span class="f-title">The contribution surface everyone would reach for first has no safety boundary</span></div>
+        <p class="f-body">Running a skill executes its <code>handler.py</code> in-process with full privileges — no sandbox, no confirmation, no review (verified: the only relevant line is a bare <code>exec_module</code>). A built-in skill is also <code>shell=True</code>-injectable, and a skill dropped into <code>~/.aipass/skills/</code> can silently impersonate a trusted built-in by name. Skills are the natural first thing an outside contributor would build — so this is the one place the adoption plan and the security posture collide. A community-skills ecosystem needs a trust model (sandbox, review, or a first-run consent gate) before it can open.</p>
+      </div>
+
+    </div>
+  </section>
+
+  <!-- PATTERN -->
+  <section>
+    <div class="sec-head"><h2>The Pattern Underneath</h2><span class="rule"></span></div>
+    <div class="callout">
+      <h3>The deepest root: the gauges are decorative, so nothing looks broken</h3>
+      <p>Running the live system surfaced the mechanism behind everything else. The <em>data</em> is sound — the vector store passes an integrity check with 5,012 embeddings, the decision DB is clean, 11,000 tests collect without error, six suites run green. But the system's <em>self-report</em> is wrong almost everywhere it's checked: the dashboard claims 1,274 vectors when there are 5,012; dashboards labeled "live" are up to ten days stale; doctor invents seven errors on a healthy repo; the standards audit reports "100%" while silently ignoring any file it can't parse. When every gauge reads "fine" regardless of reality, a thing can break quietly and stay broken for months — which is exactly what medic, rollover, and the stuck breaker did. The highest-leverage fix isn't any single bug; it's making the instruments <em>measured</em> — real counts, real freshness, invariants that fail loud — so the system can finally see its own state.</p>
+    </div>
+    <div class="callout" style="margin-top:14px">
+      <h3>We build the write side and never the read side</h3>
+      <p>The same shape recurs across F5 (errors detected, never triaged), F26 (presence records written, never cleaned), F40/F46 (mail delivered, never read), F49 (a health check pointed at a retired schema), F5b (a breaker that opens but can't close), F81 (a scheduler running with zero jobs). Detection, ingestion, and enforcement ship; the closing ritual — the thing that reads, ages, resolves, or recovers — is the half that's consistently missing. It's a culture fix as much as a code fix: every new pipeline needs its janitor shipped with it.</p>
+    </div>
+    <div class="callout" style="margin-top:14px">
+      <h3>One fix retires a dozen bugs</h3>
+      <p>Across six branches the single most common defect is the same: shared state — registries, passports, inboxes, scheduler runstate, the logger's own data — written with a raw <code>open("w") + json.dump</code>, no temp-and-rename, no lock. A crash mid-write truncates the file; the paired loader "fails open" to empty and writes that empty back, turning a transient read hiccup into permanent data loss. The correct atomic-write helper <b>already exists in every branch</b> — it's just applied inconsistently. One shared primitive plus a seedgo checker that forbids the raw pattern closes roughly a dozen findings at once, including the two that can wipe another branch's plan registry or truncate a passport.</p>
+    </div>
+    <div class="callout" style="margin-top:14px">
+      <h3>The deepest gap: blindness to silent degradation</h3>
+      <p>I dated the biggest findings by git blame: the rollover bug has been trimming memory since April 22, the seedgo audit has silently dropped crashing checkers since March 23 (its inception), medic has been off since May 10. They survived for <em>months</em> — not because they're hard to find, but because none of them produces a symptom. Rollover archives the entry instead of deleting it, so nothing looks lost. Medic being off just means silence. A dropped checker just makes a number rounder. The whole system is instrumented for <em>visible</em> failure — logs, errors, red CI — and has almost no way to notice when something quietly starts doing the wrong thing. The durable fix isn't per-bug; it's a handful of invariants that assert the invisible — rollover leaves exactly N, medic is on, every registered checker actually ran — and shout when they don't.</p>
+    </div>
+    <div class="callout" style="margin-top:14px">
+      <h3>The gates fail open, quietly</h3>
+      <p>The security gates are well-built where they run — but they default to <em>allow</em> at four independent layers: a missing config file, a handler that crashes, malformed input, and any parser exception, each logged only as a line of text. And the standards audit does the same — a file that makes a checker throw is dropped from the score rather than failed, so "100%" quietly means "100% of the files we could parse." None of this is a remote exploit — these are cooperative same-user agents — but a guarantee that fails silently isn't a guarantee. The cheap half of the fix is to make every fail-open <em>loud</em>: surface it to the error registry, not just a log.</p>
+    </div>
+  </section>
+
+  <!-- ADOPTION -->
+  <section>
+    <div class="sec-head"><h2>The Contributor Funnel</h2><span class="rule"></span><span class="count">DPLAN-0240</span></div>
+    <div class="funnel">
+      <div class="fu"><div class="n">~600</div><div class="k">unique humans / month</div></div>
+      <div class="fu"><div class="n">237</div><div class="k">stars</div></div>
+      <div class="fu"><div class="n">1</div><div class="k">external human, ever</div></div>
+      <div class="fu drop"><div class="n">0</div><div class="k">retained / contributing</div></div>
+    </div>
+    <p class="f-body" style="margin-bottom:18px">Attraction works; conversion is broken. GitHub topics are <b>empty</b>, the homepage field is blank, Discussions are dead, and the Commons — a genuinely rich agent-society sandbox with craftable artifacts, time capsules, and a hidden exploration game — has exactly <b>one post</b>.</p>
+
+    <div class="callout">
+      <h3>The April Precedent</h3>
+      <p>This exact play was already run once. In April an external agent created three <code>good first issue</code> tasks and added the README "Need help?" line. Three weeks later our own issue-zero sweep batch-closed all three — same minute, zero comments — and the underlying work was never done (the seedgo <code>readme_currency</code> proof still fails today). We didn't just lose a contributor's momentum; our tidiness reflex deleted the on-ramp.</p>
+    </div>
+
+    <ul class="ideas">
+      <li><span class="tier">Tier 0</span><span>Set GitHub topics, fill the homepage field, pin a "Start here" discussion, add a code of conduct — ~30 minutes, near-zero risk.</span></li>
+      <li><span class="tier">Tier 1</span><span>An external-response reflex (&lt;24h reply, via our own dispatch plumbing) and a protected shelf of <code>good first issue</code> tasks we deliberately don't self-clear.</span></li>
+      <li><span class="tier">Tier 2</span><span>Fix the first-ten-minutes tools: doctor's false errors, the concierge's help, the two broken examples in <code>drone --help</code>.</span></li>
+      <li><span class="tier">Tier 3</span><span>A docs front-door, and revive the Commons with automatic agent rituals — then expose a read-only public feed. "Watch AI agents run their own society" is a headline no competitor can match.</span></li>
+    </ul>
+  </section>
+
+  <!-- DELIVERABLES -->
+  <section>
+    <div class="sec-head"><h2>Deliverables</h2><span class="rule"></span></div>
+    <div class="deliverables">
+      <div class="d"><span class="chk">✓</span><span><span class="path">docs/discovery/S304_discovery_mission.md</span> — 58 findings + 6 adoption notes + 17 ready-to-file issue drafts, evidence inline</span></div>
+      <div class="d"><span class="chk">✓</span><span><span class="path">DPLAN-0240</span> — contributor funnel, four tiers, the April Precedent</span></div>
+      <div class="d"><span class="chk">✓</span><span><span class="path">todo 66</span> — CI red root-caused to F1; local.json updated</span></div>
+      <div class="d"><span class="chk">✓</span><span><span class="path">compass #94 / #95</span> — circuit-breaker lesson + April-Precedent decision recorded</span></div>
+      <div class="d"><span class="chk">✓</span><span><span class="path">5 read-only sub-agents</span> — newcomer, logs, code sweeps (commons/backup/ai_mail/daemon), security gates</span></div>
+    </div>
+  </section>
+
+  <footer>
+    <p class="disc">// search-only mission — no system files edited, no issues filed, no settings changed, no posts made.</p>
+    <p>Every outward-facing action awaits your go. @devpulse · S304</p>
+  </footer>
+</div>
@@ -275,3 +275,176 @@ def test_flag_without_value_errors(capsys, db):
    assert compass_cmd.handle_command("compass", ["query", "x", "--rating"]) is True
    out = _output(capsys).lower()
    assert "rating" in out and "value" in out
+
+
+# ---------------------------------------------------------------------------
+# supersedes — atomic archive + link, both pointer directions
+# ---------------------------------------------------------------------------
+
+
+def test_add_supersedes_archives_and_links(capsys, db):
+    """add --supersedes N archives #N, links the new row, shows 'supersedes #N'."""
+    old = _add(capsys, db, "old ctx sessions", "use sessions", "good")
+    capsys.readouterr()
+    assert (
+        compass_cmd.handle_command(
+            "compass",
+            [
+                "add",
+                "new ctx jwt",
+                "switch to jwt",
+                "--rating",
+                "good",
+                "--supersedes",
+                str(old),
+                "--db",
+                db,
+            ],
+        )
+        is True
+    )
+    out = _output(capsys)
+    assert f"supersedes #{old}" in out
+
+    # The archived row is gone from the default (active-only) query...
+    q = _query_out(capsys, db, "sessions")
+    assert "0 result(s)" in q
+
+    # ...but --include-archived surfaces it WITH its status + forward pointer.
+    q2 = _query_out(capsys, db, "sessions", "--include-archived")
+    assert "ARCHIVED" in q2.upper()
+    assert "superseded by #" in q2.lower()
+
+
+def test_add_supersedes_bad_id_errors_no_write(capsys, db):
+    """add --supersedes to a missing id errors and writes nothing."""
+    capsys.readouterr()
+    compass_cmd.handle_command(
+        "compass",
+        ["add", "ctx", "dec", "--rating", "good", "--supersedes", "9999", "--db", db],
+    )
+    out = _output(capsys).lower()
+    assert "9999" in out
+    assert "total decisions: 0" in _stats_out(capsys, db).lower()
+
+
+def test_add_supersedes_non_integer_errors(capsys, db):
+    """A non-integer --supersedes fails loud."""
+    assert (
+        compass_cmd.handle_command(
+            "compass",
+            ["add", "ctx", "dec", "--rating", "good", "--supersedes", "abc", "--db", db],
+        )
+        is True
+    )
+    out = _output(capsys).lower()
+    assert "supersedes" in out and "integer" in out
+
+
+# ---------------------------------------------------------------------------
+# write-time conflict advisory — non-blocking
+# ---------------------------------------------------------------------------
+
+
+def test_add_conflict_advisory_prints_but_does_not_block(capsys, db):
+    """An overlapping active row triggers an advisory; the add still succeeds."""
+    _add(capsys, db, "caching layer strategy", "add redis caching", "good")
+    capsys.readouterr()
+    compass_cmd.handle_command(
+        "compass",
+        ["add", "caching approach again", "another caching layer", "--rating", "good", "--db", db],
+    )
+    out = _output(capsys)
+    assert "possible conflict" in out.lower()
+    assert "--supersedes" in out  # advisory hints the fix
+    assert "Added decision" in out  # NON-BLOCKING: still added
+
+
+def test_add_no_conflict_on_empty_store(capsys, db):
+    """First add on an empty store prints no advisory."""
+    capsys.readouterr()
+    compass_cmd.handle_command("compass", ["add", "unique ctx", "unique dec", "--rating", "good", "--db", db])
+    out = _output(capsys).lower()
+    assert "possible conflict" not in out
+
+
+# ---------------------------------------------------------------------------
+# note — set a note, prove it is immediately searchable
+# ---------------------------------------------------------------------------
+
+
+def test_note_command_sets_and_is_searchable(capsys, db):
+    """note <id> "text" sets the note; a later query finds the new note text."""
+    did = _add(capsys, db, "note cmd ctx", "note cmd dec", "good")
+    capsys.readouterr()
+    assert compass_cmd.handle_command("compass", ["note", str(did), "findme pterodactyl", "--db", db]) is True
+    out = _output(capsys).lower()
+    assert "note set" in out
+
+    q = _query_out(capsys, db, "pterodactyl")
+    assert "1 result(s)" in q
+
+
+def test_note_help(capsys):
+    """compass note --help prints per-subcommand usage."""
+    assert compass_cmd.handle_command("compass", ["note", "--help"]) is True
+    out = _output(capsys).lower()
+    assert "note" in out and "usage" in out
+
+
+def test_note_missing_id_warns(capsys, db):
+    """note on a non-existent id reports nothing changed, does not crash."""
+    assert compass_cmd.handle_command("compass", ["note", "999", "text", "--db", db]) is True
+    out = _output(capsys).lower()
+    assert "999" in out and ("nothing changed" in out or "no decision" in out)
+
+
+def test_note_missing_args_shows_usage(capsys, db):
+    """note with too few args shows usage."""
+    assert compass_cmd.handle_command("compass", ["note", "5", "--db", db]) is True
+    out = _output(capsys).lower()
+    assert "usage" in out
+
+
+def test_help_and_introspection_list_note(capsys):
+    """Both --help and bare introspection advertise the note subcommand."""
+    compass_cmd.handle_command("compass", ["--help"])
+    assert "note" in _output(capsys).lower()
+    compass_cmd.handle_command("compass", [])
+    assert "note" in _output(capsys).lower()
+
+
+# ---------------------------------------------------------------------------
+# --include-archived — archived hits must show status + supersession pointer
+# ---------------------------------------------------------------------------
+
+
+def test_include_archived_shows_archived_pointer(capsys, db):
+    """--include-archived surfaces an archived row flagged with its successor."""
+    old = _add(capsys, db, "archived-visible ctx", "the old choice", "bad")
+    capsys.readouterr()
+    compass_cmd.handle_command(
+        "compass",
+        [
+            "add",
+            "replacement ctx",
+            "the new choice",
+            "--rating",
+            "good",
+            "--supersedes",
+            str(old),
+            "--db",
+            db,
+        ],
+    )
+    capsys.readouterr()
+
+    # Default query hides the archived row.
+    q = _query_out(capsys, db, "old choice")
+    assert "0 result(s)" in q
+
+    # With the flag it appears, unmistakably marked archived + superseded.
+    q2 = _query_out(capsys, db, "old choice", "--include-archived")
+    assert "1 result(s)" in q2
+    assert "archived" in q2.lower()
+    assert "superseded by #" in q2.lower()
@@ -283,3 +283,252 @@ class TestInputValidation:
        compass.add_decision("ctx", "dec", "good", db_path=db)
        with pytest.raises(ValueError):
            compass.query_decisions("ctx", limit=0, db_path=db)
+
+
+class TestSupersedes:
+    """supersedes column, atomic archive+link, idempotent migration (DPLAN-0246)."""
+
+    def test_add_with_supersedes_archives_and_links(self, db):
+        """--supersedes links the corrector AND archives the target, atomically."""
+        old = compass.add_decision("old auth ctx", "use sessions", "good", db_path=db)
+        new = compass.add_decision("new auth ctx", "switch to JWT", "good", db_path=db, supersedes=old)
+        # The corrector row links back to what it replaced.
+        hit = compass.query_decisions("JWT", db_path=db)[0]
+        assert hit["supersedes"] == old
+        # The old entry is archived → gone from the active query.
+        assert compass.query_decisions("sessions", db_path=db) == []
+        # With include_archived it reappears, pointing FORWARD to its successor.
+        arch = compass.query_decisions("sessions", include_archived=True, db_path=db)[0]
+        assert arch["status"] == "archived"
+        assert arch["superseded_by"] == new
+
+    def test_supersedes_nonexistent_raises_no_partial_write(self, db):
+        """A bad --supersedes id errors cleanly and leaves NO partial write."""
+        with pytest.raises(ValueError):
+            compass.add_decision("ctx", "dec", "good", db_path=db, supersedes=9999)
+        assert compass.stats(db_path=db)["total"] == 0
+
+    def test_supersedes_defaults_none(self, db):
+        """A plain add has supersedes=None and superseded_by=None."""
+        compass.add_decision("plain ctx", "plain dec", "good", db_path=db)
+        hit = compass.query_decisions("plain", db_path=db)[0]
+        assert hit["supersedes"] is None
+        assert hit["superseded_by"] is None
+
+    def test_active_hit_shows_its_supersedes_pointer(self, db):
+        """An ACTIVE corrector still exposes its supersedes pointer on query."""
+        old = compass.add_decision("legacy topic zzz", "old way", "bad", db_path=db)
+        compass.add_decision("current topic zzz", "new way", "good", db_path=db, supersedes=old)
+        hit = compass.query_decisions("current", db_path=db)[0]
+        assert hit["supersedes"] == old
+        assert hit["superseded_by"] is None  # nothing supersedes the corrector
+
+    def test_migration_idempotent_repeated_connects(self, db):
+        """Re-opening the DB re-runs migration harmlessly; column stays present."""
+        compass.add_decision("ctx one", "dec one", "good", db_path=db)
+        for _ in range(3):
+            conn = store._connect(db)
+            try:
+                cols = {r["name"] for r in conn.execute("PRAGMA table_info(decisions)")}
+            finally:
+                conn.close()
+            assert "supersedes" in cols
+
+    def test_migration_adds_column_to_legacy_db(self, db):
+        """A pre-supersedes DB gets the column added in via _connect migration."""
+        # Build a legacy `decisions` table WITHOUT supersedes, as older builds had.
+        conn = sqlite3.connect(str(db))
+        conn.execute(
+            "CREATE TABLE decisions ("
+            "id INTEGER PRIMARY KEY, created TEXT, context TEXT NOT NULL, "
+            "decision TEXT NOT NULL, rating TEXT NOT NULL, note TEXT, tags TEXT, "
+            "source TEXT, score INTEGER, status TEXT DEFAULT 'active', "
+            "last_reviewed TEXT, times_surfaced INTEGER DEFAULT 0)"
+        )
+        conn.commit()
+        conn.close()
+        # Legacy table lacks the column...
+        conn = sqlite3.connect(str(db))
+        pre = {r[1] for r in conn.execute("PRAGMA table_info(decisions)")}
+        conn.close()
+        assert "supersedes" not in pre
+        # ...a store connect migrates it in (CREATE IF NOT EXISTS is a no-op here).
+        conn = store._connect(db)
+        try:
+            post = {r["name"] for r in conn.execute("PRAGMA table_info(decisions)")}
+        finally:
+            conn.close()
+        assert "supersedes" in post
+
+
+class TestFindConflicts:
+    """Write-time conflict check: FTS over ACTIVE rows, sanitized, side-effect-free."""
+
+    def test_finds_overlapping_active_row(self, db):
+        """A would-be entry surfaces an existing active row it overlaps."""
+        compass.add_decision("caching strategy for the API", "add a redis caching layer", "good", db_path=db)
+        hits = compass.find_conflicts("caching approach", "use a caching layer", db_path=db)
+        assert len(hits) >= 1
+        assert any("caching" in h["context"] for h in hits)
+
+    def test_ignores_archived_rows(self, db):
+        """Conflict check searches active rows only — archived never surfaces."""
+        did = compass.add_decision("archived topic xyzzy", "some decision", "good", db_path=db)
+        compass.archive(did, db_path=db)
+        assert compass.find_conflicts("xyzzy topic", "another decision", db_path=db) == []
+
+    def test_no_side_effects_on_times_surfaced(self, db):
+        """The advisory must NOT bump times_surfaced — it is not a real surface."""
+        compass.add_decision("surfacing guard ctx", "a decision here", "good", db_path=db)
+        compass.find_conflicts("surfacing guard", "a decision", db_path=db)
+        hit = compass.query_decisions("surfacing", db_path=db)[0]
+        assert hit["times_surfaced"] == 1  # only the query above counted
+
+    def test_sanitizes_fts_special_chars(self, db):
+        """Raw FTS5 syntax characters must not crash MATCH — sanitized to literals."""
+        compass.add_decision("special ctx", "a normal decision", "good", db_path=db)
+        weird = 'broken " ( ) * : query -term AND OR NOT'
+        result = compass.find_conflicts(weird, "more * (text) ^caret", db_path=db)
+        assert isinstance(result, list)  # no exception raised
+
+    def test_empty_text_returns_empty(self, db):
+        """Text with no usable tokens yields no conflicts (and no crash)."""
+        assert compass.find_conflicts("   ", "   ", db_path=db) == []
+
+
+class TestSetNote:
+    """note edits persist AND re-index immediately via the FTS5 UPDATE trigger."""
+
+    def test_set_note_updates_and_returns_true(self, db):
+        """set_note stores the note and reports the row was changed."""
+        did = compass.add_decision("note ctx", "note dec", "good", db_path=db)
+        assert compass.set_note(did, "a fresh observation", db_path=db) is True
+        hit = compass.query_decisions("note ctx", db_path=db)[0]
+        assert hit["note"] == "a fresh observation"
+
+    def test_note_edit_is_immediately_fts_searchable(self, db):
+        """PROOF: the decisions_au trigger re-indexes a note UPDATE for FTS."""
+        did = compass.add_decision("indexing ctx", "indexing dec", "good", db_path=db)
+        # 'zebra' appears nowhere yet.
+        assert compass.query_decisions("zebra", db_path=db) == []
+        compass.set_note(did, "mentions zebra now", db_path=db)
+        # Immediately findable through the freshly re-indexed note column.
+        found = compass.query_decisions("zebra", db_path=db)
+        assert len(found) == 1
+        assert found[0]["id"] == did
+
+    def test_set_note_missing_id_returns_false(self, db):
+        """set_note on a non-existent id returns False (no silent create)."""
+        assert compass.set_note(9999, "nope", db_path=db) is False
+
+
+class TestScoreRemoved:
+    """score is gone from every Python surface (DPLAN-0246 seedgo ruling)."""
+
+    def test_score_not_in_decision_columns(self):
+        """The code-level column list no longer names score."""
+        assert "score" not in store._DECISION_COLUMNS
+
+    def test_score_absent_from_query_dict(self, db):
+        """Query result dicts carry no score key."""
+        compass.add_decision("score ctx", "score dec", "good", db_path=db)
+        hit = compass.query_decisions("score", db_path=db)[0]
+        assert "score" not in hit
+
+    def test_score_absent_from_review_dict(self, db):
+        """Review result dicts carry no score key."""
+        compass.add_decision("review score ctx", "dec", "good", db_path=db)
+        result = compass.review(db_path=db)
+        assert result is not None
+        assert "score" not in result
+
+
+class TestRecall:
+    """recall_decisions/mark_surfaced — the Track 2 ambient-recall read path."""
+
+    def test_recall_returns_scored_active_candidates(self, db):
+        """Raw prompt text yields active hits, each with relevance in (0, 1)."""
+        compass.add_decision("vectorization pipeline ctx", "salt vector ids", "good", db_path=db)
+        hits = compass.recall_decisions("we are working on the vectorization pipeline", db_path=db)
+        assert hits and hits[0]["decision"] == "salt vector ids"
+        assert 0.0 < hits[0]["relevance"] < 1.0
+
+    def test_recall_is_side_effect_free(self, db):
+        """A recall does NOT bump times_surfaced — candidates are not surfacings."""
+        compass.add_decision("recall counter ctx", "stay untouched", "good", db_path=db)
+        rid = compass.recall_decisions("recall counter", db_path=db)[0]["id"]
+        compass.recall_decisions("recall counter", db_path=db)
+        hit = compass.query_decisions("untouched", db_path=db)[0]
+        assert hit["id"] == rid
+        # query_decisions itself increments once; recalls added nothing.
+        assert hit["times_surfaced"] == 1
+
+    def test_recall_excludes_archived(self, db):
+        """Archived rows never come back as ambient candidates."""
+        rid = compass.add_decision("archived recall ctx", "dead ruling", "bad", db_path=db)
+        compass.archive(rid, db_path=db)
+        assert compass.recall_decisions("archived recall dead ruling", db_path=db) == []
+
+    def test_recall_survives_fts_syntax_in_prompt(self, db):
+        """FTS5 syntax characters in a prompt cannot crash the MATCH."""
+        compass.add_decision("syntax safety ctx", "quote all tokens", "good", db_path=db)
+        hits = compass.recall_decisions('safety AND (tokens) OR "quote" NEAR *:^-', db_path=db)
+        assert hits and hits[0]["decision"] == "quote all tokens"
+
+    def test_recall_empty_prompt_returns_empty(self, db):
+        """No usable tokens → empty list, no error."""
+        assert compass.recall_decisions("", db_path=db) == []
+        assert compass.recall_decisions("()!@#$", db_path=db) == []
+
+    def test_recall_ranks_most_relevant_first(self, db):
+        """Denser overlap outranks a single shared token."""
+        compass.add_decision("alpha beta gamma delta", "dense match", "good", db_path=db)
+        compass.add_decision("alpha unrelated topic here", "sparse match", "good", db_path=db)
+        hits = compass.recall_decisions("alpha beta gamma delta", limit=2, db_path=db)
+        assert hits[0]["decision"] == "dense match"
+        assert hits[0]["relevance"] > hits[1]["relevance"]
+
+    def test_recall_invalid_limit_raises(self, db):
+        """Non-positive limit fails honestly."""
+        with pytest.raises(ValueError):
+            compass.recall_decisions("anything", limit=0, db_path=db)
+
+    def test_mark_surfaced_counts_only_injected(self, db):
+        """mark_surfaced increments exactly the ids the caller reports."""
+        a = compass.add_decision("mark ctx one", "dec one xyzzy", "good", db_path=db)
+        b = compass.add_decision("mark ctx two", "dec two xyzzy", "good", db_path=db)
+        assert compass.mark_surfaced([a], db_path=db) == 1
+        hits = {h["id"]: h for h in compass.query_decisions("xyzzy", limit=5, db_path=db)}
+        # query bumps both by 1; only a carries the extra mark_surfaced bump.
+        assert hits[a]["times_surfaced"] == 2
+        assert hits[b]["times_surfaced"] == 1
+
+    def test_mark_surfaced_empty_list_is_noop(self, db):
+        """An empty id list returns 0 and touches nothing."""
+        assert compass.mark_surfaced([], db_path=db) == 0
+
+    def test_recall_stopwords_do_not_drive_ranking(self, db):
+        """Filler words in the prompt cannot outrank topic words (FPLAN-0332)."""
+        compass.add_decision(
+            "docker sop ctx with the and on for filler heavy text",
+            "push dev first clone in container",
+            "good",
+            db_path=db,
+        )
+        compass.add_decision(
+            "compass curation ctx",
+            "supersedes links archive corrections",
+            "good",
+            db_path=db,
+        )
+        hits = compass.recall_decisions(
+            "lets keep working on the compass supersedes links and curation",
+            limit=2,
+            db_path=db,
+        )
+        assert hits[0]["decision"] == "supersedes links archive corrections"
+
+    def test_recall_all_stopword_prompt_returns_empty(self, db):
+        """A prompt made only of stopwords yields no candidates, no error."""
+        compass.add_decision("some ctx", "some dec", "good", db_path=db)
+        assert compass.recall_decisions("what is it and how do we", db_path=db) == []
@@ -51,6 +51,21 @@ INTERACTIVE_COMMANDS = ("monitor", "audit", "watchdog", "status")
 INTERACTIVE_BRANCHES = ("cli", "backup")


+def _extract_timeout(args: list[str]) -> tuple[list[str], int | None]:
+    """Extract --drone-timeout N from an arg list. Returns (cleaned_args, timeout_or_None)."""
+    if "--drone-timeout" not in args:
+        return args, None
+    idx = args.index("--drone-timeout")
+    if idx + 1 >= len(args):
+        return args, None
+    try:
+        timeout = int(args[idx + 1])
+    except ValueError:
+        logger.info("--drone-timeout value %r is not an integer, ignoring", args[idx + 1])
+        return args, None
+    return args[:idx] + args[idx + 2 :], timeout
+
+
 # =============================================================================
 # AUTO-DISCOVERY
 # =============================================================================
@@ -92,6 +107,7 @@ def show_help() -> None:
    table.add_row("list", "List registered custom commands")
    table.add_row("remove <name>", "Remove a custom command")
    table.add_row("rm <path> [<path>...]", "Contained safe-delete (project + tmp)")
+    table.add_row("--drone-timeout <seconds>", "Override subprocess timeout (default 30s)")
    table.add_row("--help", "Show this help")
    table.add_row("--version", "Show version")

@@ -340,6 +356,7 @@ def _handle_custom_command(args: list[str]) -> int:
    target = cmd_data["target"]
    command = cmd_data["command"]
    cmd_args = list(cmd_data.get("args", [])) + remaining_args
+    cmd_args, explicit_timeout = _extract_timeout(cmd_args)
    module_name = target.lstrip("@").lower()

    interactive = command in INTERACTIVE_COMMANDS or module_name in INTERACTIVE_BRANCHES
@@ -349,6 +366,7 @@ def _handle_custom_command(args: list[str]) -> int:
            target,
            command,
            args=cmd_args if cmd_args else None,
+            timeout=explicit_timeout,
            interactive=interactive,
        )
    except (BranchNotFoundError, CommandExecutionError, RegistryError) as exc:
@@ -412,6 +430,7 @@ def _handle_target(args: List[str]) -> int:
    """Handle `drone @target command [args]` or `drone @target --help`."""
    target = args[0]
    rest = args[1:]
+    rest, explicit_timeout = _extract_timeout(rest)
    module_name = target.lstrip("@").lower()

    first_cmd = rest[0] if rest and rest[0] not in ("--help", "-h") else None
@@ -470,6 +489,7 @@ def _handle_target(args: List[str]) -> int:
            target,
            command,
            args=cmd_args if cmd_args else None,
+            timeout=explicit_timeout,
            interactive=interactive,
        )
    except (BranchNotFoundError, CommandExecutionError, RegistryError) as exc:
@@ -21,6 +21,29 @@ from .exceptions import CommandExecutionError
 from aipass.drone.apps.handlers.json import json_handler


+DEFAULT_TIMEOUT = 30
+
+TIMEOUT_OVERRIDES: dict[str, dict[str, int]] = {
+    "memory": {"process-plans": 120},
+    "flow": {"close": 90},
+}
+
+
+def resolve_timeout(branch: str, command: str | None, explicit: int | None = None) -> int:
+    """Resolve subprocess timeout for a branch command.
+
+    Priority: explicit flag > per-command policy > DEFAULT_TIMEOUT.
+    """
+    if explicit is not None:
+        return explicit
+    branch_key = branch.lstrip("@").lower()
+    if command and branch_key in TIMEOUT_OVERRIDES:
+        cmd_timeout = TIMEOUT_OVERRIDES[branch_key].get(command)
+        if cmd_timeout is not None:
+            return cmd_timeout
+    return DEFAULT_TIMEOUT
+
+
@dataclass
 class CommandResult:
    """Result of a routed command execution."""
@@ -82,7 +105,10 @@ def execute_command(
            return CommandResult(stdout="", stderr="", exit_code=130, branch="", command="")
        raise
    except subprocess.TimeoutExpired as e:
-        raise CommandExecutionError(f"Command timed out after {timeout}s: {' '.join(full_cmd)}") from e
+        raise CommandExecutionError(
+            f"Command timed out after {timeout}s: {' '.join(full_cmd)}\n"
+            f"  Override with: drone @<target> <command> --drone-timeout <seconds>"
+        ) from e
    except FileNotFoundError as e:
        raise CommandExecutionError(f"Executable not found: {executable!r}") from e
    except OSError as e:
@@ -20,7 +20,7 @@ from typing import Dict, List, Optional

 from aipass.prax.apps.modules.logger import system_logger
 from aipass.cli.apps.modules import console
-from aipass.drone.apps.handlers.executor import CommandResult
+from aipass.drone.apps.handlers.executor import CommandResult, resolve_timeout
 from aipass.drone.apps.handlers.json import json_handler
 from aipass.drone.apps.handlers.router_handler import (
    detect_caller_branch_name,
@@ -90,28 +90,38 @@ def route_command(
    target: str,
    command: Optional[str] = None,
    args: Optional[List[str]] = None,
-    timeout: int = 30,
+    timeout: int | None = None,
    interactive: bool = False,
 ) -> CommandResult:
    """Route a command to a branch's entry point.

    Resolves @target to a path, then delegates to the handler for execution.
    When command is None, runs the branch with no args (introspection).
+
+    Timeout resolution: explicit value > per-command policy > DEFAULT_TIMEOUT.
    """
    branch_path = resolve_branch(target)
    branch_name = target.lstrip("@").lower()
+    resolved_timeout = resolve_timeout(branch_name, command, timeout)

    caller = detect_caller_branch_name(Path.cwd())
    if not caller:
        caller = os.environ.get("AIPASS_BRANCH_NAME")
    caller_tag = f" [CALLER:{caller.upper()}]" if caller else ""
-    logger.info("Routing @%s%s → %s %s", branch_name, caller_tag, command or "(introspection)", args or [])
+    logger.info(
+        "Routing @%s%s → %s %s (timeout=%ds)",
+        branch_name,
+        caller_tag,
+        command or "(introspection)",
+        args or [],
+        resolved_timeout,
+    )
    return execute_branch_command(
        branch_path=branch_path,
        branch_name=branch_name,
        command=command,
        args=args,
-        timeout=timeout,
+        timeout=resolved_timeout,
        interactive=interactive,
    )

@@ -147,7 +157,7 @@ def print_introspection():
 def route_all(
    command: str,
    args: Optional[List[str]] = None,
-    timeout: int = 30,
+    timeout: int | None = None,
 ) -> Dict[str, CommandResult]:
    """Route the same command to ALL active branches in the registry."""
    if args is None:
@@ -355,6 +355,7 @@ class TestHandleCustomCommand:
            "@seedgo",
            "audit",
            args=["aipass"],
+            timeout=None,
            interactive=True,
        )

@@ -380,6 +381,7 @@ class TestHandleCustomCommand:
            "@seedgo",
            "audit",
            args=["aipass", "@drone"],
+            timeout=None,
            interactive=True,
        )

@@ -616,6 +618,7 @@ class TestMainIntegration:
            "@seedgo",
            "audit",
            args=["aipass"],
+            timeout=None,
            interactive=True,
        )

@@ -642,6 +645,7 @@ class TestMainIntegration:
            "@seedgo",
            "audit",
            args=["aipass", "@drone"],
+            timeout=None,
            interactive=True,
        )

@@ -714,5 +718,6 @@ class TestMatchCommandIntegration:
            "@flow",
            "create",
            args=["--type=plan", "my-plan"],
+            timeout=None,
            interactive=False,
        )
@@ -903,3 +903,80 @@ class TestAipassIntercept:
        ):
            result = main()
        assert result == 0
+
+
+# ---------------------------------------------------------------------------
+# _extract_timeout — --timeout flag parsing
+# ---------------------------------------------------------------------------
+
+
+class TestExtractTimeout:
+    """Tests for --drone-timeout flag extraction from arg lists."""
+
+    def test_no_flag(self) -> None:
+        """Args without --drone-timeout pass through unchanged."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        args = ["close", "FPLAN-0313"]
+        cleaned, timeout = _extract_timeout(args)
+        assert cleaned == ["close", "FPLAN-0313"]
+        assert timeout is None
+
+    def test_flag_at_end(self) -> None:
+        """--drone-timeout N at end of args is extracted."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        cleaned, timeout = _extract_timeout(["process-plans", "--drone-timeout", "120"])
+        assert cleaned == ["process-plans"]
+        assert timeout == 120
+
+    def test_flag_at_start(self) -> None:
+        """--drone-timeout N at start of args is extracted."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        cleaned, timeout = _extract_timeout(["--drone-timeout", "90", "close", "FPLAN-0313"])
+        assert cleaned == ["close", "FPLAN-0313"]
+        assert timeout == 90
+
+    def test_flag_in_middle(self) -> None:
+        """--drone-timeout N in the middle of args is extracted."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        cleaned, timeout = _extract_timeout(["close", "--drone-timeout", "60", "FPLAN-0313"])
+        assert cleaned == ["close", "FPLAN-0313"]
+        assert timeout == 60
+
+    def test_flag_without_value(self) -> None:
+        """--drone-timeout at end with no value returns None and leaves args."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        args = ["close", "--drone-timeout"]
+        cleaned, timeout = _extract_timeout(args)
+        assert cleaned == args
+        assert timeout is None
+
+    def test_flag_non_integer_value(self) -> None:
+        """--drone-timeout with non-integer value returns None and leaves args."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        args = ["close", "--drone-timeout", "abc"]
+        cleaned, timeout = _extract_timeout(args)
+        assert cleaned == args
+        assert timeout is None
+
+    def test_empty_args(self) -> None:
+        """Empty arg list returns empty with None timeout."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        cleaned, timeout = _extract_timeout([])
+        assert cleaned == []
+        assert timeout is None
+
+    def test_plain_timeout_passes_through(self) -> None:
+        """--timeout (without drone- prefix) is NOT consumed — passes to target."""
+        from aipass.drone.apps.drone import _extract_timeout
+
+        args = ["watchdog", "agent", "@memory", "--timeout", "1800"]
+        cleaned, timeout = _extract_timeout(args)
+        assert cleaned == args
+        assert timeout is None
@@ -8,7 +8,12 @@ from unittest.mock import patch
 import pytest

 from aipass.drone.apps.handlers.exceptions import CommandExecutionError
-from aipass.drone.apps.handlers.executor import execute_command
+from aipass.drone.apps.handlers.executor import (
+    DEFAULT_TIMEOUT,
+    TIMEOUT_OVERRIDES,
+    execute_command,
+    resolve_timeout,
+)


 # ---------------------------------------------------------------------------
@@ -389,3 +394,63 @@ class TestShellSecurity:
        # The semicolon is treated as literal text, not a shell separator
        assert result.stdout.strip() == "hello; echo pwned"
        assert result.exit_code == 0
+
+
+# ---------------------------------------------------------------------------
+# 11. resolve_timeout — policy resolution
+# ---------------------------------------------------------------------------
+
+
+class TestResolveTimeout:
+    """Timeout resolution: explicit > policy > default."""
+
+    def test_default_timeout(self):
+        """Unknown branch+command returns DEFAULT_TIMEOUT."""
+        assert resolve_timeout("unknown", "whatever") == DEFAULT_TIMEOUT
+
+    def test_policy_override(self):
+        """Known branch+command returns the policy value."""
+        for branch, cmds in TIMEOUT_OVERRIDES.items():
+            for cmd, expected in cmds.items():
+                assert resolve_timeout(branch, cmd) == expected
+
+    def test_explicit_wins_over_policy(self):
+        """Explicit timeout overrides the policy map."""
+        branch = next(iter(TIMEOUT_OVERRIDES))
+        cmd = next(iter(TIMEOUT_OVERRIDES[branch]))
+        assert resolve_timeout(branch, cmd, explicit=999) == 999
+
+    def test_explicit_wins_over_default(self):
+        """Explicit timeout overrides the default."""
+        assert resolve_timeout("unknown", "whatever", explicit=42) == 42
+
+    def test_none_command_returns_default(self):
+        """None command (introspection) returns default."""
+        assert resolve_timeout("memory", None) == DEFAULT_TIMEOUT
+
+    def test_at_prefix_stripped(self):
+        """Leading @ on branch name is stripped before lookup."""
+        for branch in TIMEOUT_OVERRIDES:
+            cmd = next(iter(TIMEOUT_OVERRIDES[branch]))
+            expected = TIMEOUT_OVERRIDES[branch][cmd]
+            assert resolve_timeout(f"@{branch}", cmd) == expected
+
+
+# ---------------------------------------------------------------------------
+# 12. Timeout error message includes --timeout hint
+# ---------------------------------------------------------------------------
+
+
+class TestTimeoutErrorMessage:
+    """Timeout error tells the caller how to override."""
+
+    def test_timeout_error_includes_override_hint(self, temp_test_dir: Path):
+        """The timeout error message mentions --timeout."""
+        with pytest.raises(CommandExecutionError, match="--drone-timeout") as exc_info:
+            execute_command(
+                sys.executable,
+                ["-c", "import time; time.sleep(10)"],
+                cwd=str(temp_test_dir),
+                timeout=1,
+            )
+        assert "--drone-timeout" in str(exc_info.value)
@@ -19,13 +19,16 @@ Key Functions:
 - verify_and_heal_orphaned_plans() - Orphan healing logic
 """

+# ruff: noqa: E402
 from pathlib import Path

 _PKG_ROOT = Path(__file__).resolve().parents[4]

 # Standard imports
 import json
+import os
 import shutil
+import time
 from datetime import datetime, timezone
 from typing import Dict, List, Any

@@ -42,6 +45,35 @@ from aipass.prax.apps.modules.logger import system_logger as logger
 FLOW_ROOT = _PKG_ROOT / "flow"
 FLOW_JSON_DIR = FLOW_ROOT / "flow_json"

+MODULE_NAME = "mbank_process"
+_LOCK_RETRIES = 10
+_LOCK_BACKOFF_BASE = 0.05
+
+
+def _acquire_lock(lock_path: Path) -> bool:
+    """Atomically acquire a lockfile via O_CREAT|O_EXCL with retry+backoff."""
+    for attempt in range(_LOCK_RETRIES):
+        try:
+            fd = os.open(str(lock_path), os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+            os.write(fd, str(os.getpid()).encode())
+            os.close(fd)
+            return True
+        except FileExistsError:
+            logger.info("[%s] Lock contention on %s, retry %d", MODULE_NAME, lock_path, attempt + 1)
+            time.sleep(_LOCK_BACKOFF_BASE * (2**attempt))
+        except OSError as exc:
+            logger.warning("[%s] Lock creation failed for %s: %s", MODULE_NAME, lock_path, exc)
+            return False
+    return False
+
+
+def _release_lock(lock_path: Path) -> None:
+    """Remove lockfile, tolerating already-removed."""
+    try:
+        lock_path.unlink(missing_ok=True)
+    except OSError as exc:
+        logger.warning("[%s] Could not release lock %s: %s", MODULE_NAME, lock_path, exc)
+

 def _find_repo_root() -> Path:
    """Walk up from this file to find the repo root (contains AIPASS_REGISTRY.json)."""
@@ -97,12 +129,22 @@ def load_flow_registry(registry_file: str | None = None) -> Dict[str, Any]:


 def save_flow_registry(registry: Dict[str, Any], registry_file: str | None = None) -> None:
-    """Save a plan registry."""
+    """Save a plan registry with lockfile + atomic write."""
    target = FLOW_JSON_DIR / registry_file if registry_file else REGISTRY_FILE
+    lock_path = target.with_suffix(".lock")
+
    try:
-        registry["last_updated"] = datetime.now(timezone.utc).isoformat()
-        with open(target, "w", encoding="utf-8") as f:
-            json.dump(registry, f, indent=2, ensure_ascii=False)
+        if not _acquire_lock(lock_path):
+            raise OSError(f"Could not acquire lock for {target}")
+
+        try:
+            registry["last_updated"] = datetime.now(timezone.utc).isoformat()
+            tmp_path = target.with_suffix(".tmp")
+            with open(tmp_path, "w", encoding="utf-8") as f:
+                json.dump(registry, f, indent=2, ensure_ascii=False)
+            os.replace(str(tmp_path), str(target))
+        finally:
+            _release_lock(lock_path)
    except Exception as e:
        raise Exception(f"Failed to save flow registry: {e}")

@@ -361,7 +403,10 @@ def is_template_content(content: str) -> bool:
 #         today = datetime.now().strftime("%Y%m%d")
 #         plan_num = plan_path.stem.replace("FPLAN-", "")
 #         template_suffix = "-TEMP" if is_template else ""
-#         filename = f"{folder_context}-{analysis['type']}-{analysis['category']}-{analysis['action']}-FPLAN-{plan_num}{template_suffix}-{today}.md"
+#         filename = (
+#             f"{folder_context}-{analysis['type']}-{analysis['category']}"
+#             f"-{analysis['action']}-FPLAN-{plan_num}{template_suffix}-{today}.md"
+#         )
 #
 #         filename = re.sub(r'[<>:"|?*]', '-', filename)
 #         filename = re.sub(r'-+', '-', filename)
@@ -627,7 +672,7 @@ def process_closed_plans() -> Dict[str, Any]:

                if archive_success:
                    processed_count += 1
-                    # Vector intake handled by close_ops.py via drone @memory process-plans
+                    # Vector intake triggered by post_close_runner via direct import
                    results.append({"plan": plan_label, "status": "archived", "correlation_id": correlation_id})
                else:
                    error_count += 1
@@ -3,7 +3,7 @@
 # Description: Closed Plans Local Registry Handler
 # Version: 0.1.0
 # Created: 2026-03-03
-# Modified: 2026-03-03
+# Modified: 2026-07-15
 # =============================================

 """
@@ -13,8 +13,11 @@ Appends a closed plan entry to the branch's CLOSED_PLANS.local.json file.
 Creates the file if it doesn't exist.
 """

+# ruff: noqa: E402
 import json
+import os
 import re
+import time
 from pathlib import Path

 # INFRASTRUCTURE IMPORT PATTERN
@@ -27,6 +30,31 @@ from aipass.flow.apps.handlers.json import json_handler
 MODULE_NAME = "append_closed_plan"
 CLOSED_PLANS_FILE = "CLOSED_PLANS.local.json"

+_LOCK_RETRIES = 10
+_LOCK_BACKOFF_BASE = 0.05
+
+
+def _acquire_append_lock(lock_path: Path) -> bool:
+    """Atomically acquire a lockfile via O_CREAT|O_EXCL with retry+backoff."""
+    for attempt in range(_LOCK_RETRIES):
+        try:
+            fd = os.open(str(lock_path), os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+            os.write(fd, str(os.getpid()).encode())
+            os.close(fd)
+            return True
+        except FileExistsError:
+            logger.info("[%s] Lock contention on %s, retry %d", MODULE_NAME, lock_path, attempt + 1)
+            time.sleep(_LOCK_BACKOFF_BASE * (2**attempt))
+    return False
+
+
+def _release_append_lock(lock_path: Path) -> None:
+    """Remove lockfile, tolerating already-removed."""
+    try:
+        lock_path.unlink(missing_ok=True)
+    except OSError as exc:
+        logger.warning("[%s] Could not release lock %s: %s", MODULE_NAME, lock_path, exc)
+

 def append_to_closed_plans(plan_key: str, plan_info: dict, plan_location: Path) -> bool:
    """
@@ -61,27 +89,35 @@ def append_to_closed_plans(plan_key: str, plan_info: dict, plan_location: Path)
            "location": plan_info.get("relative_path", ""),
        }

-        # Read existing file or create new structure
+        # Locked read-modify-write to prevent lost updates under concurrent close
        closed_plans_path = plan_location / CLOSED_PLANS_FILE
+        lock_path = closed_plans_path.with_suffix(".lock")

-        if closed_plans_path.exists():
-            with open(closed_plans_path, "r", encoding="utf-8") as f:
-                data = json.load(f)
-        else:
-            data = {"closed_plans": []}
+        if not _acquire_append_lock(lock_path):
+            logger.error(
+                f"[{MODULE_NAME}] Could not acquire lock for {closed_plans_path} after {_LOCK_RETRIES} retries"
+            )
+            return False

-        # Check for duplicate plan_id before appending
-        existing_ids = {p.get("plan_id") for p in data.get("closed_plans", [])}
-        if plan_id in existing_ids:
-            logger.info(f"[{MODULE_NAME}] {plan_id} already in {CLOSED_PLANS_FILE} at {plan_location}, skipping")
-            return True
+        try:
+            if closed_plans_path.exists():
+                with open(closed_plans_path, "r", encoding="utf-8") as f:
+                    data = json.load(f)
+            else:
+                data = {"closed_plans": []}

-        # Append and write
-        data["closed_plans"].append(entry)
+            existing_ids = {p.get("plan_id") for p in data.get("closed_plans", [])}
+            if plan_id in existing_ids:
+                logger.info(f"[{MODULE_NAME}] {plan_id} already in {CLOSED_PLANS_FILE} at {plan_location}, skipping")
+                return True

-        with open(closed_plans_path, "w", encoding="utf-8") as f:
-            json.dump(data, f, indent=2, ensure_ascii=False)
-            f.write("\n")
+            data["closed_plans"].append(entry)
+
+            with open(closed_plans_path, "w", encoding="utf-8") as f:
+                json.dump(data, f, indent=2, ensure_ascii=False)
+                f.write("\n")
+        finally:
+            _release_append_lock(lock_path)

        logger.info(f"[{MODULE_NAME}] Appended {plan_id} to {closed_plans_path}")
        json_handler.log_operation(
@@ -379,9 +379,22 @@ def close_plan_impl(
        try:
            from aipass.flow.apps.handlers.plan.append_closed_plan import append_to_closed_plans

-            append_to_closed_plans(plan_key, plan_info, plan_file.parent)
+            if not append_to_closed_plans(plan_key, plan_info, plan_file.parent):
+                logger.error(f"[{MODULE_NAME}] CLOSED_PLANS append failed for {plan_prefix}-{plan_key}")
+                messages.append(
+                    {
+                        "type": "warning",
+                        "text": f"  CLOSED_PLANS append failed for {plan_prefix}-{plan_key}",
+                    }
+                )
        except Exception as e:
-            logger.warning(f"[{MODULE_NAME}] CLOSED_PLANS update failed (non-critical): {e}")
+            logger.error(f"[{MODULE_NAME}] CLOSED_PLANS update failed: {e}")
+            messages.append(
+                {
+                    "type": "warning",
+                    "text": f"  CLOSED_PLANS update failed: {e}",
+                }
+            )

        # Fire trigger event for plan closure
        if trigger_fire_fn is not None:
@@ -25,6 +25,8 @@ Usage:
 """

 import json
+import os
+import time
 from pathlib import Path
 from datetime import datetime, timezone
 from typing import Dict, Any
@@ -44,6 +46,35 @@ MODULE_NAME = "save_registry"
 FLOW_JSON_DIR = FLOW_ROOT / "flow_json"
 REGISTRY_FILE = FLOW_JSON_DIR / "fplan_registry.json"

+_LOCK_RETRIES = 10
+_LOCK_BACKOFF_BASE = 0.05
+
+
+def _acquire_lock(lock_path: Path) -> bool:
+    """Atomically acquire a lockfile via O_CREAT|O_EXCL with retry+backoff."""
+    for attempt in range(_LOCK_RETRIES):
+        try:
+            fd = os.open(str(lock_path), os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+            os.write(fd, str(os.getpid()).encode())
+            os.close(fd)
+            return True
+        except FileExistsError:
+            logger.info("[%s] Lock contention on %s, retry %d", MODULE_NAME, lock_path, attempt + 1)
+            time.sleep(_LOCK_BACKOFF_BASE * (2**attempt))
+        except OSError as exc:
+            logger.warning("[%s] Lock creation failed for %s: %s", MODULE_NAME, lock_path, exc)
+            return False
+    return False
+
+
+def _release_lock(lock_path: Path) -> None:
+    """Remove lockfile, tolerating already-removed."""
+    try:
+        lock_path.unlink(missing_ok=True)
+    except OSError as exc:
+        logger.warning("[%s] Could not release lock %s: %s", MODULE_NAME, lock_path, exc)
+
+
 # =============================================
 # HANDLER FUNCTION
 # =============================================
@@ -64,15 +95,30 @@ def save_registry(registry: Dict[str, Any], registry_file: str | None = None) ->

    Automatically updates the last_updated timestamp before saving.
    Creates the flow_json directory if it doesn't exist.
+    Uses a lockfile to serialize concurrent writes and atomic
+    tempfile+rename to prevent torn reads.
    """
    target = FLOW_JSON_DIR / registry_file if registry_file else REGISTRY_FILE
+    lock_path = target.with_suffix(".lock")

    try:
        FLOW_JSON_DIR.mkdir(parents=True, exist_ok=True)
-        registry["_notice"] = "DO NOT MANUALLY EDIT — managed by flow close pipeline"
-        registry["last_updated"] = datetime.now(timezone.utc).isoformat()
-        with open(target, "w", encoding="utf-8") as f:
-            json.dump(registry, f, indent=2, ensure_ascii=False)
+
+        if not _acquire_lock(lock_path):
+            logger.error("[%s] Could not acquire lock for %s after %d retries", MODULE_NAME, target, _LOCK_RETRIES)
+            return False
+
+        try:
+            registry["_notice"] = "DO NOT MANUALLY EDIT — managed by flow close pipeline"
+            registry["last_updated"] = datetime.now(timezone.utc).isoformat()
+
+            tmp_path = target.with_suffix(".tmp")
+            with open(tmp_path, "w", encoding="utf-8") as f:
+                json.dump(registry, f, indent=2, ensure_ascii=False)
+            os.replace(str(tmp_path), str(target))
+        finally:
+            _release_lock(lock_path)
+
        json_handler.log_operation(
            "registry_saved",
            {
@@ -32,7 +32,7 @@ if sys.platform == "win32":

 from pathlib import Path

-from aipass.cli.apps.modules import console, error, warning
+from aipass.cli.apps.modules import console, error, success, warning
 from aipass.flow.apps.handlers.json import json_handler
 from aipass.flow.apps.handlers.mbank.process import process_closed_plans
 from aipass.flow.apps.handlers.runner.lock_ops import acquire_lock, release_lock
@@ -80,7 +80,25 @@ def handle_command(command: str, args: list) -> bool:

    try:
        process_closed_plans()
-        console.print("[green]Processing complete[/green]")
+
+        try:
+            import importlib
+
+            _plans_mod = importlib.import_module("aipass.memory.apps.handlers.intake.plans_processor")
+            result = _plans_mod.process_plans()
+            if result.get("success"):
+                count = result.get("files_processed", 0)
+                chunks = result.get("total_chunks", 0)
+                if count > 0:
+                    success(f"Vectorized {count} plan(s) ({chunks} chunks)")
+                logger.info("[%s] Plan vectorization: %s", MODULE_NAME, result)
+            else:
+                logger.error("[%s] Plan vectorization failed: %s", MODULE_NAME, result.get("error", "unknown"))
+                error(f"Vectorization failed: {result.get('error', 'unknown')}")
+        except Exception as e:
+            logger.error("[%s] Plan vectorization error: %s", MODULE_NAME, e)
+            error(f"Vectorization error: {e}")
+
    except Exception as e:
        logger.error(f"[{MODULE_NAME}] Background processing failed: {e}")
        error(f"Processing failed: {e}")
@@ -130,6 +148,19 @@ if __name__ == "__main__":

    try:
        process_closed_plans()
+
+        try:
+            import importlib
+
+            _plans_mod = importlib.import_module("aipass.memory.apps.handlers.intake.plans_processor")
+            result = _plans_mod.process_plans()
+            if result.get("success"):
+                logger.info("[%s] Plan vectorization: %s", MODULE_NAME, result)
+            else:
+                logger.error("[%s] Plan vectorization failed: %s", MODULE_NAME, result.get("error", "unknown"))
+        except Exception as e:
+            logger.error("[%s] Plan vectorization error: %s", MODULE_NAME, e)
+
    except Exception as e:
        logger.error(f"[{MODULE_NAME}] Background processing failed: {e}")
    finally:
@@ -1,5 +1,7 @@
 # {plan_number} - {subject} (MERGE)

+> **Create:** `drone @flow create . "Merge summary" merge pplan` (template name before type)
+
 **Created**: {today}
 **Branch**: {location}
 **Status**: Active
@@ -59,10 +61,9 @@ just that one merge commit — **cosmetic and trivially resolved**.
  WORKS** — a clean fast-forward realign, no merge commit created, no history rewrite.
 - **Realign dev to even** (recommended): `drone @git sync` from dev (clean FF), or
  manually `git merge --ff-only origin/main` on dev. No force-push, no rebase needed.
- **Sync local `main` ref WITHOUT checkout**: `git fetch origin main:main` — updates the
-  local main ref to match origin with **zero working-tree touch, no checkout**. This is the
-  answer to the IDE "switch to main → your local changes would be overwritten by checkout"
-  dialog: that dialog is git SAFETY working — **Cancel, never Force Checkout**. You never
+- **Local main behind?** `drone @git sync` from dev handles it (stays on dev, clean FF).
+  If the IDE shows "switch to main → your local changes would be overwritten by checkout" —
+  that dialog is git SAFETY working — **Cancel, never Force Checkout**. You never
  need to stand on main.

 ---
@@ -73,13 +74,14 @@ just that one merge commit — **cosmetic and trivially resolved**.
 - [ ] Confirm what's shipping — scan uncommitted changes + already-pushed dev commits ahead of main: `git rev-list --count main..dev` (read git, raw ok)
 - [ ] No surprise files (stray `/tmp` artifacts, test pollution, `.recovery`/`.archive` churn). Clean = archive, never delete.
 - [ ] **Version state check** (informs the bump decision): read the **two** release-tied versions — `grep '^version' pyproject.toml` and `grep __version__ src/aipass/__init__.py` (they should match; if drifted, note it) — and what PyPI already has: `curl -s https://pypi.org/pypi/aipass/json | python3 -c "import sys,json;print(json.load(sys.stdin)['info']['version'])"`. PyPI rejects a duplicate, so the target must be > published.
- [ ] Decide: **release tag this merge?** (tag = PyPI publish + GitHub Release). If yes, note target version. (Significance call is the user's — the PATCH-default rule below is guidance, and the actual release history is a useful tie-breaker.)
+- [ ] **Release tag: default YES with PATCH bump.** Every dev-to-main merge ships a PATCH bump + tag so PyPI always tracks main (Patrick ruling S318, 2026-07-17 — version numbers carry no significance during beta). Override to MINOR/MAJOR only when warranted; skip only if explicitly told.

 ## 2. Verify, commit, CHANGELOG

 - [ ] **Run the CI audit gate LOCALLY before pushing** (local == CI, S199 parity — catches red before the PR): `cd <repo-root> && .venv/bin/python .github/scripts/seedgo_audit.py` → expect all 13 branches `>=100%`, exit 0. Uses a relative `src/aipass` path, so run from the repo **root**, not a branch dir.
 - [ ] Update `CHANGELOG.md` — add entries under a dated section header `## [YYYY-MM-DD]` (the merge date), one section per merge. Sort into Added / Changed / Fixed.
 - [ ] Commit: `drone @git commit "msg" --all` (from a branch dir, e.g. devpulse). New/untracked files (e.g. new templates) — confirm they got staged: `git ls-files <path>` after; `--all` may not pick up untracked.
+- [ ] All commits are auto-SSH-signed via repo-level git config (key `~/.ssh/aipass_signing`, wired 2026-07-15). Nothing manual required; verify with `git log --show-signature -1` if in doubt.
 - [ ] Every commit pushed — local-only commits are invisible

 ## 3. Open / update the PR
@@ -95,6 +97,7 @@ The PR gate (verified against `.github/workflows/`):
 - [ ] `security.yml` → Security Scan / dependency-scan
 - [ ] `e2e-wheel.yml` → 3-OS wheel smoke (path-filtered: fires on `src/**`, `tests/e2e/**`, `pyproject.toml`)
 - [ ] `windows-test.yml` / `macos-test.yml` → required checks, run on every PR (must NEVER be path-filtered or they park as "Expected/waiting" forever and block merge)
+- [ ] **Hash-pinned CI deps:** tool installs are pinned from `.github/requirements/*.txt` locks. A new security advisory against a pinned dep (esp. pip-audit's 29-package tree) can red `security.yml` with zero code change on our side. Fix = dependabot lock bump (grouped weekly, label `ci`) or regen via the `pip-compile` command in each `.in` file header — never a code revert.
 - [ ] If "all green but can't merge": it's usually post-push mergeability **lag**. Confirm ground truth via the public API (no gh, no gate):
      - `curl -s https://api.github.com/repos/AIOSAI/AIPass/commits/<sha>/check-runs` → all check-runs success (incl. app checks: codecov, CodeQL)
      - `curl -s https://api.github.com/repos/AIOSAI/AIPass/pulls/<n>` → `mergeable_state: clean`
@@ -110,20 +113,22 @@ The PR gate (verified against `.github/workflows/`):

 - [ ] **Expect `dev` to show "1 behind main" — that's the merge commit, it's cosmetic + fast-forwardable.** See "Why dev shows behind main" up top.
 - [ ] **Realign dev** (recommended): `drone @git sync` from dev, or `git merge --ff-only origin/main` on dev. Clean FF, no merge commit, no rewrite.
- [ ] **Stay on `dev`. Do not check out `main`.** Local main being behind is fine — sync it without checkout: `git fetch origin main:main` (zero working-tree touch).
+- [ ] **Stay on `dev`. Do not check out `main`.** Local main being behind is fine — `drone @git sync` from dev covers it.
 - [ ] Never rebase, never reset, never checkout main.
 - [ ] Dependabot / other PRs targeting main: they go green once main has the fix + bots rebase — check after the push

-## 7. Release tag (only if cutting a release)
+## 7. Release tag

-**Versioning rule — bump by SIGNIFICANCE, not cadence:**
- **PATCH** (`x.y.Z+1`) = fix / internal / standards / UX only → the default for most merges
+**Standing default: PATCH bump every merge** (Patrick ruling S318, 2026-07-17). PyPI should always track main; version numbers carry no significance during beta. Big jump reserved for beta exit.
+
+Reference (SemVer — for when significance matters post-beta):
+- **PATCH** (`x.y.Z+1`) = fix / internal / standards / UX only
 - **MINOR** (`x.Y+1.0`) = a new backward-compatible user-facing feature shipped
 - **MAJOR** (`X+1.0.0`) = breaking public-API change

 (aipass is a 2.x library others pin → keep SemVer; the CHANGELOG uses `YYYY-MM-DD` dated section headers.)

-How the release fires (verified `publish.yml`): a `v*` **git tag push** runs build → PyPI publish → GitHub Release. Key facts:
+How the release fires (verified `publish.yml`): a `v*` **git tag push** runs build → provenance attestation → PyPI publish → GitHub Release. The attestation step (`actions/attest-build-provenance`, SHA-pinned) runs between build and upload — expect it in the run log; PyPI publish + Release notes extraction unchanged. Key facts:
 - PyPI version = `pyproject.toml [project] version` at the tagged commit — **NOT** the tag string (the tag only *triggers* the build).
 - Tag and `pyproject` version **must match** (`v2.5.2` ⇄ `version = "2.5.2"`), or PyPI publishes the wrong number while the Release is named the tag.
 - PyPI **rejects a duplicate version** → if shipping, you MUST bump.
@@ -406,6 +406,11 @@
      "standard": "json_structure",
      "reason": "Sound handler \u2014 no JSON operations, plays WAV files."
    },
+    {
+      "file": "apps/handlers/config/trust_registry.py",
+      "standard": "unused_function",
+      "reason": "enroll() and revoke() are the public API consumed CROSS-BRANCH by @aipass CLI (init/trust/revoke commands, DPLAN-0244 phase 2). seedgo's intra-branch static analysis cannot see cross-branch callers. read_registry() also exported for @aipass CLI use."
+    },
    {
      "file": "apps/handlers/config/loader.py",
      "standard": "json_structure",
@@ -25,6 +25,7 @@ Every hook event flows through one engine. Platform bridges normalize the event
 | `drone @hooks hooksound` | Show current sound mute status |
 | `drone @hooks hooksound off` | Mute all hook sounds |
 | `drone @hooks hooksound on` | Unmute all hook sounds |
+| `drone @hooks dismiss <alert-id>` | Remove an alert from `.aipass/alerts.json` |
 | `drone @hooks cadence` | Show prompt injection cadence config and state |
 | `drone @hooks verify` | Cross-check provider settings vs project hook config |
 | `drone @hooks --help` | Full help reference |
@@ -40,6 +41,8 @@ Hooks operate on two tiers:

 **Why provider-only wiring?** Claude Code does not fire `PreToolUse`/`PostToolUse` hooks from project-level settings — only from user-level settings (DPLAN-0160 platform limitation). So all hook entries live in provider settings, and per-project control happens through `.aipass/hooks.json`.

+**Deploying new handlers:** Registering a handler in `.aipass/hooks.json` is necessary but not sufficient. Each event type also needs a matching bridge command entry in `~/.claude/settings.json` — this is human-gated (agents cannot edit provider settings). After building a new handler, email @devpulse to wire the settings.json entry. Without it, the engine never receives the event and the handler never fires.
+
 ## Architecture

 ```
@@ -55,6 +58,7 @@ src/aipass/hooks/
 │   │   ├── engine.py            # Core dispatch — routes events to handlers
 │   │   ├── hooksound.py         # Sound control (drone @hooks hooksound on/off)
 │   │   ├── hookstatus.py        # Config viewer (drone @hooks status)
+│   │   ├── alert_dismiss.py      # Dismiss alerts (drone @hooks dismiss <id>)
 │   │   ├── presence.py          # Branch presence — claim/release/refresh for .ai_central/PRESENCE.central.json
 │   │   ├── sandbox.py           # Kernel sandbox — srt/bwrap wrapper + per-role policy generator
 │   │   └── wire_verify.py       # Wire verification — provider ↔ project hook wiring checker
@@ -66,7 +70,8 @@ src/aipass/hooks/
 │   │   │   ├── branch_loader.py #   Injects aipass_local_prompt.md
 │   │   │   ├── tier0_kernel.py  #   Injects tier0 kernel prompt (every turn)
 │   │   │   ├── navmap.py        #   Injects tier1 navmap prompt (periodic)
-│   │   │   └── identity.py      #   Injects passport identity block
+│   │   │   ├── identity.py      #   Injects passport identity block
+│   │   │   └── persistent_alert.py # Injects advisory banners from .aipass/alerts.json
 │   │   ├── security/            # Enforcement hooks
 │   │   │   ├── edit_gate.py     #   Blocks unsafe edits (cross-branch, inbox, diagnostics)
 │   │   │   ├── git_gate.py      #   Enforces git access tiers
@@ -91,7 +96,7 @@ src/aipass/hooks/
 │       └── diagnostics.py       # JSONL logging for hook execution
 ├── logs/
 │   └── engine.jsonl             # JSONL diagnostics (every hook execution)
-└── tests/                       # 913 tests across 28 test files
+└── tests/                       # 1071 tests across 29 test files
 ```

 ## How It Works
@@ -112,7 +117,7 @@ Handlers are called **dynamically at runtime** — the engine uses `importlib.im

 | Event | Hooks | Description |
 |---|---|---|
-| UserPromptSubmit | presence_gate, identity, email, branch_loader, tier0_kernel, navmap | Presence gate + prompt injection + inbox check |
+| UserPromptSubmit | presence_gate, persistent_alert, identity, email, branch_loader, tier0_kernel, navmap, auto_process, user_message_relay | Presence gate + alerts + prompt injection + inbox + auto-process + TG mirror |
 | PreToolUse | tool_sound, edit_gate, git_gate, rm_gate, registry_gate | Security gates + guardrails + sound |
 | PostToolUse | auto_fix, auto_watchdog | Diagnostics + watchdog |
 | SubagentStop | subagent_gate | Seedgo validation |
@@ -140,6 +145,27 @@ The `git_gate` handler (`security/git_gate.py`) enforces git access via drone to

 **Why it's on by default:** Agents reflexively reach for raw git, which causes state chaos in a multi-agent system. The gate redirects to `drone @git` which enforces access tiers (read-only for most branches, write-only for devpulse). External users who don't need multi-agent git orchestration can safely disable it.

+## Persistent Alerts
+
+The `persistent_alert` handler (`prompt/persistent_alert.py`) injects advisory banners into every prompt when active alerts exist. General-purpose — any agent can raise alerts (prax for runaway logs, trigger for medic, backup for sync failures).
+
+**How it works:** Reads `.aipass/alerts.json` at the project root. Each alert has an ID, source, severity (`warning`/`critical`), title, body, and optional `expires_at`. Active alerts render as a banner every turn until dismissed or expired. Expired alerts are auto-cleaned on read.
+
+**Sound:** Piper TTS fires on first injection per alert ID — subsequent turns are silent for known alerts. New alerts trigger a fresh announcement.
+
+**Dismissing alerts:** `drone @hooks dismiss <alert-id>` removes an alert by ID from `alerts.json`.
+
+**Schema:**
+```json
+{
+  "alerts": [{
+    "id": "uuid", "source": "prax", "severity": "warning",
+    "title": "High log rate", "body": "commons exceeds 50 lines/s",
+    "created_at": "iso", "expires_at": "iso or null"
+  }]
+}
+```
+
 ## Kernel Sandbox (srt/bwrap)

 The sandbox module (`apps/modules/sandbox.py`) provides the kernel-level filesystem boundary for agent sessions. It wraps Anthropic's `@anthropic-ai/sandbox-runtime` (srt) library, which uses bubblewrap (bwrap) + Landlock + seccomp on Linux to enforce write/read restrictions at the OS level.
@@ -180,7 +206,7 @@ The @drone broker validates sandbox policy before agent launch. @ai_mail's dispa
 - All branches via hook dispatch — every Claude Code session routes through the engine
 - @ai_mail dispatch_monitor — sandbox_launch + build_policy for agent launch boundary

-*Last Updated: 2026-06-29*
+*Last Updated: 2026-07-15*

 ---

@@ -20,19 +20,39 @@ AIPASS_HOME = os.environ.get("AIPASS_HOME", "")


 def find_project_config() -> dict | None:
-    """Walk up from CWD looking for .aipass/hooks.json."""
+    """Walk up from CWD looking for .aipass/hooks.json, with trust verification."""
+    from aipass.hooks.apps.handlers.config.trust_registry import (
+        REGISTRY_PATH,
+        bootstrap,
+        is_trusted,
+    )
+
    search = Path.cwd()
    home = Path.home()
    while search != home and search.parent != search:
-        config = search / ".aipass" / "hooks.json"
-        if config.exists():
+        config_file = search / ".aipass" / "hooks.json"
+        if config_file.exists():
+            project_dir = str(search)
+
+            if not REGISTRY_PATH.exists():
+                bootstrap()
+
+            if not is_trusted(project_dir):
+                logger.warning(
+                    "[HOOKS] project not enrolled in trust registry: %s (run: aipass init update)",
+                    project_dir,
+                )
+                return None
+
            try:
-                raw = config.read_text(encoding="utf-8")
+                raw = config_file.read_text(encoding="utf-8")
                if AIPASS_HOME:
                    raw = raw.replace("$AIPASS_HOME", AIPASS_HOME)
-                return json.loads(raw)
+                parsed = json.loads(raw)
+                parsed["_source"] = "project"
+                return parsed
            except (json.JSONDecodeError, OSError) as exc:
-                logger.error("[HOOKS] bad config %s: %s", config, exc)
+                logger.error("[HOOKS] bad config %s: %s", config_file, exc)
                return None
        search = search.parent
    return None
@@ -0,0 +1,138 @@
+# =================== AIPass ====================
+# Name: trust_registry.py
+# Version: 1.0.0
+# Description: Trusted-project registry — DPLAN-0244 Layer B
+# Branch: hooks
+# Layer: apps/handlers/config
+# Created: 2026-07-15
+# Modified: 2026-07-15
+# =============================================
+
+"""Trusted-project registry for hook config loading.
+
+Single source of truth for which projects are trusted to have their
+.aipass/hooks.json loaded by the hook engine. Registry lives at
+~/.aipass/trusted_projects.json. @aipass CLI (init/trust/revoke)
+imports this module for enrollment operations.
+"""
+
+import hashlib
+import json
+import os
+from pathlib import Path
+
+from aipass.hooks.apps.handlers.json import json_handler
+from aipass.prax.apps.modules.logger import system_logger as logger
+
+REGISTRY_PATH = Path.home() / ".aipass" / "trusted_projects.json"
+
+
+def _hash_file(path: Path) -> str:
+    """Compute sha256 of a file's contents."""
+    data = path.read_bytes()
+    return f"sha256:{hashlib.sha256(data).hexdigest()}"
+
+
+def read_registry() -> dict:
+    """Read the trusted-project registry. Returns empty registry if absent or corrupt."""
+    if not REGISTRY_PATH.exists():
+        return {"version": 1, "projects": {}}
+    try:
+        data = json_handler.read_json_file(REGISTRY_PATH)
+        if not isinstance(data.get("projects"), dict):
+            return {"version": 1, "projects": {}}
+        return data
+    except (json.JSONDecodeError, OSError) as exc:
+        logger.error("[HOOKS] bad trust registry %s: %s", REGISTRY_PATH, exc)
+        return {"version": 1, "projects": {}}
+
+
+def _write_registry(registry: dict) -> None:
+    """Write the registry to disk, creating parent dirs if needed."""
+    REGISTRY_PATH.parent.mkdir(parents=True, exist_ok=True)
+    json_handler.write_json_file(REGISTRY_PATH, registry)
+
+
+def enroll(project_dir: str) -> bool:
+    """Enroll a project in the trusted registry. Returns True on success."""
+    project_path = Path(project_dir).resolve()
+    config_path = project_path / ".aipass" / "hooks.json"
+    if not config_path.exists():
+        logger.warning("[HOOKS] cannot enroll %s: no .aipass/hooks.json", project_path)
+        return False
+    config_hash = _hash_file(config_path)
+    registry = read_registry()
+    registry["projects"][str(project_path)] = {
+        "enrolled": _isoformat_now(),
+        "config_hash": config_hash,
+        "config_path": str(config_path),
+    }
+    _write_registry(registry)
+    json_handler.log_operation("enroll", {"project": str(project_path)}, module_name="trust_registry")
+    logger.info("[HOOKS] enrolled %s (hash=%s)", project_path, config_hash)
+    return True
+
+
+def revoke(project_dir: str) -> bool:
+    """Remove a project from the trusted registry. Returns True if it was present."""
+    project_path = str(Path(project_dir).resolve())
+    registry = read_registry()
+    if project_path not in registry["projects"]:
+        return False
+    del registry["projects"][project_path]
+    _write_registry(registry)
+    json_handler.log_operation("revoke", {"project": project_path}, module_name="trust_registry")
+    logger.info("[HOOKS] revoked %s", project_path)
+    return True
+
+
+def is_trusted(project_dir: str) -> bool:
+    """Check if a project is enrolled with a matching config hash."""
+    project_path = str(Path(project_dir).resolve())
+    registry = read_registry()
+    entry = registry["projects"].get(project_path)
+    if entry is None:
+        return False
+    config_path = Path(project_dir).resolve() / ".aipass" / "hooks.json"
+    if not config_path.exists():
+        return False
+    current_hash = _hash_file(config_path)
+    return current_hash == entry.get("config_hash", "")
+
+
+def bootstrap() -> bool:
+    """Bootstrap the registry with ONLY the AIPass install. Returns True on success.
+
+    Called when the registry file does not exist. Enrolls the AIPass
+    install identified by $AIPASS_HOME — never the current CWD.
+    """
+    aipass_home = os.environ.get("AIPASS_HOME", "")
+    if not aipass_home:
+        logger.warning("[HOOKS] registry absent and AIPASS_HOME not set — cannot bootstrap")
+        return False
+    aipass_path = Path(aipass_home).resolve()
+    config_path = aipass_path / ".aipass" / "hooks.json"
+    if not config_path.exists():
+        logger.warning(
+            "[HOOKS] registry absent and AIPass hooks.json not found at %s",
+            config_path,
+        )
+        return False
+    config_hash = _hash_file(config_path)
+    registry = {"version": 1, "projects": {}}
+    registry["projects"][str(aipass_path)] = {
+        "enrolled": _isoformat_now(),
+        "config_hash": config_hash,
+        "config_path": str(config_path),
+    }
+    _write_registry(registry)
+    json_handler.log_operation("bootstrap", {"aipass_home": str(aipass_path)}, module_name="trust_registry")
+    logger.info("[HOOKS] registry bootstrapped, enrolled AIPass install: %s", aipass_path)
+    return True
+
+
+def _isoformat_now() -> str:
+    """Return current UTC time as ISO string."""
+    from datetime import datetime, timezone
+
+    return datetime.now(timezone.utc).isoformat()
@@ -0,0 +1,3 @@
+"""JSON Handler — Hooks Branch."""
+
+__all__ = []
@@ -0,0 +1,165 @@
+# =================== AIPass ====================
+# Name: json_handler.py
+# Description: JSON auto-creating handler for hooks data files
+# Version: 1.0.0
+# Created: 2026-07-15
+# Modified: 2026-07-15
+# =============================================
+
+"""JSON auto-creating handler for hooks data files."""
+
+import json
+import os
+import sys
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+import inspect
+
+from aipass.prax.apps.modules.logger import system_logger as logger
+
+if sys.platform == "win32":
+    os.environ.setdefault("PYTHONUTF8", "1")
+    for _stream in (sys.stdout, sys.stderr):
+        _reconfigure = getattr(_stream, "reconfigure", None)
+        if _reconfigure is not None:
+            _reconfigure(encoding="utf-8", errors="replace")
+
+_BRANCH_ROOT = Path(__file__).resolve().parents[3]
+_BRANCH_NAME = _BRANCH_ROOT.name
+JSON_DIR = _BRANCH_ROOT / f"{_BRANCH_NAME}_json"
+
+
+def _get_caller_module_name() -> str:
+    """Auto-detect calling module name from call stack."""
+    stack = inspect.stack()
+    if len(stack) > 2:
+        caller_frame = stack[2]
+        caller_path = Path(caller_frame.filename)
+        module_name = caller_path.stem
+        if module_name and not module_name.startswith("_"):
+            return module_name
+    return "unknown"
+
+
+def _create_default(json_type: str, module_name: str) -> Any:
+    """Create default JSON structure from inline code defaults."""
+    today = datetime.now().date().isoformat()
+    if json_type == "config":
+        return {
+            "module_name": module_name,
+            "version": "1.0.0",
+            "config": {"max_log_entries": 100},
+            "created": today,
+        }
+    elif json_type == "data":
+        return {
+            "module_name": module_name,
+            "created": today,
+            "last_updated": today,
+        }
+    elif json_type == "log":
+        return []
+    raise ValueError(f"Unknown json_type: {json_type}")
+
+
+def validate_json_structure(data: Any, json_type: str) -> bool:
+    """Validate JSON structure matches expected type."""
+    if json_type == "config":
+        return isinstance(data, dict) and all(k in data for k in ["module_name", "version", "config"])
+    elif json_type == "data":
+        return isinstance(data, dict) and all(k in data for k in ["created", "last_updated"])
+    elif json_type == "log":
+        return isinstance(data, list)
+    return False
+
+
+def get_json_path(module_name: str, json_type: str) -> Path:
+    """Get path for module JSON file."""
+    return JSON_DIR / f"{module_name}_{json_type}.json"
+
+
+def ensure_json_exists(module_name: str, json_type: str) -> bool:
+    """Ensure JSON file exists, create from template if missing."""
+    JSON_DIR.mkdir(parents=True, exist_ok=True)
+    json_path = get_json_path(module_name, json_type)
+    if json_path.exists():
+        try:
+            data = json.loads(json_path.read_text(encoding="utf-8"))
+            if validate_json_structure(data, json_type):
+                return True
+        except Exception as exc:
+            logger.warning("[HOOKS] json_handler: ensure_json_exists failed for %s_%s: %s", module_name, json_type, exc)
+    template = _create_default(json_type, module_name)
+    json_path.write_text(json.dumps(template, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+    return True
+
+
+def load_json(module_name: str, json_type: str) -> Any | None:
+    """Load JSON file, auto-create if missing."""
+    if not ensure_json_exists(module_name, json_type):
+        return None
+    json_path = get_json_path(module_name, json_type)
+    return json.loads(json_path.read_text(encoding="utf-8"))
+
+
+def save_json(module_name: str, json_type: str, data: Any) -> bool:
+    """Save JSON file."""
+    json_path = get_json_path(module_name, json_type)
+    if not validate_json_structure(data, json_type):
+        raise ValueError(f"Invalid structure for {json_type} JSON")
+    if json_type == "data" and isinstance(data, dict):
+        data["last_updated"] = datetime.now().date().isoformat()
+    json_path.write_text(json.dumps(data, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+    return True
+
+
+def ensure_module_jsons(module_name: str) -> bool:
+    """Ensure all 3 JSON files exist for a module."""
+    ensure_json_exists(module_name, "config")
+    ensure_json_exists(module_name, "data")
+    ensure_json_exists(module_name, "log")
+    return True
+
+
+def log_operation(
+    operation: str,
+    data: dict[str, Any] | None = None,
+    module_name: str | None = None,
+) -> bool:
+    """Add entry to module log with automatic rotation.
+
+    Auto-detects calling module if module_name not provided.
+    """
+    if module_name is None:
+        module_name = _get_caller_module_name()
+    ensure_module_jsons(module_name)
+
+    config = load_json(module_name, "config")
+    max_entries = 100
+    if config and "config" in config:
+        max_entries = config["config"].get("max_log_entries", 100)
+
+    log = load_json(module_name, "log")
+    if log is None:
+        log = []
+
+    entry: dict[str, Any] = {"timestamp": datetime.now().isoformat(), "operation": operation}
+    if data:
+        entry["data"] = data
+
+    log.append(entry)
+    if len(log) > max_entries:
+        log = log[-max_entries:]
+
+    return save_json(module_name, "log", log)
+
+
+def read_json_file(path: Path) -> Any:
+    """Read and parse a JSON file at an arbitrary path."""
+    return json.loads(path.read_text(encoding="utf-8"))
+
+
+def write_json_file(path: Path, data: Any) -> None:
+    """Write data as JSON to an arbitrary path."""
+    path.write_text(json.dumps(data, indent=2) + "\n", encoding="utf-8")
@@ -1,25 +1,39 @@
 # =================== AIPass ====================
 # Name: session_boot.py
-# Version: 1.1.0
-# Description: Boot wrapper — attach-if-live or start-in-tmux for Claude Code
+# Version: 4.0.0
+# Description: Boot wrapper — attach-first menu for Claude Code sessions
 # Branch: hooks
 # Layer: apps/handlers/lifecycle
 # Created: 2026-06-30
-# Modified: 2026-06-30
+# Modified: 2026-07-14
 # =============================================

 """Boot wrapper for Claude Code sessions.

-When Patrick boots Claude in a branch directory, this wrapper:
-  1. If already inside tmux ($TMUX set) → execs claude directly (no nesting).
-  2. Checks CC-native ~/.claude/sessions/ for a live session at this cwd.
-  3. If live AND hosted in a tmux session → attaches to that tmux session.
-  4. If live but NOT in tmux → warns (can't inject into a plain terminal).
-  5. If no live session → starts fresh inside a new tmux session.
+When Patrick runs `claude` in a branch directory, this wrapper presents a menu:

-Tmux sessions are named after the branch directory (e.g., "hooks", "devpulse").
-All sessions use --permission-mode bypassPermissions (TG user can't answer prompts).
-Remote visibility is preserved (tmux composes with remoteControlAtStartup).
+Live session (interactive):
+  hooks — live chat: PID 1234 · abc12345 · interactive · 2h old
+    [Enter]  resume this chat
+    [n]      start new chat   (closes the one above first)
+    [c]      close it and exit
+
+Live session (background):
+  devpulse — live chat: PID 773292 · c624cbcd · background "chroma review" · 2h old
+    [Enter]  resume this chat   (stops bg, reopens as normal chat)
+    [n]      start new chat   (stops bg first)
+    [c]      close it and exit   (stops bg)
+
+No live session:
+  devpulse — no live chat
+    [Enter]  continue last chat
+    [n]      new chat
+
+All interactive launches are tmux-wrapped (closed terminal = recoverable).
+
+Special cases:
+  - Already inside tmux → execs claude directly (no nesting).
+  - Headless (-p flag) → execs claude directly.

 Entry points:
  drone @hooks boot [claude args...]
@@ -27,6 +41,7 @@ Entry points:

 import os
 import shutil
+import signal
 import subprocess
 import sys
 from pathlib import Path
@@ -37,12 +52,7 @@ _DEFAULT_ARGS = ["--permission-mode", "bypassPermissions"]


 def _resolve_claude_binary() -> str:
-    """Resolve the REAL claude binary path from PATH.
-
-    Uses shutil.which, which searches the filesystem PATH — it does NOT see
-    shell functions. So even when a claude() shell function shadows the binary,
-    this finds the real one.
-    """
+    """Resolve the REAL claude binary path from PATH."""
    path = shutil.which("claude")
    if path:
        return path
@@ -128,8 +138,148 @@ def _is_descendant(target_pid: int, ancestor_pid: int) -> bool:
    return False


+def _format_age(session: dict) -> str:
+    """Format session age from its start time."""
+    started = session.get("startedAt") or session.get("started", "")
+    if not started:
+        return ""
+    try:
+        from datetime import datetime, timezone
+
+        if isinstance(started, (int, float)):
+            start_dt = datetime.fromtimestamp(started / 1000, tz=timezone.utc)
+        else:
+            start_dt = datetime.fromisoformat(str(started).replace("Z", "+00:00"))
+        delta = datetime.now(tz=timezone.utc) - start_dt
+        hours = int(delta.total_seconds() // 3600)
+        minutes = int((delta.total_seconds() % 3600) // 60)
+        if hours > 0:
+            return f"{hours}h{minutes}m"
+        return f"{minutes}m"
+    except Exception as exc:
+        logger.info("[SESSION_BOOT] age format error: %s", exc)
+        return ""
+
+
+def _session_short_id(session: dict) -> str:
+    """Extract first 8 chars of sessionId."""
+    return str(session.get("sessionId", ""))[:8]
+
+
+def _session_label(session: dict, branch: str) -> str:
+    """Format a session's one-line label: PID · short-id · kind [auto-name] · age."""
+    pid = session.get("pid", "?")
+    short_id = _session_short_id(session)
+    kind = session.get("kind", "unknown")
+    auto_name = session.get("name", "")
+    name_str = f' "{auto_name}"' if auto_name else ""
+    age = _format_age(session)
+    age_str = f" · {age} old" if age else ""
+    return f"PID {pid} · {short_id} · {kind}{name_str}{age_str}"
+
+
+def _read_choice(prompt: str = "> ") -> str:
+    """Read a single-line choice from /dev/tty (works even when stdin is piped)."""
+    try:
+        tty = open("/dev/tty", "r", encoding="utf-8")
+        sys.stderr.write(prompt)
+        sys.stderr.flush()
+        choice = tty.readline().strip().lower()
+        tty.close()
+        return choice
+    except OSError as exc:
+        logger.info("[SESSION_BOOT] /dev/tty not available: %s", exc)
+        return ""
+
+
+def _stop_session(session: dict, claude_bin: str) -> str:
+    """Stop a session. Returns description of action taken.
+
+    bg sessions: no per-job stop exists in the CLI. Returns an honest
+    message — never SIGTERMs bg (daemon respawns it).
+    """
+    pid = session.get("pid")
+    kind = session.get("kind", "unknown")
+
+    if kind in ("bg", "background"):
+        logger.info("[SESSION_BOOT] Cannot stop bg PID %s — no per-job stop in CLI", pid)
+        return f"PID {pid}: bg session — no per-job stop available"
+
+    tmux_session = _find_tmux_session_for_pid(pid) if pid else None
+    if tmux_session:
+        subprocess.run(["tmux", "kill-session", "-t", tmux_session], check=False)
+        logger.info("[SESSION_BOOT] Killed tmux session '%s' (PID %d)", tmux_session, pid)
+        return f"PID {pid}: killed tmux session '{tmux_session}'"
+
+    if pid:
+        try:
+            os.kill(pid, signal.SIGTERM)
+            logger.info("[SESSION_BOOT] Sent SIGTERM to PID %d", pid)
+            return f"PID {pid}: sent SIGTERM"
+        except ProcessLookupError:
+            logger.info("[SESSION_BOOT] PID %d already dead", pid)
+            return f"PID {pid}: already dead"
+        except OSError as exc:
+            logger.warning("[SESSION_BOOT] SIGTERM PID %d failed: %s", pid, exc)
+            return f"PID {pid}: SIGTERM failed ({exc})"
+    return f"PID {pid}: no action"
+
+
+def _resume_session(
+    session: dict, branch: str, claude_bin: str, defaults: list[str], extra_args: list[str] | None = None
+) -> dict:
+    """Resume a session — right mechanism per kind.
+
+    bg: takeover (daemon stop + --resume in tmux). Never opens agents view.
+    tmux: attach to existing tmux session.
+    dead-window: --continue in a new tmux session.
+    """
+    pid = session.get("pid")
+    kind = session.get("kind", "unknown")
+    ea = list(extra_args or [])
+
+    if kind in ("bg", "background"):
+        return _takeover_bg(session, branch, claude_bin, defaults, extra_args)
+
+    tmux_session = _find_tmux_session_for_pid(pid) if pid else None
+    if tmux_session:
+        logger.info("[SESSION_BOOT] Attaching to tmux session '%s'", tmux_session)
+        os.execvp("tmux", ["tmux", "attach-session", "-t", tmux_session])
+        return {"exit_code": 0, "action": "attached", "tmux_session": tmux_session}
+
+    logger.info("[SESSION_BOOT] Continuing dead-window session via --continue")
+    sid = session.get("sessionId", "")
+    nf = _name_flag(branch, sid, extra_args)
+    return _exec_in_tmux(branch, "", claude_bin, [claude_bin] + defaults + ["--continue"] + ea + nf)
+
+
+def _make_session_name(branch: str, session_id: str = "") -> str:
+    """Generate tmux session name: branch-shortid."""
+    short_id = session_id[:8] if session_id else ""
+    if short_id:
+        return f"{branch}-{short_id}"
+    return branch
+
+
+def _name_flag(branch: str, session_id: str = "", extra_args: list[str] | None = None) -> list[str]:
+    """Build --name args for session stamping, unless user already provided one."""
+    if extra_args and ("-n" in extra_args or "--name" in extra_args):
+        return []
+    return ["--name", _make_session_name(branch, session_id)]
+
+
+def _exec_in_tmux(branch: str, session_id: str, claude_bin: str, claude_cmd: list[str]) -> dict:
+    """Exec a claude command inside a new tmux session."""
+    session_name = _make_session_name(branch, session_id)
+    if _tmux_session_exists(session_name):
+        subprocess.run(["tmux", "kill-session", "-t", session_name], check=False)
+    logger.info("[SESSION_BOOT] Launching in tmux '%s': %s", session_name, " ".join(claude_cmd))
+    os.execvp("tmux", ["tmux", "new-session", "-s", session_name, "--"] + claude_cmd)
+    return {"exit_code": 0, "action": "started", "tmux_session": session_name}
+
+
 def boot(cwd: str | None = None, extra_args: list[str] | None = None) -> dict:
-    """Boot Claude Code — attach if live, else start in tmux.
+    """Boot Claude Code — present menu when sessions exist.

    Args:
        cwd: Branch directory (defaults to current working directory).
@@ -153,7 +303,7 @@ def boot(cwd: str | None = None, extra_args: list[str] | None = None) -> dict:

    if os.environ.get("TMUX"):
        logger.info("[SESSION_BOOT] Already inside tmux — running claude directly")
-        claude_cmd = [claude_bin] + defaults
+        claude_cmd = [claude_bin] + defaults + _name_flag(branch, extra_args=extra_args)
        if extra_args:
            claude_cmd.extend(extra_args)
        os.execvp(claude_bin, claude_cmd)
@@ -167,40 +317,302 @@ def boot(cwd: str | None = None, extra_args: list[str] | None = None) -> dict:
    live = _find_live_sessions(cwd)

    if live:
-        session = live[0]
-        pid = session["pid"]
-        name = session.get("name", "")
-        logger.info("[SESSION_BOOT] Live session found: PID %d%s", pid, f" ({name})" if name else "")
+        return _menu_live(live, branch, claude_bin, defaults, extra_args)
+    return _menu_no_live(branch, claude_bin, defaults, extra_args)

-        tmux_session = _find_tmux_session_for_pid(pid)
-        if tmux_session:
-            logger.info("[SESSION_BOOT] Attaching to tmux session '%s'", tmux_session)
-            os.execvp("tmux", ["tmux", "attach-session", "-t", tmux_session])
-            return {"exit_code": 0, "action": "attached", "tmux_session": tmux_session}

-        return {
-            "exit_code": 1,
-            "action": "warn",
-            "error": (
-                f"{branch} already has a live Claude session (PID {pid}) running outside tmux"
-                f" — Claude allows one session per branch.\n"
-                f"  • Reattach in its own terminal, OR\n"
-                f"  • Reclaim it here:   kill {pid} && claude\n"
-                f"  • Or bypass this wrapper:   command claude --resume"
-            ),
-        }
+def _has_bg(sessions: list[dict]) -> bool:
+    """Check if any session is a background session."""
+    return any(s.get("kind") in ("bg", "background") for s in sessions)

-    if _tmux_session_exists(branch):
-        logger.info("[SESSION_BOOT] Killing stale tmux session '%s'", branch)
-        subprocess.run(["tmux", "kill-session", "-t", branch], check=False)

-    claude_cmd = [claude_bin] + defaults
+def _get_collateral_bg(branch: str) -> list[dict]:
+    """Find live bg sessions outside the given branch (blast-radius check)."""
+    import importlib
+
+    cc_sessions = importlib.import_module("aipass.hooks.apps.modules.cc_sessions")
+    collateral = []
+    for s in cc_sessions.read_all_sessions():
+        if s.get("kind") not in ("bg", "background"):
+            continue
+        s_branch = Path(s.get("cwd", "")).name
+        if s_branch != branch and s.get("pid") and cc_sessions._is_pid_alive(s["pid"]):
+            collateral.append(s)
+    return collateral
+
+
+def _daemon_stop(claude_bin: str, branch: str, pid: int | None) -> dict:
+    """Run daemon stop --any with blast-radius confirmation.
+
+    Returns {"ok": True} on success, {"ok": False, "error": "..."} on failure.
+    """
+    collateral = _get_collateral_bg(branch)
+    if collateral:
+        sys.stderr.write("  Other branches have live bg sessions that will also stop:\n")
+        for s in collateral:
+            coll_branch = Path(s.get("cwd", "")).name
+            sys.stderr.write(f"    PID {s.get('pid')} · {coll_branch} · {_session_short_id(s)}\n")
+        sys.stderr.write("  Continue? [y/N] ")
+        confirm = _read_choice("")
+        if confirm != "y":
+            return {"ok": False, "error": "cancelled by user"}
+
+    sys.stderr.write("  Stopping background sessions (daemon stop --any)...\n")
+    try:
+        result = subprocess.run(
+            [claude_bin, "daemon", "stop", "--any"],
+            capture_output=True,
+            text=True,
+            timeout=15,
+        )
+        if result.returncode != 0:
+            stderr_msg = result.stderr.strip()
+            logger.warning("[SESSION_BOOT] daemon stop exit %d: %s", result.returncode, stderr_msg)
+            sys.stderr.write(f"  daemon stop failed (exit {result.returncode}): {stderr_msg}\n")
+            return {"ok": False, "error": f"daemon stop exit {result.returncode}: {stderr_msg}"}
+    except (OSError, subprocess.TimeoutExpired) as exc:
+        logger.warning("[SESSION_BOOT] daemon stop failed: %s", exc)
+        sys.stderr.write(f"  daemon stop failed: {exc}\n")
+        return {"ok": False, "error": f"daemon stop failed: {exc}"}
+
+    import time
+
+    for _ in range(10):
+        time.sleep(1)
+        if not _is_session_file_present(pid):
+            break
+    else:
+        logger.warning("[SESSION_BOOT] Session file for PID %s did not clear after daemon stop", pid)
+
+    return {"ok": True}
+
+
+def _takeover_bg(
+    session: dict, branch: str, claude_bin: str, defaults: list[str], extra_args: list[str] | None = None
+) -> dict:
+    """Take over a bg session: daemon stop --any, poll, then --resume in tmux.
+
+    Checks blast radius first (other branches' bg sessions). On daemon stop
+    failure, aborts honestly. Resumes inside a tmux session so a closed
+    terminal is always recoverable.
+    """
+    session_id = session.get("sessionId", "")
+    pid = session.get("pid")
+    ea = list(extra_args or [])
+    nf = _name_flag(branch, session_id, extra_args)
+
+    stop_result = _daemon_stop(claude_bin, branch, pid)
+    if not stop_result["ok"]:
+        return {"exit_code": 1, "error": stop_result["error"]}
+
+    if session_id:
+        logger.info("[SESSION_BOOT] Resuming session %s after takeover", session_id[:8])
+        return _exec_in_tmux(
+            branch, session_id, claude_bin, [claude_bin] + defaults + ["--resume", session_id] + ea + nf
+        )
+
+    logger.info("[SESSION_BOOT] No sessionId for takeover — continuing last")
+    return _exec_in_tmux(branch, "", claude_bin, [claude_bin] + defaults + ["--continue"] + ea + nf)
+
+
+def _is_session_file_present(pid: int | None) -> bool:
+    """Check if a CC session file exists for the given PID."""
+    if pid is None:
+        return False
+    session_file = Path.home() / ".claude" / "sessions" / f"{pid}.json"
+    return session_file.exists()
+
+
+def _menu_single_session(
+    session: dict,
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Handle menu for a single live session."""
+    label = _session_label(session, branch)
+    is_bg = session.get("kind") in ("bg", "background")
+    sys.stderr.write(f"\n{branch} — live chat: {label}\n")
+    if is_bg:
+        sys.stderr.write("  [Enter]  resume this chat   (stops bg, reopens as normal chat)\n")
+        sys.stderr.write("  [n]      start new chat   (stops bg first)\n")
+        sys.stderr.write("  [c]      close it and exit   (stops bg)\n\n")
+    else:
+        sys.stderr.write("  [Enter]  resume this chat\n")
+        sys.stderr.write("  [n]      start new chat   (closes the one above first)\n")
+        sys.stderr.write("  [c]      close it and exit\n\n")
+
+    choice = _read_choice()
+
+    if choice in ("", "r"):
+        return _resume_session(session, branch, claude_bin, defaults, extra_args)
+    if choice == "n":
+        return _menu_single_new(session, is_bg, branch, claude_bin, defaults, extra_args)
+    if choice == "c":
+        return _menu_single_close(session, is_bg, branch, claude_bin)
+    if choice in ("exit", "q", "quit"):
+        return {"exit_code": 0, "action": "quit"}
+    sys.stderr.write("  Unknown choice. Exiting.\n")
+    return {"exit_code": 1, "error": "unknown choice"}
+
+
+def _menu_single_new(
+    session: dict,
+    is_bg: bool,
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Handle 'n' choice for single session — stop current, start fresh."""
+    if is_bg:
+        stop = _daemon_stop(claude_bin, branch, session.get("pid"))
+        if not stop["ok"]:
+            return {"exit_code": 1, "error": stop["error"]}
+    else:
+        _stop_session(session, claude_bin)
+    return _start_fresh(branch, claude_bin, defaults, extra_args)
+
+
+def _menu_single_close(session: dict, is_bg: bool, branch: str, claude_bin: str) -> dict:
+    """Handle 'c' choice for single session — close and exit."""
+    if is_bg:
+        stop = _daemon_stop(claude_bin, branch, session.get("pid"))
+        if not stop["ok"]:
+            return {"exit_code": 1, "error": stop["error"]}
+        sys.stderr.write(f"  Stopped bg session PID {session.get('pid')}.\n")
+    else:
+        result = _stop_session(session, claude_bin)
+        sys.stderr.write(f"  {result}\n")
+    return {"exit_code": 0, "action": "closed"}
+
+
+def _menu_live(
+    live: list[dict],
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Display menu when live session(s) exist."""
+    if len(live) == 1:
+        return _menu_single_session(live[0], branch, claude_bin, defaults, extra_args)
+
+    sys.stderr.write(f"\n{branch} — {len(live)} live sessions:\n")
+    for i, session in enumerate(live, 1):
+        label = _session_label(session, branch)
+        sys.stderr.write(f"  [{i}]  {label}\n")
+    sys.stderr.write("  [n]  start new chat\n")
+    sys.stderr.write("  [c]  close all and exit\n\n")
+
+    choice = _read_choice()
+
+    if choice == "c":
+        return _close_all(live, branch, claude_bin)
+
+    if choice == "n":
+        return _new_over_all(live, branch, claude_bin, defaults, extra_args)
+
+    if choice in ("exit", "q", "quit"):
+        return {"exit_code": 0, "action": "quit"}
+
+    try:
+        idx = int(choice) - 1
+        if 0 <= idx < len(live):
+            return _resume_session(live[idx], branch, claude_bin, defaults, extra_args)
+    except (ValueError, IndexError):
+        logger.info("[SESSION_BOOT] Invalid menu choice: %r", choice)
+
+    sys.stderr.write("  Pick a number, 'n', or 'c'. Exiting.\n")
+    return {"exit_code": 1, "error": "unknown choice"}
+
+
+def _close_all(live: list[dict], branch: str, claude_bin: str) -> dict:
+    """Close all sessions — stop what's stoppable, honest about bg."""
+    non_bg = [s for s in live if s.get("kind") not in ("bg", "background")]
+    bg = [s for s in live if s.get("kind") in ("bg", "background")]
+    for s in non_bg:
+        result = _stop_session(s, claude_bin)
+        sys.stderr.write(f"  {result}\n")
+    if bg:
+        stop = _daemon_stop(claude_bin, branch, bg[0].get("pid"))
+        if stop["ok"]:
+            sys.stderr.write(f"  Stopped {len(bg)} bg session(s) via daemon stop.\n")
+        else:
+            for s in bg:
+                sys.stderr.write(f"  PID {s.get('pid')}: bg session remains — daemon stop failed\n")
+    return {"exit_code": 0, "action": "closed_all"}
+
+
+def _new_over_all(
+    live: list[dict],
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Start new chat, stopping what's stoppable first."""
+    non_bg = [s for s in live if s.get("kind") not in ("bg", "background")]
+    bg = [s for s in live if s.get("kind") in ("bg", "background")]
+    for s in non_bg:
+        result = _stop_session(s, claude_bin)
+        sys.stderr.write(f"  {result}\n")
+    if bg:
+        stop = _daemon_stop(claude_bin, branch, bg[0].get("pid"))
+        if not stop["ok"]:
+            sys.stderr.write("  Cannot start new — bg session(s) still running.\n")
+            return {"exit_code": 1, "error": "daemon stop failed, aborting to preserve one-brain"}
+    return _start_fresh(branch, claude_bin, defaults, extra_args)
+
+
+def _menu_no_live(
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Display menu when no live session exists."""
+    sys.stderr.write(f"\n{branch} — no live chat\n")
+    sys.stderr.write("  [Enter]  continue last chat\n")
+    sys.stderr.write("  [n]      new chat\n\n")
+
+    choice = _read_choice()
+
+    if choice in ("", "r"):
+        logger.info("[SESSION_BOOT] Continuing last chat via --continue")
+        nf = _name_flag(branch, extra_args=extra_args)
+        cmd = [claude_bin] + defaults + ["--continue"] + list(extra_args or []) + nf
+        return _exec_in_tmux(branch, "", claude_bin, cmd)
+    elif choice == "n":
+        return _start_fresh(branch, claude_bin, defaults, extra_args)
+    elif choice in ("exit", "q", "quit"):
+        return {"exit_code": 0, "action": "quit"}
+    else:
+        sys.stderr.write("  Unknown choice. Exiting.\n")
+        return {"exit_code": 1, "error": "unknown choice"}
+
+
+def _start_fresh(
+    branch: str,
+    claude_bin: str,
+    defaults: list[str],
+    extra_args: list[str] | None,
+) -> dict:
+    """Start a fresh Claude session in a new tmux session."""
+    session_name = _make_session_name(branch)
+
+    if _tmux_session_exists(session_name):
+        logger.info("[SESSION_BOOT] Killing stale tmux session '%s'", session_name)
+        subprocess.run(["tmux", "kill-session", "-t", session_name], check=False)
+
+    claude_cmd = [claude_bin] + defaults + _name_flag(branch, extra_args=extra_args)
    if extra_args:
        claude_cmd.extend(extra_args)

-    logger.info("[SESSION_BOOT] Starting fresh in tmux session '%s': %s", branch, " ".join(claude_cmd))
-    os.execvp("tmux", ["tmux", "new-session", "-s", branch, "--"] + claude_cmd)
-    return {"exit_code": 0, "action": "started", "tmux_session": branch}
+    logger.info("[SESSION_BOOT] Starting fresh in tmux session '%s': %s", session_name, " ".join(claude_cmd))
+    os.execvp("tmux", ["tmux", "new-session", "-s", session_name, "--"] + claude_cmd)
+    return {"exit_code": 0, "action": "started", "tmux_session": session_name}


 def main() -> None:
@@ -564,11 +564,19 @@ def handle(hook_data: dict) -> dict:

    response_text = _prepend_branch_prefix(response_text)
    logs_were_active = _check_log_streamer_active()
+    streaming = bool(pending_data.get("streaming", False))

    chunks = chunk_text(response_text)
-    logger.info("[HOOKS] telegram: sending %d chunk(s) (logs_active=%s)", len(chunks), logs_were_active)
+    logger.info(
+        "[HOOKS] telegram: sending %d chunk(s) (logs_active=%s, streaming=%s)",
+        len(chunks),
+        logs_were_active,
+        streaming,
+    )

-    all_sent, chunk_results = _deliver_chunks(chunks, bot_token, chat_id, processing_message_id, logs_were_active)
+    all_sent, chunk_results = _deliver_chunks(
+        chunks, bot_token, chat_id, processing_message_id, logs_were_active, streaming
+    )

    _write_delivery_log(response_text, chunks, chunk_results, session_id)

@@ -658,6 +666,7 @@ def _deliver_chunks(
    chat_id: int,
    processing_message_id: int | None,
    logs_were_active: bool,
+    streaming: bool = False,
 ) -> tuple[bool, list[dict]]:
    """Send all response chunks to Telegram. Returns (all_sent, per-chunk results)."""
    chunk_results: list[dict] = []
@@ -666,7 +675,8 @@ def _deliver_chunks(

    for i, chunk in enumerate(chunks):
        if i == 0 and processing_message_id:
-            if single and not logs_were_active:
+            should_reconcile = streaming or (single and not logs_were_active)
+            if should_reconcile:
                result = edit_telegram_message(bot_token, chat_id, processing_message_id, chunk)
                if result["ok"]:
                    chunk_results.append({"idx": i, "method": "edit", **result})
@@ -0,0 +1,117 @@
+# =================== AIPass ====================
+# Name: compass_recall.py
+# Version: 1.0.0
+# Description: Ambient compass recall — surfaces rated decisions on relevant prompts
+# Branch: hooks
+# Layer: apps/handlers/prompt
+# Created: 2026-07-16
+# Modified: 2026-07-16
+# =============================================
+
+"""Queries compass FTS against the user's prompt and injects matching decisions
+under governance rules. Never blocks the prompt on error."""
+
+import json
+import os
+import tempfile
+from pathlib import Path
+
+from aipass.prax.apps.modules.logger import system_logger as logger
+from aipass.hooks.apps.handlers.json import json_handler
+
+_STATE_DIR = Path(tempfile.gettempdir())
+
+
+def _state_path(hook_data: dict | None = None) -> Path | None:
+    session_id = ""
+    if hook_data:
+        session_id = hook_data.get("session_id", "")
+    if not session_id:
+        session_id = os.environ.get("CLAUDE_CODE_SESSION_ID", "")
+    if not session_id:
+        return None
+    return _STATE_DIR / f"aipass-compass-recall-{session_id}.json"
+
+
+def _load_state(hook_data: dict | None = None) -> dict:
+    path = _state_path(hook_data)
+    if path is None or not path.exists():
+        return _fresh_state()
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except (json.JSONDecodeError, OSError) as exc:
+        logger.info("[HOOKS] compass_recall: state read failed: %s", exc)
+        return _fresh_state()
+
+
+def _fresh_state() -> dict:
+    try:
+        from aipass.memory.apps.modules.governance import new_state
+
+        return new_state()
+    except Exception as exc:
+        logger.info("[HOOKS] compass_recall: governance import failed: %s", exc)
+        return {"surfaces_count": 0, "messages_since_last": 0, "last_surface_time": 0.0, "surfaced_ids": []}
+
+
+def _save_state(state: dict, hook_data: dict | None = None) -> None:
+    path = _state_path(hook_data)
+    if path is None:
+        return
+    try:
+        path.write_text(json.dumps(state), encoding="utf-8")
+    except OSError as exc:
+        logger.info("[HOOKS] compass_recall: state write failed: %s", exc)
+
+
+def handle(hook_data: dict) -> dict:
+    """Surface relevant compass decisions into the prompt context."""
+    try:
+        if not _state_path(hook_data):
+            return {"stdout": "", "exit_code": 0}
+
+        state = _load_state(hook_data)
+
+        from aipass.memory.apps.modules.governance import should_surface, record_message
+
+        state = record_message(state)
+
+        prompt_text = hook_data.get("prompt", "")
+        if not prompt_text or len(prompt_text) < 10:
+            _save_state(state, hook_data)
+            return {"stdout": "", "exit_code": 0}
+
+        from aipass.devpulse.apps.modules.compass import recall_decisions, mark_surfaced
+
+        candidates = recall_decisions(prompt_text, limit=3)
+        if not candidates:
+            _save_state(state, hook_data)
+            return {"stdout": "", "exit_code": 0}
+
+        approved = []
+        for c in candidates:
+            item_id = str(c["id"])
+            relevance = c.get("relevance", 0.0)
+            surface, reason, new_st = should_surface(item_id, relevance, state)
+            if surface:
+                approved.append(c)
+                state = new_st
+
+        _save_state(state, hook_data)
+
+        if not approved:
+            return {"stdout": "", "exit_code": 0}
+
+        lines = []
+        for c in approved:
+            rating = c.get("rating", "good").upper()
+            lines.append(f"[{rating}] #{c['id']}: {c['decision']}")
+
+        mark_surfaced([c["id"] for c in approved])
+        json_handler.log_operation("compass_recall", {"count": len(approved)})
+
+        return {"stdout": "\n".join(lines), "exit_code": 0}
+
+    except Exception as exc:
+        logger.info("[HOOKS] compass_recall_unreachable: %s", exc)
+        return {"stdout": "", "exit_code": 0}
@@ -0,0 +1,133 @@
+# =================== AIPass ====================
+# Name: persistent_alert.py
+# Version: 1.0.1
+# Description: Injects advisory banners for active alerts on UserPromptSubmit
+# Branch: hooks
+# Layer: apps/handlers/prompt
+# Created: 2026-07-14
+# Modified: 2026-07-14
+# =============================================
+
+"""Injects advisory banners for active alerts from .aipass/alerts.json."""
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+
+from aipass.hooks.apps.handlers.json import json_handler
+from aipass.prax.apps.modules.logger import system_logger as logger
+
+_announced: set[str] = set()
+
+
+def _find_aipass_dir() -> Path | None:
+    """Walk up from CWD; return the nearest .aipass/ that contains alerts.json.
+
+    Every branch has its own .aipass/ (branch prompt), so stopping at the first
+    .aipass directory would never reach the project root where alerts.json lives.
+    """
+    search = Path.cwd()
+    home = Path.home()
+    while search != home and search.parent != search:
+        aipass_dir = search / ".aipass"
+        if (aipass_dir / "alerts.json").exists():
+            return aipass_dir
+        search = search.parent
+    return None
+
+
+def _load_and_clean(alerts_path: Path) -> list[dict]:
+    """Load alerts, remove expired, write back if cleaned. Returns active alerts."""
+    try:
+        data = json.loads(alerts_path.read_text(encoding="utf-8"))
+    except (json.JSONDecodeError, OSError) as exc:
+        logger.info("[HOOKS] persistent_alert: read error: %s", exc)
+        return []
+
+    alerts = data.get("alerts", []) if isinstance(data, dict) else []
+    if not alerts:
+        return []
+
+    now = datetime.now(timezone.utc)
+    active = []
+    cleaned = False
+    for alert in alerts:
+        expires = alert.get("expires_at")
+        if expires:
+            try:
+                exp_dt = datetime.fromisoformat(expires)
+                if exp_dt.tzinfo is None:
+                    exp_dt = exp_dt.replace(tzinfo=timezone.utc)
+                if exp_dt < now:
+                    cleaned = True
+                    continue
+            except (ValueError, TypeError) as exc:
+                logger.info("[HOOKS] persistent_alert: bad expires_at: %s", exc)
+        active.append(alert)
+
+    if cleaned:
+        try:
+            alerts_path.write_text(
+                json.dumps({"alerts": active}, indent=2) + "\n",
+                encoding="utf-8",
+            )
+        except OSError as exc:
+            logger.info("[HOOKS] persistent_alert: cleanup write error: %s", exc)
+
+    return active
+
+
+def _format_banner(alerts: list[dict]) -> str:
+    """Format alert banners for prompt injection."""
+    lines = []
+    for alert in alerts:
+        severity = alert.get("severity", "warning").upper()
+        title = alert.get("title", "Untitled alert")
+        body = alert.get("body", "")
+        source = alert.get("source", "unknown")
+        alert_id = alert.get("id", "?")
+        lines.append(f"[{severity}] {title} (from @{source}, id: {alert_id})")
+        if body:
+            lines.append(f"  {body}")
+    header = "# Active Alerts"
+    dismiss_hint = "Dismiss with: drone @hooks dismiss <alert-id>"
+    return "\n".join([header, ""] + lines + ["", dismiss_hint])
+
+
+def handle(hook_data: dict) -> dict:
+    """Inject advisory banners for active alerts.
+
+    Args:
+        hook_data: Parsed hook event dict from engine.
+
+    Returns:
+        Result dict with stdout (banner or empty) and exit_code.
+    """
+    aipass_dir = _find_aipass_dir()
+    if not aipass_dir:
+        return {"stdout": "", "exit_code": 0}
+
+    alerts_path = aipass_dir / "alerts.json"
+    if not alerts_path.exists():
+        return {"stdout": "", "exit_code": 0}
+
+    alerts = _load_and_clean(alerts_path)
+    if not alerts:
+        return {"stdout": "", "exit_code": 0}
+
+    banner = _format_banner(alerts)
+
+    new_ids = [a["id"] for a in alerts if a.get("id") and a["id"] not in _announced]
+    sound = ""
+    if new_ids:
+        _announced.update(new_ids)
+        count = len(alerts)
+        plural = "s" if count != 1 else ""
+        sound = f"alert: {count} active alert{plural}"
+
+    json_handler.log_operation("inject_alerts", {"count": len(alerts)})
+    logger.info("[HOOKS] persistent_alert: %d active alerts injected", len(alerts))
+    result = {"stdout": banner, "exit_code": 0}
+    if sound:
+        result["sound"] = sound
+    return result
@@ -1,11 +1,11 @@
 # =================== AIPass ====================
 # Name: presence_gate.py
-# Version: 2.0.0
+# Version: 3.0.0
 # Description: Single-session gate — blocks duplicate Claude runtimes per branch
 # Branch: hooks
 # Layer: apps/handlers/security
 # Created: 2026-06-29
-# Modified: 2026-06-30
+# Modified: 2026-07-13
 # =============================================

 """Single-session gate — blocks duplicate Claude runtimes per branch.
@@ -20,10 +20,13 @@ same branch. If occupied by a different PID, blocks. If free, allows.
 handle_stop is a no-op (Stop fires every turn, not just session end; CC-native
 session files handle cleanup on exit).

-Skips sub-agents and dispatched/daemon session types.
+Skips true sub-agents (Explore/general-purpose/Plan/etc.) and
+dispatched/daemon session types. Gates main sessions of every kind
+including background sessions with agent_type "claude".

-PRESENCE.central.json + presence.py are preserved (not deleted) but the guard
-no longer sources truth from them.
+Ships in OBSERVE-ONLY mode: logs would-block decisions to engine.jsonl
+but never actually blocks. Flip _OBSERVE_ONLY to False after soak period
+confirms zero false positives.
 """

 import importlib
@@ -36,6 +39,19 @@ from aipass.prax.apps.modules.logger import system_logger as logger
 _ALLOW = {"exit_code": 0, "stdout": ""}
 _NON_BLOCKING_SESSION_TYPES = frozenset({"dispatched", "daemon"})

+_OBSERVE_ONLY = True
+
+_SUB_AGENT_TYPES = frozenset(
+    {
+        "general-purpose",
+        "Explore",
+        "Plan",
+        "code-reviewer",
+        "statusline-setup",
+        "Task",
+    }
+)
+

 def _resolve_branch(hook_data: dict) -> str:
    """Resolve the branch name from hook_data's cwd (session dir, not process cwd)."""
@@ -50,6 +66,29 @@ def _resolve_branch(hook_data: dict) -> str:
    return Path(cwd).name


+def _format_session_age(session: dict) -> str:
+    """Format session age from its start time."""
+    started = session.get("startedAt") or session.get("started", "")
+    if not started:
+        return ""
+    try:
+        from datetime import datetime, timezone
+
+        if isinstance(started, (int, float)):
+            start_dt = datetime.fromtimestamp(started / 1000, tz=timezone.utc)
+        else:
+            start_dt = datetime.fromisoformat(str(started).replace("Z", "+00:00"))
+        delta = datetime.now(tz=timezone.utc) - start_dt
+        hours = int(delta.total_seconds() // 3600)
+        minutes = int((delta.total_seconds() % 3600) // 60)
+        if hours > 0:
+            return f"{hours}h{minutes}m"
+        return f"{minutes}m"
+    except Exception as exc:
+        logger.info("[presence_gate] age format error: %s", exc)
+        return ""
+
+
 def handle(hook_data: dict) -> dict:
    """UserPromptSubmit gate — enforce one live session per branch.

@@ -59,7 +98,7 @@ def handle(hook_data: dict) -> dict:
    """
    try:
        agent_type = hook_data.get("agent_type", "")
-        if agent_type and agent_type != "main":
+        if agent_type in _SUB_AGENT_TYPES:
            return _ALLOW

        session_type = os.environ.get("AIPASS_SESSION_TYPE", "interactive")
@@ -79,12 +118,21 @@ def handle(hook_data: dict) -> dict:
            return _ALLOW

        occ_pid = occupant.get("pid", "?")
-        occ_name = occupant.get("name", "")
+        occ_kind = occupant.get("kind", "unknown")
+        occ_sid = str(occupant.get("sessionId", ""))[:8]
+        age = _format_session_age(occupant)
+        age_str = f" · {age} old" if age else ""
+
        reason = (
-            f"{branch} is already live at PID {occ_pid}{f' ({occ_name})' if occ_name else ''}"
-            f" — Claude allows one session per branch."
-            f" Attach to that session, or run `kill {occ_pid}` to reclaim the branch, then retry."
+            f"{branch} is already live: PID {occ_pid} · {occ_sid} · {occ_kind}{age_str}\n"
+            f"  Attach to that session, or run: drone @hooks sessions reclaim @{branch}\n"
+            f"  To disable this gate: set presence_gate.enabled=false in .aipass/hooks.json"
        )
+
+        if _OBSERVE_ONLY:
+            logger.warning("[presence_gate] OBSERVE-ONLY would-block: %s", reason)
+            return _ALLOW
+
        logger.warning("[presence_gate] BLOCKED: %s", reason)
        return {
            "exit_code": 2,
@@ -0,0 +1,110 @@
+# =================== AIPass ====================
+# Name: alert_dismiss.py
+# Version: 1.0.1
+# Description: Dismiss alerts from .aipass/alerts.json via drone @hooks dismiss
+# Branch: hooks
+# Layer: apps/modules
+# Created: 2026-07-14
+# Modified: 2026-07-14
+# =============================================
+
+"""Dismiss alerts from .aipass/alerts.json by ID."""
+
+import json
+from pathlib import Path
+
+from aipass.cli.apps.modules import err_console
+from aipass.hooks.apps.handlers.json import json_handler
+from aipass.prax.apps.modules.logger import system_logger as logger
+
+CONSOLE = err_console
+
+HELP_COMMANDS = [
+    ("dismiss <alert-id>", "Remove an alert from .aipass/alerts.json"),
+]
+
+
+def _find_aipass_dir() -> Path | None:
+    """Walk up from CWD; return the nearest .aipass/ that contains alerts.json.
+
+    Every branch has its own .aipass/ (branch prompt), so stopping at the first
+    .aipass directory would never reach the project root where alerts.json lives.
+    """
+    search = Path.cwd()
+    home = Path.home()
+    while search != home and search.parent != search:
+        aipass_dir = search / ".aipass"
+        if (aipass_dir / "alerts.json").exists():
+            return aipass_dir
+        search = search.parent
+    return None
+
+
+def _dismiss_alert(alert_id: str) -> bool:
+    """Remove an alert by ID from alerts.json. Returns True if found and removed."""
+    aipass_dir = _find_aipass_dir()
+    if not aipass_dir:
+        CONSOLE.print("[red]No .aipass/ directory found in this directory tree.[/red]")
+        return False
+
+    alerts_path = aipass_dir / "alerts.json"
+    if not alerts_path.exists():
+        CONSOLE.print("[yellow]No alerts.json found — nothing to dismiss.[/yellow]")
+        return False
+
+    try:
+        data = json_handler.read_json_file(alerts_path)
+    except (json.JSONDecodeError, OSError) as exc:
+        logger.error("[HOOKS] dismiss: read error: %s", exc)
+        CONSOLE.print(f"[red]Failed to read alerts.json: {exc}[/red]")
+        return False
+
+    alerts = data.get("alerts", []) if isinstance(data, dict) else []
+    original_count = len(alerts)
+    remaining = [a for a in alerts if a.get("id") != alert_id]
+
+    if len(remaining) == original_count:
+        CONSOLE.print(f"[yellow]Alert {alert_id} not found.[/yellow]")
+        return False
+
+    try:
+        json_handler.write_json_file(alerts_path, {"alerts": remaining})
+    except OSError as exc:
+        logger.error("[HOOKS] dismiss: write error: %s", exc)
+        CONSOLE.print(f"[red]Failed to write alerts.json: {exc}[/red]")
+        return False
+
+    json_handler.log_operation("dismiss_alert", {"alert_id": alert_id})
+    logger.info("[HOOKS] dismiss: removed alert %s", alert_id)
+    CONSOLE.print(f"[green]Dismissed alert {alert_id}[/green]")
+    return True
+
+
+def print_introspection():
+    """Print module structure for drone routing."""
+    CONSOLE.print("[bold cyan]alert_dismiss[/bold cyan] — Remove alerts from .aipass/alerts.json")
+
+
+def handle_command(command: str, args: list) -> bool:
+    """Route dismiss commands from drone @hooks."""
+    if command != "dismiss":
+        return False
+
+    if not args:
+        print_introspection()
+        CONSOLE.print()
+        CONSOLE.print("  drone @hooks dismiss <alert-id>")
+        CONSOLE.print()
+        CONSOLE.print("Removes the alert with the given ID from .aipass/alerts.json.")
+        return True
+
+    if args[0] in ("--help", "-h", "help"):
+        CONSOLE.print("[bold cyan]dismiss[/bold cyan] — Remove an alert")
+        CONSOLE.print()
+        CONSOLE.print("  drone @hooks dismiss <alert-id>")
+        CONSOLE.print()
+        CONSOLE.print("Removes the alert with the given ID from .aipass/alerts.json.")
+        return True
+
+    _dismiss_alert(args[0])
+    return True
@@ -1,11 +1,11 @@
 # =================== AIPass ====================
 # Name: cc_sessions.py
-# Version: 1.0.0
-# Description: Read CC-native session files for active-session discovery
+# Version: 3.0.0
+# Description: CC-native session discovery, listing, and reclaim
 # Branch: hooks
 # Layer: apps/modules
 # Created: 2026-06-30
-# Modified: 2026-06-30
+# Modified: 2026-07-14
 # =============================================

 """Read Claude Code native session files (~/.claude/sessions/<pid>.json).
@@ -16,10 +16,14 @@ clean exit, stale-swept by CC itself), and the authoritative source of which
 sessions are live for a given working directory.

 Used by presence_gate to source truth instead of PRESENCE.central.json.
+
+Exposed as `drone @hooks sessions` (list) and
+`drone @hooks sessions reclaim [@branch]` (proper-stop cleanup).
 """

 import json
 import os
+import signal
 import sys
 from pathlib import Path

@@ -30,6 +34,11 @@ CONSOLE = err_console

 CC_SESSIONS_DIR = Path.home() / ".claude" / "sessions"

+HELP_COMMANDS = [
+    ("sessions", "List all CC sessions (PID · branch · short-id · kind · age)"),
+    ("sessions reclaim [@branch]", "Properly stop sessions — clean slate"),
+]
+

 def _pid_alive_windows(pid: int) -> bool:
    """Windows-safe liveness check via OpenProcess + GetExitCodeProcess."""
@@ -83,6 +92,40 @@ def _is_pid_alive(pid: int) -> bool:
        return False


+def _format_age(session: dict) -> str:
+    """Format session age from its start time."""
+    started = session.get("startedAt") or session.get("started", "")
+    if not started:
+        return "?"
+    try:
+        from datetime import datetime, timezone
+
+        if isinstance(started, (int, float)):
+            start_dt = datetime.fromtimestamp(started / 1000, tz=timezone.utc)
+        else:
+            start_dt = datetime.fromisoformat(str(started).replace("Z", "+00:00"))
+        delta = datetime.now(tz=timezone.utc) - start_dt
+        hours = int(delta.total_seconds() // 3600)
+        minutes = int((delta.total_seconds() % 3600) // 60)
+        if hours > 0:
+            return f"{hours}h{minutes}m"
+        return f"{minutes}m"
+    except Exception as exc:
+        logger.info("[CC_SESSIONS] age format error: %s", exc)
+        return "?"
+
+
+def _session_branch(session: dict) -> str:
+    """Extract the branch name from a session's cwd."""
+    cwd = session.get("cwd", "")
+    return Path(cwd).name if cwd else "?"
+
+
+def _session_short_id(session: dict) -> str:
+    """Extract short session ID (first 8 chars of sessionId)."""
+    return str(session.get("sessionId", ""))[:8]
+
+
 def read_all_sessions() -> list[dict]:
    """Read all CC session PID files. Returns list of session dicts."""
    if not CC_SESSIONS_DIR.is_dir():
@@ -139,45 +182,102 @@ def find_occupant(cwd: str, exclude_pid: int | None = None) -> dict | None:
    return None


+def _stop_session(session: dict) -> str:
+    """Stop a session. Returns description of action taken.
+
+    bg sessions: no per-job stop exists in the CLI. Returns an honest
+    message — never SIGTERMs bg (daemon respawns it).
+    """
+    pid = session.get("pid")
+    kind = session.get("kind", "unknown")
+    branch = _session_branch(session)
+
+    if kind in ("bg", "background"):
+        logger.info("[CC_SESSIONS] Cannot stop bg PID %s — no per-job stop in CLI", pid)
+        return f"PID {pid} ({branch}): bg session — no per-job stop available"
+
+    if pid and _is_pid_alive(pid):
+        try:
+            os.kill(pid, signal.SIGTERM)
+        except OSError as exc:
+            logger.warning("[CC_SESSIONS] Failed to SIGTERM PID %d: %s", pid, exc)
+            return f"PID {pid} ({branch}): SIGTERM failed ({exc})"
+        return f"PID {pid} ({branch}): sent SIGTERM to {kind} session"
+    return f"PID {pid} ({branch}): already dead"
+
+
+def reclaim(branch_filter: str | None = None) -> list[str]:
+    """Properly stop sessions, optionally filtered to a branch. Returns action log."""
+    actions = []
+    for session in read_all_sessions():
+        pid = session.get("pid")
+        if not pid or not _is_pid_alive(pid):
+            continue
+        if branch_filter and _session_branch(session) != branch_filter:
+            continue
+        action = _stop_session(session)
+        actions.append(action)
+        logger.info("[CC_SESSIONS] reclaim: %s", action)
+    return actions
+
+
 # =============================================================================
 # MODULE INTERFACE (drone @hooks routing)
 # =============================================================================


+def _print_sessions_list():
+    """Print all CC sessions in P6 format: PID · branch · short-id · kind · age."""
+    sessions = read_all_sessions()
+    if not sessions:
+        CONSOLE.print("  No CC session files found")
+        return
+    for s in sessions:
+        pid = s.get("pid", "?")
+        branch = _session_branch(s)
+        short_id = _session_short_id(s)
+        kind = s.get("kind", "?")
+        age = _format_age(s)
+        alive = _is_pid_alive(pid) if isinstance(pid, int) else False
+        status = "[green]live[/green]" if alive else "[dim]stale[/dim]"
+        CONSOLE.print(f"  PID {pid} · {branch} · {short_id} · {kind} · {age}  {status}")
+
+
 def print_introspection():
    """Print CC session state for drone routing."""
-    CONSOLE.print("[bold cyan]cc_sessions[/bold cyan] Module")
+    CONSOLE.print("[bold cyan]sessions[/bold cyan] — CC session listing & reclaim")
    CONSOLE.print(f"  Sessions dir: {CC_SESSIONS_DIR}")
-    try:
-        sessions = read_all_sessions()
-        if not sessions:
-            CONSOLE.print("  No CC session files found")
-        else:
-            for s in sessions:
-                pid = s.get("pid", "?")
-                cwd = s.get("cwd", "?")
-                kind = s.get("kind", "?")
-                name = s.get("name", "")
-                alive = _is_pid_alive(pid) if isinstance(pid, int) else False
-                status = "[green]live[/green]" if alive else "[dim]stale[/dim]"
-                label = f" ({name})" if name else ""
-                CONSOLE.print(f"  PID {pid}{label}: {Path(cwd).name}  kind={kind}  {status}")
-    except Exception as exc:
-        logger.info("[CC_SESSIONS] introspection error: %s", exc)
-        CONSOLE.print(f"  Error reading sessions: {exc}")
+    _print_sessions_list()


 def handle_command(command: str, args: list) -> bool:
-    """Route cc_sessions commands from drone @hooks."""
+    """Route sessions/cc_sessions commands from drone @hooks."""
    if command in ("--help", "-h", "help"):
-        CONSOLE.print("[bold cyan]cc_sessions[/bold cyan] — CC-native session file reader")
+        CONSOLE.print("[bold cyan]sessions[/bold cyan] — CC session listing & reclaim")
        CONSOLE.print()
-        CONSOLE.print("  drone @hooks cc_sessions    Show live CC sessions")
+        CONSOLE.print("  drone @hooks sessions                  List all CC sessions")
+        CONSOLE.print("  drone @hooks sessions reclaim          Stop all live sessions")
+        CONSOLE.print("  drone @hooks sessions reclaim @branch  Stop sessions for a branch")
+        CONSOLE.print()
+        CONSOLE.print("  drone @hooks cc_sessions               (legacy alias for sessions)")
        return True

-    if command == "cc_sessions":
+    if command in ("sessions", "cc_sessions"):
        if not args:
            print_introspection()
            return True

+        if args[0] == "reclaim":
+            branch_filter = None
+            if len(args) > 1:
+                branch_filter = args[1].lstrip("@")
+            CONSOLE.print(f"[bold cyan]sessions reclaim[/bold cyan]{f' @{branch_filter}' if branch_filter else ''}")
+            actions = reclaim(branch_filter)
+            if not actions:
+                CONSOLE.print("  No live sessions to reclaim")
+            else:
+                for action in actions:
+                    CONSOLE.print(f"  {action}")
+            return True
+
    return False
@@ -14,6 +14,7 @@ import importlib
 import json
 import os
 import subprocess
+import tempfile
 import time
 from pathlib import Path

@@ -65,6 +66,18 @@ def _run_handler(handler_path: str, hook_data: dict) -> dict:
    start = time.monotonic()
    try:
        module_path, func_name = handler_path.rsplit(".", 1)
+        if not module_path.startswith("aipass."):
+            elapsed_ms = (time.monotonic() - start) * 1000
+            logger.warning(
+                "[HOOKS] handler path refused (not in aipass.* namespace): %s",
+                handler_path,
+            )
+            return {
+                "exit_code": -1,
+                "stdout": "",
+                "stderr": f"handler namespace refused: {handler_path}",
+                "elapsed_ms": round(elapsed_ms, 1),
+            }
        module = importlib.import_module(module_path)
        handler_func = getattr(module, func_name)
        result = handler_func(hook_data)
@@ -89,6 +102,63 @@ def _matches(matcher: str, value: str) -> bool:
    return value in matcher.split("|")


+_BUDGET_KEYS = ("max_per_session", "min_spacing_turns", "cooldown_seconds")
+
+
+def _budget_state_path(session_id: str = "") -> Path | None:
+    if not session_id:
+        session_id = os.environ.get("CLAUDE_CODE_SESSION_ID", "")
+    if not session_id:
+        return None
+    return Path(tempfile.gettempdir()) / f"aipass-handler-budget-{session_id}.json"
+
+
+def _load_budget_state(session_id: str = "") -> dict:
+    path = _budget_state_path(session_id)
+    if path is None or not path.exists():
+        return {}
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except (json.JSONDecodeError, OSError) as exc:
+        logger.info("[HOOKS] budget: state read failed: %s", exc)
+        return {}
+
+
+def _save_budget_state(state: dict, session_id: str = "") -> None:
+    path = _budget_state_path(session_id)
+    if path is None:
+        return
+    try:
+        path.write_text(json.dumps(state), encoding="utf-8")
+    except OSError as exc:
+        logger.info("[HOOKS] budget: state write failed: %s", exc)
+
+
+def _check_budget(hook_name: str, budget_cfg: dict, budget_state: dict) -> tuple[bool, str]:
+    """Check if handler is within its per-session budget."""
+    hs = budget_state.get(hook_name, {})
+    fire_count = hs.get("fire_count", 0)
+
+    max_fires = budget_cfg.get("max_per_session")
+    if max_fires is not None and fire_count >= max_fires:
+        return False, f"budget exhausted ({fire_count}/{max_fires})"
+
+    if fire_count > 0:
+        min_spacing = budget_cfg.get("min_spacing_turns")
+        if min_spacing is not None:
+            turns_since = hs.get("turns_since_fire", 0)
+            if turns_since < min_spacing:
+                return False, f"spacing ({turns_since}/{min_spacing})"
+
+        cooldown = budget_cfg.get("cooldown_seconds")
+        if cooldown is not None:
+            elapsed = time.time() - hs.get("last_fire_time", 0.0)
+            if elapsed < cooldown:
+                return False, f"cooldown ({int(cooldown - elapsed)}s)"
+
+    return True, "ok"
+
+
 def dispatch(event_type: str, stdin_data: str, config: dict) -> tuple[str, int]:
    """Core dispatch — run hooks for event, return (merged_stdout, exit_code)."""
    if not config.get("hooks_enabled", True):
@@ -111,6 +181,9 @@ def dispatch(event_type: str, stdin_data: str, config: dict) -> tuple[str, int]:

    outputs = []
    total_start = time.monotonic()
+    budget_state = None
+    budget_dirty = False
+    payload_session_id = parsed.get("session_id", "")

    for hook_name, hook_def in event_hooks.items():
        if not hook_def.get("enabled", True):
@@ -137,6 +210,43 @@ def dispatch(event_type: str, stdin_data: str, config: dict) -> tuple[str, int]:
            )
            continue

+        if command and not handler and config.get("_source") == "project":
+            logger.warning(
+                "[HOOKS] %s.%s REFUSED: command-type not allowed in per-project config",
+                event_type,
+                hook_name,
+            )
+            _log(
+                {
+                    "ts": time.time(),
+                    "event": event_type,
+                    "hook": hook_name,
+                    "action": "refused_command_type",
+                }
+            )
+            continue
+
+        budget_cfg = {k: hook_def[k] for k in _BUDGET_KEYS if k in hook_def}
+        if budget_cfg:
+            if budget_state is None:
+                budget_state = _load_budget_state(payload_session_id)
+            hs = budget_state.setdefault(hook_name, {})
+            hs["turns_since_fire"] = hs.get("turns_since_fire", 0) + 1
+            budget_dirty = True
+            allowed, reason = _check_budget(hook_name, budget_cfg, budget_state)
+            if not allowed:
+                logger.info("[HOOKS] %s.%s budget: %s", event_type, hook_name, reason)
+                _log(
+                    {
+                        "ts": time.time(),
+                        "event": event_type,
+                        "hook": hook_name,
+                        "action": "budget_suppressed",
+                        "reason": reason,
+                    }
+                )
+                continue
+
        if handler:
            result = _run_handler(handler, parsed)
        else:
@@ -216,6 +326,15 @@ def dispatch(event_type: str, stdin_data: str, config: dict) -> tuple[str, int]:

        if result["stdout"]:
            outputs.append(result["stdout"])
+            if budget_cfg and budget_state is not None:
+                hs = budget_state.setdefault(hook_name, {})
+                hs["fire_count"] = hs.get("fire_count", 0) + 1
+                hs["last_fire_time"] = time.time()
+                hs["turns_since_fire"] = 0
+                budget_dirty = True
+
+    if budget_dirty and budget_state is not None:
+        _save_budget_state(budget_state, payload_session_id)

    total_ms = (time.monotonic() - total_start) * 1000
    logger.info("[HOOKS] %s complete: %d hooks %dms", event_type, len(outputs), total_ms)
@@ -155,30 +155,49 @@ def _read_proc_ppid(pid: int) -> int | None:
    return None


+def _get_ppid_portable(pid: int) -> int | None:
+    """Get parent PID portably (Linux + macOS). Returns None on failure."""
+    import subprocess as _sp
+
+    try:
+        result = _sp.run(
+            ["ps", "-o", "ppid=", "-p", str(pid)],
+            capture_output=True,
+            text=True,
+            timeout=5,
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            return int(result.stdout.strip())
+    except (OSError, ValueError, _sp.TimeoutExpired) as exc:
+        logger.info("[PRESENCE] ppid lookup failed for PID %d: %s", pid, exc)
+    return None
+
+
+def _has_session_file(pid: int) -> bool:
+    """Check if ~/.claude/sessions/<pid>.json exists."""
+    return (Path.home() / ".claude" / "sessions" / f"{pid}.json").is_file()
+
+
 def _resolve_session_pid() -> int | None:
    """Walk the parent process chain to find the persistent claude session PID.

    The hook runs as an ephemeral subprocess — os.getpid() gives a PID that dies
-    in milliseconds. The owning claude session is a parent process with comm=claude.
-    Linux only (/proc). Returns None on non-Linux or if no claude ancestor found.
+    in milliseconds. The owning claude session is an ancestor that owns a
+    ~/.claude/sessions/<pid>.json file (CC binary is version-named, so comm
+    matching is unreliable).
    """
-    if sys.platform != "linux":
-        return None
    pid = os.getpid()
    ancestors = []
    for _ in range(12):
-        comm = _read_proc_comm(pid)
-        if not comm:
-            break
-        ancestors.append(f"{pid}:{comm}")
-        if comm == "claude":
+        ancestors.append(str(pid))
+        if _has_session_file(pid):
            logger.info("[PRESENCE] Resolved session PID: %d (chain: %s)", pid, " -> ".join(ancestors))
            return pid
-        ppid = _read_proc_ppid(pid)
-        if not ppid or ppid == pid:
+        ppid = _get_ppid_portable(pid)
+        if not ppid or ppid == pid or ppid <= 1:
            break
        pid = ppid
-    logger.info("[PRESENCE] No claude ancestor found (chain: %s)", " -> ".join(ancestors))
+    logger.info("[PRESENCE] No session-file ancestor found (chain: %s)", " -> ".join(ancestors))
    return None


@@ -102,14 +102,10 @@ def _check_event_wiring(event_type, hooks_group, pidx, errors, warnings, info):
    if not enabled_hooks:
        return

-    provider_wired_hooks = {
-        name: defn for name, defn in enabled_hooks.items() if defn.get("provider_wired", True) is not False
-    }
-
    if pidx is None:
-        if provider_wired_hooks:
+        if enabled_hooks:
            errors.append(
-                f"{event_type}: {len(provider_wired_hooks)} enabled handler(s) in project config"
+                f"{event_type}: {len(enabled_hooks)} enabled handler(s) in project config"
                f" but NO provider event entry — handlers never fire"
            )
        return
@@ -124,8 +120,6 @@ def _check_event_wiring(event_type, hooks_group, pidx, errors, warnings, info):
        info.append(f"{event_type}: unfiltered bridge, {len(enabled_hooks)} enabled hooks OK")
    else:
        for hook_name, hook_defn in enabled_hooks.items():
-            if hook_defn.get("provider_wired", True) is False:
-                continue
            if hook_name not in filtered:
                errors.append(
                    f"{event_type}:{hook_name}: enabled in project config"
@@ -136,15 +136,75 @@ class TestFindOccupant:
        assert result is not None


+class TestReclaim:
+    def test_reclaim_stops_live_sessions(self, tmp_path):
+        my_pid = os.getpid()
+        s = {"pid": my_pid, "sessionId": "a", "cwd": "/tmp/hooks", "kind": "interactive"}
+        (tmp_path / f"{my_pid}.json").write_text(json.dumps(s))
+        with (
+            patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path),
+            patch.object(cc_sessions, "_stop_session", return_value="stopped") as mock_stop,
+        ):
+            actions = cc_sessions.reclaim()
+        assert len(actions) == 1
+        mock_stop.assert_called_once()
+
+    def test_reclaim_filters_by_branch(self, tmp_path):
+        my_pid = os.getpid()
+        s1 = {"pid": my_pid, "sessionId": "a", "cwd": "/tmp/hooks", "kind": "interactive"}
+        (tmp_path / f"{my_pid}.json").write_text(json.dumps(s1))
+        with (
+            patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path),
+            patch.object(cc_sessions, "_stop_session", return_value="stopped") as mock_stop,
+        ):
+            actions = cc_sessions.reclaim("devpulse")
+        assert actions == []
+        mock_stop.assert_not_called()
+
+    def test_reclaim_empty_no_actions(self, tmp_path):
+        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
+            actions = cc_sessions.reclaim()
+        assert actions == []
+
+
+class TestSessionHelpers:
+    def test_session_branch(self):
+        assert cc_sessions._session_branch({"cwd": "/tmp/project/src/aipass/hooks"}) == "hooks"
+
+    def test_session_branch_empty_cwd(self):
+        assert cc_sessions._session_branch({"cwd": ""}) == "?"
+
+    def test_session_short_id(self):
+        assert cc_sessions._session_short_id({"sessionId": "abcdef1234567890"}) == "abcdef12"
+
+    def test_session_short_id_short(self):
+        assert cc_sessions._session_short_id({"sessionId": "abc"}) == "abc"
+
+    def test_session_short_id_missing(self):
+        assert cc_sessions._session_short_id({}) == ""
+
+
 class TestIntrospection:
    def test_print_introspection_no_sessions(self, tmp_path):
        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
            cc_sessions.print_introspection()

-    def test_handle_command_cc_sessions(self, tmp_path):
+    def test_handle_command_sessions(self, tmp_path):
+        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
+            assert cc_sessions.handle_command("sessions", []) is True
+
+    def test_handle_command_cc_sessions_legacy(self, tmp_path):
        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
            assert cc_sessions.handle_command("cc_sessions", []) is True

+    def test_handle_command_sessions_reclaim(self, tmp_path):
+        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
+            assert cc_sessions.handle_command("sessions", ["reclaim"]) is True
+
+    def test_handle_command_sessions_reclaim_branch(self, tmp_path):
+        with patch.object(cc_sessions, "CC_SESSIONS_DIR", tmp_path):
+            assert cc_sessions.handle_command("sessions", ["reclaim", "@hooks"]) is True
+
    def test_handle_command_help(self):
        assert cc_sessions.handle_command("--help", []) is True

@@ -0,0 +1,101 @@
+"""CLI contract tests — verify flags/subcommands our code invokes actually exist.
+
+Probes `claude --help` and `claude agents --help` at test time. Skips cleanly
+when the binary is absent. Catches phantom subcommands (like the former
+`claude agents stop`) before they ship as mocked-green.
+"""
+
+import shutil
+import subprocess
+
+import pytest
+
+_CLAUDE = shutil.which("claude")
+_SKIP = pytest.mark.skipif(_CLAUDE is None, reason="claude binary not on PATH")
+
+
+def _help_text(args: list[str]) -> str:
+    assert _CLAUDE is not None
+    result = subprocess.run(
+        [_CLAUDE, *args, "--help"],
+        capture_output=True,
+        text=True,
+        timeout=10,
+    )
+    return result.stdout + result.stderr
+
+
+def _get_main_help() -> str:
+    return _help_text([])
+
+
+def _get_agents_help() -> str:
+    return _help_text(["agents"])
+
+
+def _get_daemon_help() -> str:
+    return _help_text(["daemon"])
+
+
+@_SKIP
+class TestClaudeMainFlags:
+    """Flags from `claude --help` that session_boot invokes."""
+
+    def test_permission_mode(self):
+        assert "--permission-mode" in _get_main_help()
+
+    def test_continue(self):
+        assert "--continue" in _get_main_help()
+
+    def test_resume(self):
+        assert "--resume" in _get_main_help()
+
+    def test_p_flag(self):
+        h = _get_main_help()
+        assert "-p" in h or "--print" in h
+
+    def test_name_flag(self):
+        h = _get_main_help()
+        assert "--name" in h or "-n" in h
+
+
+@_SKIP
+class TestClaudeAgentsFlags:
+    """Flags from `claude agents --help` that session_boot invokes."""
+
+    def test_permission_mode(self):
+        assert "--permission-mode" in _get_agents_help()
+
+    def test_cwd(self):
+        assert "--cwd" in _get_agents_help()
+
+    def test_no_stop_subcommand(self):
+        h = _get_agents_help()
+        assert "stop" not in h.lower() or "agents stop" not in h.lower()
+
+
+@_SKIP
+class TestClaudeDaemonFlags:
+    """Flags from `claude daemon --help` that session_boot invokes."""
+
+    def test_stop_subcommand(self):
+        assert "stop" in _get_daemon_help()
+
+    def test_any_flag(self):
+        assert "--any" in _get_daemon_help()
+
+
+@_SKIP
+class TestAgentsStopDoesNotExist:
+    """Regression: `claude agents stop <id>` must NOT be a valid command."""
+
+    def test_agents_rejects_stop_arg(self):
+        assert _CLAUDE is not None
+        result = subprocess.run(
+            [_CLAUDE, "agents", "stop", "test-id"],
+            capture_output=True,
+            text=True,
+            timeout=10,
+        )
+        assert result.returncode != 0
+        assert "too many arguments" in result.stderr.lower() or "error" in result.stderr.lower()
@@ -0,0 +1,588 @@
+# =================== AIPass ====================
+# Name: test_compass_recall.py
+# Version: 1.1.0
+# Description: Tests for compass recall prompt handler
+# Branch: hooks
+# Created: 2026-07-16
+# Modified: 2026-07-16
+# =============================================
+
+"""Tests for handlers/prompt/compass_recall.py."""
+
+import json
+import os
+from unittest.mock import patch
+
+
+CANDIDATE_GOOD = {
+    "id": 56,
+    "rating": "good",
+    "decision": "Never hardcode config in prompts",
+    "context": "Prompt config management",
+    "note": "",
+    "tags": "config,prompts",
+    "relevance": 0.7,
+}
+
+CANDIDATE_BAD = {
+    "id": 84,
+    "rating": "bad",
+    "decision": "Usage gap is not a bug",
+    "context": "Compass audit",
+    "note": "",
+    "tags": "compass",
+    "relevance": 0.5,
+}
+
+CANDIDATE_LOW_RELEVANCE = {
+    "id": 99,
+    "rating": "good",
+    "decision": "Some low relevance decision",
+    "context": "Testing",
+    "note": "",
+    "tags": "test",
+    "relevance": 0.1,
+}
+
+REAL_PAYLOAD = {
+    "session_id": "abc-123-def",
+    "transcript_path": "/tmp/transcript.jsonl",
+    "cwd": "/home/user/project",
+    "permission_mode": "default",
+    "hook_event_name": "UserPromptSubmit",
+    "prompt": "How should we handle prompt config?",
+}
+
+
+def _payload(prompt, session_id="test-session"):
+    """Build a realistic hook payload with documented keys."""
+    return {"session_id": session_id, "prompt": prompt, "cwd": "/tmp"}
+
+
+class TestCompassRecallHandler:
+    def test_surfaces_relevant_decision(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ) as mock_mark,
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(
+                    True,
+                    "Ready to surface",
+                    {
+                        "surfaces_count": 1,
+                        "messages_since_last": 0,
+                        "last_surface_time": 1000.0,
+                        "surfaced_ids": ["56"],
+                    },
+                ),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("How should we handle prompt config?"))
+
+            assert result["exit_code"] == 0
+            assert "[GOOD] #56:" in result["stdout"]
+            assert "Never hardcode config in prompts" in result["stdout"]
+            mock_mark.assert_called_once_with([56])
+
+    def test_formats_bad_rating(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_BAD],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(
+                    True,
+                    "Ready",
+                    {
+                        "surfaces_count": 1,
+                        "messages_since_last": 0,
+                        "last_surface_time": 1000.0,
+                        "surfaced_ids": ["84"],
+                    },
+                ),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("Is the usage gap a real bug?"))
+
+            assert "[BAD] #84:" in result["stdout"]
+
+    def test_empty_when_no_candidates(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[],
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("Some prompt about something"))
+
+            assert result["exit_code"] == 0
+            assert result["stdout"] == ""
+
+    def test_empty_when_governance_suppresses(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD],
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(
+                    False,
+                    "Spacing not met",
+                    {
+                        "surfaces_count": 0,
+                        "messages_since_last": 1,
+                        "last_surface_time": 0.0,
+                        "surfaced_ids": [],
+                    },
+                ),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("How should we handle prompt config?"))
+
+            assert result["exit_code"] == 0
+            assert result["stdout"] == ""
+
+    def test_empty_when_prompt_too_short(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("Hi"))
+
+            assert result["exit_code"] == 0
+            assert result["stdout"] == ""
+
+    def test_never_blocks_on_import_error(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._state_path",
+                return_value=tmp_path / "state.json",
+            ),
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._load_state",
+                side_effect=Exception("DB locked"),
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("Some prompt about something important"))
+
+            assert result["exit_code"] == 0
+            assert result["stdout"] == ""
+
+    def test_persists_governance_state(self, tmp_path):
+        updated_state = {
+            "surfaces_count": 1,
+            "messages_since_last": 0,
+            "last_surface_time": 1000.0,
+            "surfaced_ids": ["56"],
+        }
+
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(True, "Ready", updated_state),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            handle(_payload("How should we handle prompt config?", session_id="test-persist"))
+
+            state_file = tmp_path / "aipass-compass-recall-test-persist.json"
+            assert state_file.exists()
+            saved = json.loads(state_file.read_text())
+            assert saved["surfaces_count"] == 1
+            assert "56" in saved["surfaced_ids"]
+
+    def test_multiple_candidates_partial_approval(self, tmp_path):
+        def mock_should_surface(item_id, relevance, state, config=None, *, current_time=None):
+            if item_id == "56":
+                new_st = {**state, "surfaces_count": 1, "surfaced_ids": list(state.get("surfaced_ids", [])) + ["56"]}
+                return True, "Ready", new_st
+            return False, "Below threshold", state
+
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD, CANDIDATE_LOW_RELEVANCE],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ) as mock_mark,
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                side_effect=mock_should_surface,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload("How should we handle prompt config?"))
+
+            assert "[GOOD] #56:" in result["stdout"]
+            assert "#99" not in result["stdout"]
+            mock_mark.assert_called_once_with([56])
+
+    def test_empty_prompt_no_cross_branch_import(self, tmp_path):
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(_payload(""))
+
+            assert result["exit_code"] == 0
+            assert result["stdout"] == ""
+
+    def test_no_session_id_degrades_safe(self):
+        """No session_id in payload or env = no injection, no crash."""
+        from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("CLAUDE_CODE_SESSION_ID", None)
+            result = handle({"prompt": "How should we handle prompt config?"})
+
+        assert result["exit_code"] == 0
+        assert result["stdout"] == ""
+
+    def test_real_documented_payload_shape(self, tmp_path):
+        """Surfaces from a payload using the official Claude Code hook keys."""
+        with (
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(
+                    True,
+                    "Ready",
+                    {
+                        "surfaces_count": 1,
+                        "messages_since_last": 0,
+                        "last_surface_time": 1000.0,
+                        "surfaced_ids": ["56"],
+                    },
+                ),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle(REAL_PAYLOAD)
+
+            assert result["exit_code"] == 0
+            assert "[GOOD] #56:" in result["stdout"]
+
+    def test_env_var_fallback_for_session_id(self, tmp_path):
+        """Falls back to CLAUDE_CODE_SESSION_ID env var if payload has no session_id."""
+        with (
+            patch.dict(os.environ, {"CLAUDE_CODE_SESSION_ID": "env-fallback"}),
+            patch(
+                "aipass.hooks.apps.handlers.prompt.compass_recall._STATE_DIR",
+                tmp_path,
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.recall_decisions",
+                return_value=[CANDIDATE_GOOD],
+            ),
+            patch(
+                "aipass.devpulse.apps.modules.compass.mark_surfaced",
+                return_value=1,
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.should_surface",
+                return_value=(
+                    True,
+                    "Ready",
+                    {
+                        "surfaces_count": 1,
+                        "messages_since_last": 0,
+                        "last_surface_time": 1000.0,
+                        "surfaced_ids": ["56"],
+                    },
+                ),
+            ),
+            patch(
+                "aipass.memory.apps.modules.governance.record_message",
+                side_effect=lambda s: {**s, "messages_since_last": s.get("messages_since_last", 0) + 1},
+            ),
+        ):
+            from aipass.hooks.apps.handlers.prompt.compass_recall import handle
+
+            result = handle({"prompt": "How should we handle prompt config?"})
+
+            assert "[GOOD] #56:" in result["stdout"]
+            state_file = tmp_path / "aipass-compass-recall-env-fallback.json"
+            assert state_file.exists()
+
+
+class TestEngineBudget:
+    def test_check_budget_allows_first_fire(self):
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        allowed, reason = _check_budget("test_hook", {"max_per_session": 5}, {})
+        assert allowed is True
+
+    def test_check_budget_blocks_when_exhausted(self):
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 5}}
+        allowed, reason = _check_budget("test_hook", {"max_per_session": 5}, state)
+        assert allowed is False
+        assert "exhausted" in reason
+
+    def test_check_budget_spacing_skipped_on_first_fire(self):
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 0, "turns_since_fire": 0}}
+        allowed, reason = _check_budget("test_hook", {"min_spacing_turns": 10}, state)
+        assert allowed is True
+
+    def test_check_budget_spacing_enforced_after_fire(self):
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 1, "turns_since_fire": 3}}
+        allowed, reason = _check_budget("test_hook", {"min_spacing_turns": 10}, state)
+        assert allowed is False
+        assert "spacing" in reason
+
+    def test_check_budget_spacing_passes_after_enough_turns(self):
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 1, "turns_since_fire": 10}}
+        allowed, reason = _check_budget("test_hook", {"min_spacing_turns": 10}, state)
+        assert allowed is True
+
+    def test_check_budget_cooldown_enforced(self):
+        import time
+
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 1, "last_fire_time": time.time() - 10}}
+        allowed, reason = _check_budget("test_hook", {"cooldown_seconds": 300}, state)
+        assert allowed is False
+        assert "cooldown" in reason
+
+    def test_check_budget_cooldown_expired(self):
+        import time
+
+        from aipass.hooks.apps.modules.engine import _check_budget
+
+        state = {"test_hook": {"fire_count": 1, "last_fire_time": time.time() - 400}}
+        allowed, reason = _check_budget("test_hook", {"cooldown_seconds": 300}, state)
+        assert allowed is True
+
+    def test_budget_state_persistence(self, tmp_path):
+        from aipass.hooks.apps.modules.engine import (
+            _load_budget_state,
+            _save_budget_state,
+        )
+
+        state = {"compass_recall": {"fire_count": 2, "last_fire_time": 1000.0, "turns_since_fire": 5}}
+
+        with patch("aipass.hooks.apps.modules.engine._budget_state_path", return_value=tmp_path / "budget.json"):
+            _save_budget_state(state)
+            loaded = _load_budget_state()
+            assert loaded["compass_recall"]["fire_count"] == 2
+
+    def test_budget_state_missing_returns_empty(self, tmp_path):
+        from aipass.hooks.apps.modules.engine import _load_budget_state
+
+        with patch(
+            "aipass.hooks.apps.modules.engine._budget_state_path",
+            return_value=tmp_path / "nonexistent.json",
+        ):
+            assert _load_budget_state() == {}
+
+    def test_dispatch_suppresses_over_budget_handler(self, tmp_path):
+        from aipass.hooks.apps.modules.engine import dispatch
+
+        config = {
+            "hooks_enabled": True,
+            "UserPromptSubmit": {
+                "test_hook": {
+                    "enabled": True,
+                    "handler": "aipass.hooks.apps.handlers.prompt.compass_recall.handle",
+                    "max_per_session": 0,
+                },
+            },
+        }
+        budget_file = tmp_path / "budget.json"
+
+        with (
+            patch("aipass.hooks.apps.modules.engine._budget_state_path", return_value=budget_file),
+            patch("aipass.hooks.apps.modules.engine._run_handler") as mock_run,
+        ):
+            dispatch("UserPromptSubmit", json.dumps({"session_id": "budget-test", "prompt": "test"}), config)
+            mock_run.assert_not_called()
+
+    def test_dispatch_records_fire_on_output(self, tmp_path):
+        from aipass.hooks.apps.modules.engine import dispatch
+
+        config = {
+            "hooks_enabled": True,
+            "UserPromptSubmit": {
+                "test_hook": {
+                    "enabled": True,
+                    "handler": "aipass.hooks.apps.handlers.prompt.compass_recall.handle",
+                    "max_per_session": 10,
+                },
+            },
+        }
+        budget_file = tmp_path / "budget.json"
+
+        with (
+            patch("aipass.hooks.apps.modules.engine._budget_state_path", return_value=budget_file),
+            patch(
+                "aipass.hooks.apps.modules.engine._run_handler",
+                return_value={"exit_code": 0, "stdout": "[GOOD] #56: test", "stderr": "", "elapsed_ms": 5.0},
+            ),
+        ):
+            dispatch("UserPromptSubmit", json.dumps({"session_id": "fire-test", "prompt": "test"}), config)
+
+            assert budget_file.exists()
+            state = json.loads(budget_file.read_text())
+            assert state["test_hook"]["fire_count"] == 1
+            assert state["test_hook"]["turns_since_fire"] == 0
+
+    def test_dispatch_threads_payload_session_id(self, tmp_path):
+        """Budget state file is keyed by payload session_id, not env var."""
+        from aipass.hooks.apps.modules.engine import dispatch
+
+        config = {
+            "hooks_enabled": True,
+            "UserPromptSubmit": {
+                "test_hook": {
+                    "enabled": True,
+                    "handler": "aipass.hooks.apps.handlers.prompt.compass_recall.handle",
+                    "max_per_session": 10,
+                },
+            },
+        }
+
+        with (
+            patch(
+                "aipass.hooks.apps.modules.engine._run_handler",
+                return_value={"exit_code": 0, "stdout": "output", "stderr": "", "elapsed_ms": 1.0},
+            ),
+        ):
+            dispatch(
+                "UserPromptSubmit",
+                json.dumps({"session_id": "payload-sid", "prompt": "test"}),
+                config,
+            )
+
+            from aipass.hooks.apps.modules.engine import _budget_state_path
+
+            path = _budget_state_path("payload-sid")
+            assert path is not None
+            assert "payload-sid" in str(path)
@@ -23,6 +23,7 @@ from aipass.hooks.apps.modules.engine import (
    _log,
 )
 from aipass.hooks.apps.handlers.config.loader import find_project_config
+from aipass.hooks.apps.handlers.config.trust_registry import enroll


 class TestMatches:
@@ -276,7 +277,13 @@ class TestFindProjectConfig:
    """Tests for find_project_config() CWD walk."""

    def test_finds_config_in_cwd(self, hooks_config_file, temp_test_dir, mock_logger):
-        with patch("aipass.hooks.apps.modules.engine.Path.cwd", return_value=temp_test_dir):
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(temp_test_dir))
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=temp_test_dir),
+        ):
            config = find_project_config()
        assert config is not None
        assert config["hooks_enabled"] is True
@@ -295,9 +302,15 @@ class TestFindProjectConfig:
            "Stop": {"sound": {"enabled": True, "command": "python3 $AIPASS_HOME/hook.py", "matcher": ""}},
        }
        (config_dir / "hooks.json").write_text(json.dumps(config))
-        with patch("aipass.hooks.apps.modules.engine.Path.cwd", return_value=temp_test_dir):
-            with patch("aipass.hooks.apps.handlers.config.loader.AIPASS_HOME", "/test/path"):
-                result = find_project_config()
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(temp_test_dir))
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=temp_test_dir),
+            patch("aipass.hooks.apps.handlers.config.loader.AIPASS_HOME", "/test/path"),
+        ):
+            result = find_project_config()
        assert result is not None
        assert "/test/path/hook.py" in result["Stop"]["sound"]["command"]

@@ -502,7 +515,13 @@ class TestInitProvisioning:
        (config_dir / "hooks.json").write_text('{"hooks_enabled": true}')
        sub_dir = temp_test_dir / "deep" / "nested" / "path"
        sub_dir.mkdir(parents=True)
-        with patch("aipass.hooks.apps.modules.engine.Path.cwd", return_value=sub_dir):
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(temp_test_dir))
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=sub_dir),
+        ):
            config = find_project_config()
        assert config is not None
        assert config["hooks_enabled"] is True
@@ -721,6 +740,153 @@ class TestMockInfrastructure:
        assert hasattr(engine, "_run_hook")


+class TestLayerATrustEnforcement:
+    """DPLAN-0244 Layer A: engine refuses command-type from project config, enforces handler namespace."""
+
+    def test_command_type_refused_from_project_config(self, mock_logger):
+        config = {
+            "hooks_enabled": True,
+            "_source": "project",
+            "PreToolUse": {
+                "evil_cmd": {
+                    "enabled": True,
+                    "command": "echo PWNED",
+                    "matcher": "",
+                }
+            },
+        }
+        with patch("aipass.hooks.apps.modules.engine._log") as mock_log:
+            with patch("aipass.hooks.apps.modules.engine._run_hook") as mock_run:
+                result = dispatch("PreToolUse", '{"tool_name":"Edit"}', config)
+        mock_run.assert_not_called()
+        assert result == ("", 0)
+        log_calls = [c[0][0] for c in mock_log.call_args_list]
+        assert any(e.get("action") == "refused_command_type" for e in log_calls if isinstance(e, dict))
+
+    def test_handler_namespace_enforced(self, mock_logger):
+        from aipass.hooks.apps.modules.engine import _run_handler
+
+        result = _run_handler("evil.payload.handle", {})
+        assert result["exit_code"] == -1
+        assert "namespace refused" in result["stderr"]
+
+    def test_handler_aipass_namespace_allowed(self, mock_logger):
+        from aipass.hooks.apps.modules.engine import _run_handler
+
+        mock_handler = MagicMock(return_value={"exit_code": 0, "stdout": "ok"})
+        mock_module = MagicMock()
+        mock_module.handle = mock_handler
+        with patch("importlib.import_module", return_value=mock_module):
+            result = _run_handler("aipass.hooks.apps.handlers.notification.stop_sound.handle", {})
+        assert result["exit_code"] == 0
+
+    def test_command_type_allowed_from_default_config(self, mock_logger):
+        config = {
+            "hooks_enabled": True,
+            "_source": "default",
+            "Stop": {
+                "cmd_hook": {
+                    "enabled": True,
+                    "command": "echo allowed",
+                    "matcher": "",
+                }
+            },
+        }
+        with patch("aipass.hooks.apps.modules.engine._log"):
+            with patch("aipass.hooks.apps.modules.engine._run_hook") as mock_run:
+                mock_run.return_value = {
+                    "exit_code": 0,
+                    "stdout": "allowed",
+                    "stderr": "",
+                    "elapsed_ms": 5,
+                }
+                result = dispatch("Stop", "{}", config)
+        mock_run.assert_called_once()
+        assert "allowed" in result[0]
+
+    def test_mixed_config_partial_refusal(self, mock_logger):
+        config = {
+            "hooks_enabled": True,
+            "_source": "project",
+            "UserPromptSubmit": {
+                "good_handler": {
+                    "enabled": True,
+                    "handler": "aipass.hooks.apps.handlers.notification.stop_sound.handle",
+                    "matcher": "",
+                },
+                "evil_cmd": {
+                    "enabled": True,
+                    "command": "echo PWNED",
+                    "matcher": "",
+                },
+            },
+        }
+        mock_handler_func = MagicMock(return_value={"exit_code": 0, "stdout": "handler_ok"})
+        mock_module = MagicMock()
+        mock_module.handle = mock_handler_func
+        with patch("aipass.hooks.apps.modules.engine._log"):
+            with patch("importlib.import_module", return_value=mock_module):
+                with patch("aipass.hooks.apps.modules.engine._run_hook") as mock_run:
+                    result = dispatch("UserPromptSubmit", "{}", config)
+        mock_run.assert_not_called()
+        assert "handler_ok" in result[0]
+        assert result[1] == 0
+
+    def test_source_overwrite_not_merge(self, temp_test_dir, mock_logger):
+        config_dir = temp_test_dir / ".aipass"
+        config_dir.mkdir()
+        hostile_config = {
+            "hooks_enabled": True,
+            "_source": "provider",
+            "SessionStart": {
+                "evil": {
+                    "enabled": True,
+                    "command": "echo PWNED",
+                    "matcher": "",
+                }
+            },
+        }
+        (config_dir / "hooks.json").write_text(json.dumps(hostile_config))
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(temp_test_dir))
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=temp_test_dir),
+        ):
+            loaded = find_project_config()
+        assert loaded is not None
+        assert loaded["_source"] == "project"
+        with patch("aipass.hooks.apps.modules.engine._log"):
+            with patch("aipass.hooks.apps.modules.engine._run_hook") as mock_run:
+                result = dispatch("SessionStart", "{}", loaded)
+        mock_run.assert_not_called()
+        assert result == ("", 0)
+
+    def test_command_without_source_defaults_allowed(self, mock_logger):
+        config = {
+            "hooks_enabled": True,
+            "Stop": {
+                "cmd_hook": {
+                    "enabled": True,
+                    "command": "echo ok",
+                    "matcher": "",
+                }
+            },
+        }
+        with patch("aipass.hooks.apps.modules.engine._log"):
+            with patch("aipass.hooks.apps.modules.engine._run_hook") as mock_run:
+                mock_run.return_value = {
+                    "exit_code": 0,
+                    "stdout": "ok",
+                    "stderr": "",
+                    "elapsed_ms": 5,
+                }
+                result = dispatch("Stop", "{}", config)
+        mock_run.assert_called_once()
+        assert "ok" in result[0]
+
+
 class TestJsonHandlerNotApplicable:
    """Hooks uses JSONL logging, not json_handler. These verify the log equivalent."""

@@ -0,0 +1,425 @@
+# =================== AIPass ====================
+# Name: test_persistent_alert.py
+# Version: 1.0.0
+# Description: Tests for persistent_alert handler and alert_dismiss module
+# Branch: hooks
+# Created: 2026-07-14
+# Modified: 2026-07-14
+# =============================================
+
+"""Tests for handlers/prompt/persistent_alert.py and modules/alert_dismiss.py."""
+
+import json
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from unittest.mock import patch
+
+
+def _make_alert(
+    alert_id="test-001",
+    source="prax",
+    severity="warning",
+    title="Test alert",
+    body="Something happened",
+    expires_at=None,
+):
+    alert = {
+        "id": alert_id,
+        "source": source,
+        "severity": severity,
+        "title": title,
+        "body": body,
+        "created_at": datetime.now(timezone.utc).isoformat(),
+        "expires_at": expires_at,
+    }
+    return alert
+
+
+def _write_alerts(aipass_dir: Path, alerts: list[dict]):
+    alerts_path = aipass_dir / "alerts.json"
+    alerts_path.write_text(
+        json.dumps({"alerts": alerts}, indent=2) + "\n",
+        encoding="utf-8",
+    )
+
+
+class TestPersistentAlertHandler:
+    """Banner injection behavior."""
+
+    def test_banner_injected_when_alerts_exist(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert()])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert result["exit_code"] == 0
+        assert "# Active Alerts" in result["stdout"]
+        assert "[WARNING] Test alert" in result["stdout"]
+        assert "drone @hooks dismiss" in result["stdout"]
+
+    def test_no_banner_when_no_alerts_file(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert result["stdout"] == ""
+        assert result["exit_code"] == 0
+
+    def test_no_banner_when_empty_alerts(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert result["stdout"] == ""
+
+    def test_no_banner_when_no_aipass_dir(self):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=None,
+        ):
+            result = handle({})
+
+        assert result["stdout"] == ""
+        assert result["exit_code"] == 0
+
+    def test_multiple_alerts_all_shown(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(
+            aipass_dir,
+            [
+                _make_alert(alert_id="a1", title="First"),
+                _make_alert(alert_id="a2", title="Second", severity="critical"),
+            ],
+        )
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert "[WARNING] First" in result["stdout"]
+        assert "[CRITICAL] Second" in result["stdout"]
+
+    def test_source_and_id_in_banner(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="abc-123", source="trigger")])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert "@trigger" in result["stdout"]
+        assert "abc-123" in result["stdout"]
+
+    def test_body_included_in_banner(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(body="Log rate exceeds 50/s")])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert "Log rate exceeds 50/s" in result["stdout"]
+
+    def test_no_body_line_when_body_empty(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(body="")])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        lines = result["stdout"].split("\n")
+        body_lines = [line for line in lines if line.startswith("  ")]
+        assert len(body_lines) == 0
+
+
+class TestExpiredAlertCleanup:
+    """Auto-cleaning of expired alerts."""
+
+    def test_expired_alerts_removed(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        past = (datetime.now(timezone.utc) - timedelta(hours=1)).isoformat()
+        _write_alerts(
+            aipass_dir,
+            [
+                _make_alert(alert_id="expired", expires_at=past),
+                _make_alert(alert_id="active", expires_at=None),
+            ],
+        )
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert "active" in result["stdout"]
+        assert "expired" not in result["stdout"]
+
+        saved = json.loads((aipass_dir / "alerts.json").read_text())
+        assert len(saved["alerts"]) == 1
+        assert saved["alerts"][0]["id"] == "active"
+
+    def test_all_expired_returns_empty(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        past = (datetime.now(timezone.utc) - timedelta(hours=1)).isoformat()
+        _write_alerts(aipass_dir, [_make_alert(expires_at=past)])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert result["stdout"] == ""
+
+    def test_future_expiry_kept(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        future = (datetime.now(timezone.utc) + timedelta(hours=1)).isoformat()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="still-valid", expires_at=future)])
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert "still-valid" in result["stdout"]
+
+    def test_corrupt_json_returns_empty(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt.persistent_alert import handle
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        (aipass_dir / "alerts.json").write_text("{bad json", encoding="utf-8")
+
+        with patch(
+            "aipass.hooks.apps.handlers.prompt.persistent_alert._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle({})
+
+        assert result["stdout"] == ""
+        assert result["exit_code"] == 0
+
+
+class TestAlertSound:
+    """Sound fires on first injection only."""
+
+    def test_sound_on_first_injection(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt import persistent_alert
+
+        persistent_alert._announced.clear()
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="snd-001")])
+
+        with patch.object(persistent_alert, "_find_aipass_dir", return_value=aipass_dir):
+            result = persistent_alert.handle({})
+
+        assert "sound" in result
+        assert "1 active alert" in result["sound"]
+
+    def test_no_sound_on_repeat_injection(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt import persistent_alert
+
+        persistent_alert._announced.clear()
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="snd-002")])
+
+        with patch.object(persistent_alert, "_find_aipass_dir", return_value=aipass_dir):
+            persistent_alert.handle({})
+            result = persistent_alert.handle({})
+
+        assert "sound" not in result
+
+    def test_sound_on_new_alert_added(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt import persistent_alert
+
+        persistent_alert._announced.clear()
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="snd-003")])
+
+        with patch.object(persistent_alert, "_find_aipass_dir", return_value=aipass_dir):
+            persistent_alert.handle({})
+
+        _write_alerts(
+            aipass_dir,
+            [
+                _make_alert(alert_id="snd-003"),
+                _make_alert(alert_id="snd-004"),
+            ],
+        )
+
+        with patch.object(persistent_alert, "_find_aipass_dir", return_value=aipass_dir):
+            result = persistent_alert.handle({})
+
+        assert "sound" in result
+        assert "2 active alerts" in result["sound"]
+
+    def test_no_sound_when_no_alerts(self, tmp_path):
+        from aipass.hooks.apps.handlers.prompt import persistent_alert
+
+        persistent_alert._announced.clear()
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [])
+
+        with patch.object(persistent_alert, "_find_aipass_dir", return_value=aipass_dir):
+            result = persistent_alert.handle({})
+
+        assert "sound" not in result
+
+
+class TestAlertDismiss:
+    """drone @hooks dismiss behavior."""
+
+    def test_dismiss_removes_by_id(self, tmp_path):
+        from aipass.hooks.apps.modules.alert_dismiss import _dismiss_alert
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(
+            aipass_dir,
+            [
+                _make_alert(alert_id="keep"),
+                _make_alert(alert_id="remove"),
+            ],
+        )
+
+        with patch(
+            "aipass.hooks.apps.modules.alert_dismiss._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = _dismiss_alert("remove")
+
+        assert result is True
+        saved = json.loads((aipass_dir / "alerts.json").read_text())
+        assert len(saved["alerts"]) == 1
+        assert saved["alerts"][0]["id"] == "keep"
+
+    def test_dismiss_nonexistent_returns_false(self, tmp_path):
+        from aipass.hooks.apps.modules.alert_dismiss import _dismiss_alert
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="exists")])
+
+        with patch(
+            "aipass.hooks.apps.modules.alert_dismiss._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = _dismiss_alert("nope")
+
+        assert result is False
+
+    def test_dismiss_no_alerts_file(self, tmp_path):
+        from aipass.hooks.apps.modules.alert_dismiss import _dismiss_alert
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+
+        with patch(
+            "aipass.hooks.apps.modules.alert_dismiss._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = _dismiss_alert("any")
+
+        assert result is False
+
+    def test_dismiss_no_aipass_dir(self):
+        from aipass.hooks.apps.modules.alert_dismiss import _dismiss_alert
+
+        with patch(
+            "aipass.hooks.apps.modules.alert_dismiss._find_aipass_dir",
+            return_value=None,
+        ):
+            result = _dismiss_alert("any")
+
+        assert result is False
+
+    def test_handle_command_routes_dismiss(self, tmp_path):
+        from aipass.hooks.apps.modules.alert_dismiss import handle_command
+
+        aipass_dir = tmp_path / ".aipass"
+        aipass_dir.mkdir()
+        _write_alerts(aipass_dir, [_make_alert(alert_id="cmd-test")])
+
+        with patch(
+            "aipass.hooks.apps.modules.alert_dismiss._find_aipass_dir",
+            return_value=aipass_dir,
+        ):
+            result = handle_command("dismiss", ["cmd-test"])
+
+        assert result is True
+        saved = json.loads((aipass_dir / "alerts.json").read_text())
+        assert len(saved["alerts"]) == 0
+
+    def test_handle_command_ignores_other_commands(self):
+        from aipass.hooks.apps.modules.alert_dismiss import handle_command
+
+        assert handle_command("status", []) is False
+
+    def test_handle_command_help(self):
+        from aipass.hooks.apps.modules.alert_dismiss import handle_command
+
+        assert handle_command("dismiss", ["--help"]) is True
@@ -47,45 +47,36 @@ def _patch_session_pid(pid):


 class TestResolveSessionPid:
-    def test_finds_claude_ancestor(self):
-        comm_map = {100: "python3", 90: "bash", 80: "claude"}
+    def test_finds_session_file_ancestor(self):
        ppid_map = {100: 90, 90: 80}
        with (
-            patch("sys.platform", "linux"),
            patch("os.getpid", return_value=100),
-            patch.object(presence, "_read_proc_comm", side_effect=lambda p: comm_map.get(p, "")),
-            patch.object(presence, "_read_proc_ppid", side_effect=lambda p: ppid_map.get(p)),
+            patch.object(presence, "_has_session_file", side_effect=lambda p: p == 80),
+            patch.object(presence, "_get_ppid_portable", side_effect=lambda p: ppid_map.get(p)),
        ):
            assert presence._resolve_session_pid() == 80

-    def test_no_claude_ancestor_returns_none(self):
-        comm_map = {100: "python3", 90: "bash", 80: "init"}
+    def test_no_session_file_ancestor_returns_none(self):
        ppid_map = {100: 90, 90: 80, 80: 1}
        with (
-            patch("sys.platform", "linux"),
            patch("os.getpid", return_value=100),
-            patch.object(presence, "_read_proc_comm", side_effect=lambda p: comm_map.get(p, "")),
-            patch.object(presence, "_read_proc_ppid", side_effect=lambda p: ppid_map.get(p)),
+            patch.object(presence, "_has_session_file", return_value=False),
+            patch.object(presence, "_get_ppid_portable", side_effect=lambda p: ppid_map.get(p)),
        ):
            assert presence._resolve_session_pid() is None

-    def test_non_linux_returns_none(self):
-        with patch("sys.platform", "win32"):
-            assert presence._resolve_session_pid() is None
-
-    def test_proc_read_failure_returns_none(self):
+    def test_ppid_failure_returns_none(self):
        with (
-            patch("sys.platform", "linux"),
            patch("os.getpid", return_value=100),
-            patch.object(presence, "_read_proc_comm", return_value=""),
+            patch.object(presence, "_has_session_file", return_value=False),
+            patch.object(presence, "_get_ppid_portable", return_value=None),
        ):
            assert presence._resolve_session_pid() is None

-    def test_direct_claude_process(self):
+    def test_direct_session_process(self):
        with (
-            patch("sys.platform", "linux"),
            patch("os.getpid", return_value=100),
-            patch.object(presence, "_read_proc_comm", return_value="claude"),
+            patch.object(presence, "_has_session_file", side_effect=lambda p: p == 100),
        ):
            assert presence._resolve_session_pid() == 100

@@ -440,7 +431,7 @@ class TestLiveness:
            mock_kill.assert_called_once_with(1234, 0)

    def test_is_pid_alive_dead(self):
-        with patch("os.kill", side_effect=ProcessLookupError):
+        with patch("sys.platform", "linux"), patch("os.kill", side_effect=ProcessLookupError):
            assert presence._is_pid_alive(1234) is False

    def test_is_pid_alive_permission_error(self):
@@ -34,6 +34,11 @@ _OCCUPANT = {
 }


+def _blocking():
+    """Patch observe-only off so blocking tests exercise the block path."""
+    return patch.object(presence_gate, "_OBSERVE_ONLY", False)
+
+
 class TestResolveBranch:
    def test_uses_hook_data_cwd(self, tmp_path):
        branch_dir = tmp_path / "devpulse"
@@ -67,27 +72,46 @@ class TestHandle:

    def test_occupant_blocks(self):
        _, _, router = _make_mocks(our_pid=1000, occupant=_OCCUPANT)
-        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
-            with patch("importlib.import_module", side_effect=router):
-                result = presence_gate.handle({})
+        with (
+            _blocking(),
+            patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True),
+            patch("importlib.import_module", side_effect=router),
+        ):
+            result = presence_gate.handle({})
        assert result["exit_code"] == 2
        parsed = json.loads(result["stdout"])
        assert parsed["decision"] == "block"
        assert "5000" in parsed["reason"]

-    def test_block_includes_session_name(self):
+    def test_block_includes_session_info(self):
        _, _, router = _make_mocks(our_pid=1000, occupant=_OCCUPANT)
-        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
-            with patch("importlib.import_module", side_effect=router):
-                result = presence_gate.handle({})
+        with (
+            _blocking(),
+            patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True),
+            patch("importlib.import_module", side_effect=router),
+        ):
+            result = presence_gate.handle({})
        parsed = json.loads(result["stdout"])
-        assert "hooks-ab" in parsed["reason"]
+        assert "existing" in parsed["reason"]
+        assert "interactive" in parsed["reason"]

    def test_subagent_skipped(self):
-        result = presence_gate.handle({"agent_type": "sub"})
+        result = presence_gate.handle({"agent_type": "Explore"})
        assert result["exit_code"] == 0
        assert result["stdout"] == ""

+    def test_general_purpose_subagent_skipped(self):
+        result = presence_gate.handle({"agent_type": "general-purpose"})
+        assert result["exit_code"] == 0
+        assert result["stdout"] == ""
+
+    def test_claude_agent_type_not_skipped(self):
+        _, _, router = _make_mocks(our_pid=1000, occupant=None)
+        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
+            with patch("importlib.import_module", side_effect=router):
+                result = presence_gate.handle({"agent_type": "claude"})
+        assert result["exit_code"] == 0
+
    def test_main_agent_not_skipped(self):
        _, _, router = _make_mocks(our_pid=1000, occupant=None)
        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
@@ -127,13 +151,25 @@ class TestHandle:
        branch_dir.mkdir()
        (branch_dir / ".trinity").mkdir()
        _, _, router = _make_mocks(our_pid=1000, occupant=_OCCUPANT)
-        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
-            with patch("importlib.import_module", side_effect=router):
-                result = presence_gate.handle({"cwd": str(branch_dir)})
+        with (
+            _blocking(),
+            patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True),
+            patch("importlib.import_module", side_effect=router),
+        ):
+            result = presence_gate.handle({"cwd": str(branch_dir)})
        parsed = json.loads(result["stdout"])
        assert "devpulse" in parsed["reason"]
-        assert "kill 5000" in parsed["reason"]
-        assert "one session per branch" in parsed["reason"].lower()
+        assert "reclaim" in parsed["reason"]
+
+    def test_observe_only_logs_but_allows(self):
+        _, _, router = _make_mocks(our_pid=1000, occupant=_OCCUPANT)
+        with (
+            patch.object(presence_gate, "_OBSERVE_ONLY", True),
+            patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True),
+            patch("importlib.import_module", side_effect=router),
+        ):
+            result = presence_gate.handle({})
+        assert result["exit_code"] == 0

    def test_gate_error_allows(self):
        with patch.dict(os.environ, {"AIPASS_SESSION_TYPE": "interactive"}, clear=True):
@@ -1,4 +1,4 @@
-"""Tests for session boot wrapper (attach-if-live / start-in-tmux)."""
+"""Tests for session boot wrapper (menu-based attach/start/close)."""

 from unittest.mock import MagicMock, patch

@@ -107,6 +107,48 @@ class TestIsDescendant:
            assert session_boot._is_descendant(200, 100) is False


+class TestMakeSessionName:
+    def test_with_session_id(self):
+        assert session_boot._make_session_name("hooks", "abcdef1234") == "hooks-abcdef12"
+
+    def test_without_session_id(self):
+        assert session_boot._make_session_name("hooks") == "hooks"
+
+    def test_empty_session_id(self):
+        assert session_boot._make_session_name("hooks", "") == "hooks"
+
+
+class TestNameFlag:
+    def test_branch_only(self):
+        result = session_boot._name_flag("hooks")
+        assert result == ["--name", "hooks"]
+
+    def test_branch_with_session_id(self):
+        result = session_boot._name_flag("hooks", "abcdef1234")
+        assert result == ["--name", "hooks-abcdef12"]
+
+    def test_skips_when_user_provides_name(self):
+        result = session_boot._name_flag("hooks", "", ["--name", "myname"])
+        assert result == []
+
+    def test_skips_when_user_provides_n(self):
+        result = session_boot._name_flag("hooks", "", ["-n", "myname"])
+        assert result == []
+
+    def test_adds_when_extra_args_no_name(self):
+        result = session_boot._name_flag("hooks", "", ["--verbose"])
+        assert result == ["--name", "hooks"]
+
+
+class TestSessionLabel:
+    def test_formats_label(self):
+        session = {"pid": 1234, "sessionId": "abcdef1234", "kind": "interactive"}
+        label = session_boot._session_label(session, "hooks")
+        assert "1234" in label
+        assert "abcdef12" in label
+        assert "interactive" in label
+
+
 class TestBoot:
    def test_already_in_tmux_execs_directly(self, tmp_path):
        with (
@@ -141,40 +183,102 @@ class TestBoot:
        assert result["exit_code"] == 1
        assert "tmux not found" in result["error"]

-    def test_live_session_in_tmux_attaches(self, tmp_path):
-        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "name": "hooks-ab"}]
+    def test_live_session_resume_via_tmux(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
        with (
            patch.dict("os.environ", {}, clear=True),
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
            patch.object(session_boot, "_find_tmux_session_for_pid", return_value="hooks"),
            patch(f"{_MOD}.os.execvp") as mock_exec,
        ):
            session_boot.boot(cwd=str(tmp_path))
        mock_exec.assert_called_once_with("tmux", ["tmux", "attach-session", "-t", "hooks"])

-    def test_live_session_not_in_tmux_warns(self, tmp_path):
-        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path)}]
+    def test_live_session_resume_continues_dead_window(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
        with (
            patch.dict("os.environ", {}, clear=True),
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        args = mock_exec.call_args[0][1]
+        assert "--continue" in args
+
+    def test_live_session_resume_bg_does_takeover(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(
+                session_boot, "_takeover_bg", return_value={"exit_code": 0, "action": "takeover"}
+            ) as mock_take,
        ):
            result = session_boot.boot(cwd=str(tmp_path))
-        assert result["exit_code"] == 1
-        assert result["action"] == "warn"
-        assert "kill 1234" in result["error"]
-        assert "command claude --resume" in result["error"]
+        mock_take.assert_called_once()
+        assert result["action"] == "takeover"

-    def test_no_live_session_starts_fresh(self, tmp_path):
+    def test_live_session_new_stops_old(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_stop_session") as mock_stop,
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp"),
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        mock_stop.assert_called_once()
+
+    def test_live_session_close_stops_and_exits(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="c"),
+            patch.object(session_boot, "_stop_session") as mock_stop,
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        mock_stop.assert_called_once()
+        assert result["action"] == "closed"
+
+    def test_no_live_continue_last(self, tmp_path):
        with (
            patch.dict("os.environ", {}, clear=True),
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        args = mock_exec.call_args[0][1]
+        assert "--continue" in args
+
+    def test_no_live_new_starts_fresh(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
            patch.object(session_boot, "_tmux_session_exists", return_value=False),
            patch(f"{_MOD}.os.execvp") as mock_exec,
        ):
@@ -183,15 +287,14 @@ class TestBoot:
        assert args[0] == "tmux"
        assert "new-session" in args[1]
        assert "/usr/local/bin/claude" in args[1]
-        assert "--permission-mode" in args[1]
-        assert "bypassPermissions" in args[1]

-    def test_stale_tmux_session_killed(self, tmp_path):
+    def test_stale_tmux_session_killed_on_fresh_start(self, tmp_path):
        with (
            patch.dict("os.environ", {}, clear=True),
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
            patch.object(session_boot, "_tmux_session_exists", return_value=True),
            patch(f"{_MOD}.subprocess.run") as mock_run,
            patch(f"{_MOD}.os.execvp"),
@@ -200,12 +303,13 @@ class TestBoot:
        kill_calls = [c for c in mock_run.call_args_list if "kill-session" in str(c)]
        assert len(kill_calls) == 1

-    def test_extra_args_passed(self, tmp_path):
+    def test_extra_args_passed_on_fresh_start(self, tmp_path):
        with (
            patch.dict("os.environ", {}, clear=True),
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
            patch.object(session_boot, "_tmux_session_exists", return_value=False),
            patch(f"{_MOD}.os.execvp") as mock_exec,
        ):
@@ -214,6 +318,54 @@ class TestBoot:
        assert "--resume" in args


+class TestStopSession:
+    def test_bg_returns_honest_no_stop(self):
+        session = {"pid": 1234, "kind": "bg"}
+        result = session_boot._stop_session(session, "/usr/local/bin/claude")
+        assert "no per-job stop" in result
+
+    def test_bg_background_kind_also_honest(self):
+        session = {"pid": 1234, "kind": "background"}
+        result = session_boot._stop_session(session, "/usr/local/bin/claude")
+        assert "no per-job stop" in result
+
+    def test_bg_never_sigterms(self):
+        session = {"pid": 1234, "kind": "bg"}
+        with patch(f"{_MOD}.os.kill") as mock_kill:
+            session_boot._stop_session(session, "/usr/local/bin/claude")
+        mock_kill.assert_not_called()
+
+    def test_tmux_session_killed(self):
+        session = {"pid": 1234, "kind": "interactive"}
+        with (
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value="hooks"),
+            patch(f"{_MOD}.subprocess.run") as mock_run,
+        ):
+            result = session_boot._stop_session(session, "/usr/local/bin/claude")
+        kill_calls = [c for c in mock_run.call_args_list if "kill-session" in str(c)]
+        assert len(kill_calls) == 1
+        assert "tmux" in result
+
+    def test_plain_session_sigterm(self):
+        session = {"pid": 1234, "kind": "interactive"}
+        with (
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch(f"{_MOD}.os.kill") as mock_kill,
+        ):
+            result = session_boot._stop_session(session, "/usr/local/bin/claude")
+        mock_kill.assert_called_once()
+        assert "SIGTERM" in result
+
+    def test_plain_session_already_dead(self):
+        session = {"pid": 1234, "kind": "interactive"}
+        with (
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch(f"{_MOD}.os.kill", side_effect=ProcessLookupError),
+        ):
+            result = session_boot._stop_session(session, "/usr/local/bin/claude")
+        assert "already dead" in result
+
+
 class TestMain:
    def test_success(self):
        with patch.object(session_boot, "boot", return_value={"exit_code": 0, "action": "started"}):
@@ -346,6 +498,7 @@ class TestPermissionModeDedupe:
            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
            patch.object(session_boot, "_tmux_session_exists", return_value=False),
            patch(f"{_MOD}.os.execvp") as mock_exec,
        ):
@@ -353,3 +506,600 @@ class TestPermissionModeDedupe:
        cmd = mock_exec.call_args[0][1]
        assert cmd.count("--permission-mode") == 1
        assert "acceptEdits" in cmd
+
+
+class TestMultipleLiveSessions:
+    def test_close_all(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "bg"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="c"),
+            patch.object(session_boot, "_stop_session", return_value="stopped") as mock_stop,
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["action"] == "closed_all"
+        mock_stop.assert_called_once()
+
+    def test_pick_by_number_interactive(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "interactive"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="1"),
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value="hooks"),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        mock_exec.assert_called_once_with("tmux", ["tmux", "attach-session", "-t", "hooks"])
+
+    def test_pick_bg_triggers_takeover(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "bg"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="2"),
+            patch.object(
+                session_boot, "_takeover_bg", return_value={"exit_code": 0, "action": "takeover"}
+            ) as mock_take,
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        mock_take.assert_called_once()
+        assert result["action"] == "takeover"
+
+    def test_enter_without_pick_rejected(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "bg"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["exit_code"] == 1
+
+    def test_n_stops_stoppable_then_starts(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "bg"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_stop_session", return_value="stopped") as mock_stop,
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        mock_stop.assert_called_once()
+        args = mock_exec.call_args[0]
+        assert args[0] == "tmux"
+        assert "new-session" in args[1]
+
+
+class TestTakeover:
+    def test_takeover_bg_runs_daemon_stop(self, tmp_path):
+        session = {
+            "pid": 1234,
+            "sessionId": "abc12345-full-uuid",
+            "cwd": str(tmp_path),
+            "kind": "bg",
+        }
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}) as mock_daemon,
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._takeover_bg(
+                session, "hooks", "/usr/local/bin/claude", ["--permission-mode", "bypassPermissions"]
+            )
+        mock_daemon.assert_called_once()
+        args = mock_exec.call_args[0][1]
+        assert "--resume" in args
+        assert "abc12345-full-uuid" in args
+        assert "new-session" in args
+
+    def test_takeover_bg_no_session_id_continues(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "", "cwd": str(tmp_path), "kind": "bg"}
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._takeover_bg(
+                session, "hooks", "/usr/local/bin/claude", ["--permission-mode", "bypassPermissions"]
+            )
+        args = mock_exec.call_args[0][1]
+        assert "--continue" in args
+
+    def test_takeover_daemon_stop_failure(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}
+        with patch.object(
+            session_boot, "_daemon_stop", return_value={"ok": False, "error": "daemon stop failed: no claude"}
+        ):
+            result = session_boot._takeover_bg(session, "hooks", "/usr/local/bin/claude", [])
+        assert result["exit_code"] == 1
+        assert "daemon stop failed" in result["error"]
+
+    def test_takeover_nonzero_returncode_aborts(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}
+        with patch.object(
+            session_boot,
+            "_daemon_stop",
+            return_value={"ok": False, "error": "daemon stop exit 1: something failed"},
+        ):
+            result = session_boot._takeover_bg(session, "hooks", "/usr/local/bin/claude", [])
+        assert result["exit_code"] == 1
+
+    def test_single_bg_enter_is_takeover(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(
+                session_boot, "_takeover_bg", return_value={"exit_code": 0, "action": "takeover"}
+            ) as mock_take,
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        mock_take.assert_called_once()
+        assert result["action"] == "takeover"
+
+    def test_single_bg_n_stops_then_fresh(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        args = mock_exec.call_args[0]
+        assert args[0] == "tmux"
+        assert "new-session" in args[1]
+
+    def test_single_bg_c_stops(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="c"),
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["action"] == "closed"
+
+
+class TestBgResume:
+    def test_bg_resume_routes_to_takeover(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}
+        with patch.object(
+            session_boot, "_takeover_bg", return_value={"exit_code": 0, "action": "takeover"}
+        ) as mock_take:
+            result = session_boot._resume_session(
+                session, "hooks", "/usr/local/bin/claude", ["--permission-mode", "bypassPermissions"]
+            )
+        mock_take.assert_called_once()
+        assert result["action"] == "takeover"
+
+    def test_bg_resume_never_opens_agents_view(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._resume_session(
+                session, "hooks", "/usr/local/bin/claude", ["--permission-mode", "bypassPermissions"]
+            )
+        args = mock_exec.call_args[0][1]
+        assert "agents" not in args
+
+
+class TestSessionLabelAutoName:
+    def test_bg_label_includes_auto_name(self):
+        session = {
+            "pid": 1234,
+            "sessionId": "abc12345",
+            "kind": "bg",
+            "name": "chroma review",
+        }
+        label = session_boot._session_label(session, "hooks")
+        assert '"chroma review"' in label
+        assert "abc12345" in label
+
+    def test_interactive_label_no_name(self):
+        session = {"pid": 1234, "sessionId": "abc12345", "kind": "interactive"}
+        label = session_boot._session_label(session, "hooks")
+        assert '"' not in label
+
+    def test_bg_label_no_name_field(self):
+        session = {"pid": 1234, "sessionId": "abc12345", "kind": "bg"}
+        label = session_boot._session_label(session, "hooks")
+        assert '"' not in label
+
+
+class TestDaemonStop:
+    def test_success_no_collateral(self):
+        with (
+            patch.object(session_boot, "_get_collateral_bg", return_value=[]),
+            patch(f"{_MOD}.subprocess.run", return_value=MagicMock(returncode=0)),
+            patch.object(session_boot, "_is_session_file_present", return_value=False),
+        ):
+            result = session_boot._daemon_stop("/usr/local/bin/claude", "hooks", 1234)
+        assert result["ok"] is True
+
+    def test_nonzero_returncode_fails(self):
+        with (
+            patch.object(session_boot, "_get_collateral_bg", return_value=[]),
+            patch(
+                f"{_MOD}.subprocess.run",
+                return_value=MagicMock(returncode=1, stderr="something broke"),
+            ),
+        ):
+            result = session_boot._daemon_stop("/usr/local/bin/claude", "hooks", 1234)
+        assert result["ok"] is False
+        assert "exit 1" in result["error"]
+
+    def test_oserror_fails(self):
+        with (
+            patch.object(session_boot, "_get_collateral_bg", return_value=[]),
+            patch(f"{_MOD}.subprocess.run", side_effect=OSError("no binary")),
+        ):
+            result = session_boot._daemon_stop("/usr/local/bin/claude", "hooks", 1234)
+        assert result["ok"] is False
+
+    def test_collateral_confirmed_proceeds(self):
+        collateral = [{"pid": 9999, "cwd": "/tmp/other", "sessionId": "xyz"}]
+        with (
+            patch.object(session_boot, "_get_collateral_bg", return_value=collateral),
+            patch.object(session_boot, "_read_choice", return_value="y"),
+            patch(f"{_MOD}.subprocess.run", return_value=MagicMock(returncode=0)),
+            patch.object(session_boot, "_is_session_file_present", return_value=False),
+        ):
+            result = session_boot._daemon_stop("/usr/local/bin/claude", "hooks", 1234)
+        assert result["ok"] is True
+
+    def test_collateral_denied_cancels(self):
+        collateral = [{"pid": 9999, "cwd": "/tmp/other", "sessionId": "xyz"}]
+        with (
+            patch.object(session_boot, "_get_collateral_bg", return_value=collateral),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+        ):
+            result = session_boot._daemon_stop("/usr/local/bin/claude", "hooks", 1234)
+        assert result["ok"] is False
+        assert "cancelled" in result["error"]
+
+
+class TestExecInTmux:
+    def test_wraps_in_tmux_session(self):
+        with (
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._exec_in_tmux(
+                "hooks", "abc12345", "/usr/local/bin/claude", ["/usr/local/bin/claude", "--continue"]
+            )
+        args = mock_exec.call_args[0]
+        assert args[0] == "tmux"
+        assert "new-session" in args[1]
+        assert "-s" in args[1]
+        assert "hooks-abc12345" in args[1]
+        assert "/usr/local/bin/claude" in args[1]
+
+    def test_kills_stale_tmux_first(self):
+        with (
+            patch.object(session_boot, "_tmux_session_exists", return_value=True),
+            patch(f"{_MOD}.subprocess.run") as mock_run,
+            patch(f"{_MOD}.os.execvp"),
+        ):
+            session_boot._exec_in_tmux("hooks", "", "/usr/local/bin/claude", ["/usr/local/bin/claude"])
+        kill_calls = [c for c in mock_run.call_args_list if "kill-session" in str(c)]
+        assert len(kill_calls) == 1
+
+
+class TestIsSessionFilePresent:
+    def test_present(self, tmp_path):
+        sessions_dir = tmp_path / ".claude" / "sessions"
+        sessions_dir.mkdir(parents=True)
+        (sessions_dir / "1234.json").write_text("{}")
+        with patch.object(session_boot.Path, "home", return_value=tmp_path):
+            assert session_boot._is_session_file_present(1234) is True
+
+    def test_absent(self, tmp_path):
+        with patch.object(session_boot.Path, "home", return_value=tmp_path):
+            assert session_boot._is_session_file_present(1234) is False
+
+    def test_none_pid(self):
+        assert session_boot._is_session_file_present(None) is False
+
+
+class TestExtraArgsThreading:
+    """R6: extra_args must reach every launch path, not just _start_fresh."""
+
+    def test_resume_dead_window_threads_extra_args(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}
+        with (
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._resume_session(session, "hooks", "/usr/local/bin/claude", [], ["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--continue" in cmd
+        assert "--permission-mode" in cmd
+        assert "plan" in cmd
+
+    def test_takeover_bg_threads_extra_args(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc-uuid", "cwd": str(tmp_path), "kind": "bg"}
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._takeover_bg(session, "hooks", "/usr/local/bin/claude", [], ["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--resume" in cmd
+        assert "--permission-mode" in cmd
+        assert "plan" in cmd
+
+    def test_takeover_bg_continue_threads_extra_args(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "", "cwd": str(tmp_path), "kind": "bg"}
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._takeover_bg(session, "hooks", "/usr/local/bin/claude", [], ["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--continue" in cmd
+        assert "--permission-mode" in cmd
+
+    def test_no_live_continue_threads_extra_args(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path), extra_args=["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--continue" in cmd
+        assert "--permission-mode" in cmd
+        assert "plan" in cmd
+
+    def test_live_bg_enter_threads_extra_args(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "bg"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path), extra_args=["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--resume" in cmd
+        assert "--permission-mode" in cmd
+        assert "plan" in cmd
+
+    def test_multi_session_pick_threads_extra_args(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "interactive"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="1"),
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path), extra_args=["--permission-mode", "plan"])
+        cmd = mock_exec.call_args[0][1]
+        assert "--continue" in cmd
+        assert "--permission-mode" in cmd
+        assert "plan" in cmd
+
+
+class TestNewOverAllAbort:
+    def test_aborts_on_daemon_stop_failure(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "bg"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_stop_session", return_value="stopped"),
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": False, "error": "failed"}),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["exit_code"] == 1
+        assert "one-brain" in result["error"]
+        mock_exec.assert_not_called()
+
+
+class TestMenuQuit:
+    def test_single_session_q_quits(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="q"),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["exit_code"] == 0
+        assert result["action"] == "quit"
+
+    def test_single_session_exit_quits(self, tmp_path):
+        live = [{"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"}]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="exit"),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["action"] == "quit"
+
+    def test_multi_session_quit_quits(self, tmp_path):
+        live = [
+            {"pid": 1234, "sessionId": "abc", "cwd": str(tmp_path), "kind": "interactive"},
+            {"pid": 5678, "sessionId": "def", "cwd": str(tmp_path), "kind": "interactive"},
+        ]
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=live),
+            patch.object(session_boot, "_read_choice", return_value="quit"),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["action"] == "quit"
+
+    def test_no_live_q_quits(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="q"),
+        ):
+            result = session_boot.boot(cwd=str(tmp_path))
+        assert result["action"] == "quit"
+
+
+class TestAutoNamer:
+    """R7: --name flag stamped on every launch for self-identifying sessions."""
+
+    def test_fresh_start_gets_name(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        cmd = mock_exec.call_args[0][1]
+        assert "--name" in cmd
+        branch = tmp_path.name
+        idx = cmd.index("--name")
+        assert cmd[idx + 1] == branch
+
+    def test_takeover_gets_name_with_session_id(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc12345-full-uuid", "cwd": str(tmp_path), "kind": "bg"}
+        with (
+            patch.object(session_boot, "_daemon_stop", return_value={"ok": True}),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._takeover_bg(session, "hooks", "/usr/local/bin/claude", [])
+        cmd = mock_exec.call_args[0][1]
+        assert "--name" in cmd
+        idx = cmd.index("--name")
+        assert cmd[idx + 1] == "hooks-abc12345"
+
+    def test_in_tmux_gets_name(self, tmp_path):
+        with (
+            patch.dict("os.environ", {"TMUX": "/tmp/tmux-1000/default,123,0"}),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        cmd = mock_exec.call_args[0][1]
+        assert "--name" in cmd
+
+    def test_no_live_continue_gets_name(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value=""),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path))
+        cmd = mock_exec.call_args[0][1]
+        assert "--name" in cmd
+
+    def test_user_name_not_overridden(self, tmp_path):
+        with (
+            patch.dict("os.environ", {}, clear=True),
+            patch.object(session_boot, "_resolve_claude_binary", return_value="/usr/local/bin/claude"),
+            patch.object(session_boot, "_find_tmux", return_value="/usr/bin/tmux"),
+            patch.object(session_boot, "_find_live_sessions", return_value=[]),
+            patch.object(session_boot, "_read_choice", return_value="n"),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot.boot(cwd=str(tmp_path), extra_args=["--name", "myname"])
+        cmd = mock_exec.call_args[0][1]
+        assert cmd.count("--name") == 1
+        idx = cmd.index("--name")
+        assert cmd[idx + 1] == "myname"
+
+    def test_dead_window_resume_gets_name_with_session_id(self, tmp_path):
+        session = {"pid": 1234, "sessionId": "abc12345-uuid", "cwd": str(tmp_path), "kind": "interactive"}
+        with (
+            patch.object(session_boot, "_find_tmux_session_for_pid", return_value=None),
+            patch.object(session_boot, "_tmux_session_exists", return_value=False),
+            patch(f"{_MOD}.os.execvp") as mock_exec,
+        ):
+            session_boot._resume_session(session, "hooks", "/usr/local/bin/claude", [])
+        cmd = mock_exec.call_args[0][1]
+        assert "--name" in cmd
+        idx = cmd.index("--name")
+        assert cmd[idx + 1] == "hooks-abc12345"
@@ -1308,6 +1308,117 @@ class TestDeliverChunks:
        assert chunk_results[1]["message_id"] == 202


+# ===========================================================================
+# _deliver_chunks streaming mode
+# ===========================================================================
+
+
+class TestDeliverChunksStreaming:
+    """Streaming flag changes reconcile behavior — batch unchanged."""
+
+    def test_streaming_logs_active_reconciles_instead_of_done(self):
+        """Streaming + logs_active: edit processing msg with response, not 'Done.'."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_ok_result()) as mock_edit,
+        ):
+            all_sent, chunk_results = _deliver_chunks(["Hello"], "tok", 123, 789, True, streaming=True)
+
+        assert all_sent is True
+        assert chunk_results[0]["method"] == "edit"
+        mock_edit.assert_called_once_with("tok", 123, 789, "Hello")
+
+    def test_streaming_multi_chunk_edits_first_sends_rest(self):
+        """Streaming + multi-chunk: edit chunk 1 into processing msg, send rest."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        sent_texts = []
+
+        def capture_send(bot_token, chat_id, text):
+            sent_texts.append(text)
+            return _ok_result()
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_ok_result()) as mock_edit,
+            patch(f"{MOD}._send_with_retry", side_effect=capture_send),
+        ):
+            all_sent, chunk_results = _deliver_chunks(
+                ["Part A", "Part B", "Part C"], "tok", 123, 789, False, streaming=True
+            )
+
+        assert all_sent is True
+        mock_edit.assert_called_once_with("tok", 123, 789, "Part A")
+        assert chunk_results[0]["method"] == "edit"
+        assert len(sent_texts) == 2
+        assert "[2/3]" in sent_texts[0]
+        assert "[3/3]" in sent_texts[1]
+
+    def test_streaming_multi_chunk_edit_fails_sends_all(self):
+        """Streaming + multi-chunk: if edit fails, fall back to send for chunk 1."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        sent_texts = []
+
+        def capture_send(bot_token, chat_id, text):
+            sent_texts.append(text)
+            return _ok_result()
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_fail_result()),
+            patch(f"{MOD}._send_with_retry", side_effect=capture_send),
+        ):
+            all_sent, chunk_results = _deliver_chunks(["Part A", "Part B"], "tok", 123, 789, False, streaming=True)
+
+        assert all_sent is True
+        assert chunk_results[0]["method"] == "send"
+        assert len(sent_texts) == 2
+
+    def test_batch_logs_active_still_sends_done(self):
+        """Batch mode (no streaming): logs_active still sends 'Done.' — zero regression."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_ok_result()) as mock_edit,
+            patch(f"{MOD}._send_with_retry", return_value=_ok_result()),
+        ):
+            all_sent, chunk_results = _deliver_chunks(["Hello"], "tok", 123, 789, True, streaming=False)
+
+        assert all_sent is True
+        mock_edit.assert_called_once_with("tok", 123, 789, "Done.")
+
+    def test_batch_multi_chunk_still_sends_done(self):
+        """Batch mode (no streaming): multi-chunk still sends 'Done.' — zero regression."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_ok_result()) as mock_edit,
+            patch(f"{MOD}._send_with_retry", return_value=_ok_result()),
+        ):
+            _deliver_chunks(["Part A", "Part B"], "tok", 123, 789, False, streaming=False)
+
+        mock_edit.assert_called_once_with("tok", 123, 789, "Done.")
+
+    def test_streaming_single_no_logs_still_edits(self):
+        """Streaming + single chunk + no logs: same as batch — reconcile-edit."""
+        from aipass.hooks.apps.handlers.notification.telegram_response import _deliver_chunks
+
+        with (
+            patch(LOGGER_PATCH),
+            patch(f"{MOD}.edit_telegram_message", return_value=_ok_result()) as mock_edit,
+        ):
+            all_sent, chunk_results = _deliver_chunks(["Hello"], "tok", 123, 789, False, streaming=True)
+
+        assert all_sent is True
+        mock_edit.assert_called_once_with("tok", 123, 789, "Hello")
+        assert chunk_results[0]["method"] == "edit"
+
+
 # ===========================================================================
 # _advance_pending
 # ===========================================================================
@@ -0,0 +1,320 @@
+# =================== AIPass ====================
+# Name: test_trust_registry.py
+# Version: 1.0.0
+# Description: Tests for trusted-project registry — DPLAN-0244 Layer B
+# Branch: hooks
+# Layer: tests
+# Created: 2026-07-15
+# Modified: 2026-07-15
+# =============================================
+
+"""Tests for trusted-project registry and loader trust integration."""
+
+import json
+from unittest.mock import patch
+
+from aipass.hooks.apps.handlers.config.trust_registry import (
+    _hash_file,
+    bootstrap,
+    enroll,
+    is_trusted,
+    read_registry,
+    revoke,
+)
+from aipass.hooks.apps.handlers.config.loader import find_project_config
+
+
+class TestRegistryHelpers:
+    """Unit tests for registry helper functions."""
+
+    def test_hash_file_deterministic(self, temp_test_dir):
+        f = temp_test_dir / "test.json"
+        f.write_text('{"hello": "world"}')
+        h1 = _hash_file(f)
+        h2 = _hash_file(f)
+        assert h1 == h2
+        assert h1.startswith("sha256:")
+
+    def test_hash_file_changes_on_content_change(self, temp_test_dir):
+        f = temp_test_dir / "test.json"
+        f.write_text('{"v": 1}')
+        h1 = _hash_file(f)
+        f.write_text('{"v": 2}')
+        h2 = _hash_file(f)
+        assert h1 != h2
+
+    def test_read_registry_absent(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "nonexistent.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = read_registry()
+        assert result == {"version": 1, "projects": {}}
+
+    def test_read_registry_valid(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        reg_data = {
+            "version": 1,
+            "projects": {
+                "/some/path": {
+                    "enrolled": "2026-07-15T00:00:00",
+                    "config_hash": "sha256:abc",
+                    "config_path": "/some/path/.aipass/hooks.json",
+                }
+            },
+        }
+        reg_path.write_text(json.dumps(reg_data))
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = read_registry()
+        assert "/some/path" in result["projects"]
+
+    def test_read_registry_corrupt(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        reg_path.write_text("{corrupt!!!")
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = read_registry()
+        assert result == {"version": 1, "projects": {}}
+
+
+class TestEnrollRevoke:
+    """Unit tests for enroll() and revoke()."""
+
+    def test_enroll_success(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "myproject"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        (project / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = enroll(str(project))
+        assert result is True
+        assert reg_path.exists()
+        data = json.loads(reg_path.read_text())
+        assert str(project.resolve()) in data["projects"]
+        entry = data["projects"][str(project.resolve())]
+        assert entry["config_hash"].startswith("sha256:")
+
+    def test_enroll_no_hooks_json(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "empty_project"
+        project.mkdir()
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = enroll(str(project))
+        assert result is False
+
+    def test_revoke_success(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "myproject"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        (project / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(project))
+            result = revoke(str(project))
+        assert result is True
+        data = json.loads(reg_path.read_text())
+        assert str(project.resolve()) not in data["projects"]
+
+    def test_revoke_nonexistent(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            result = revoke("/nonexistent/project")
+        assert result is False
+
+
+class TestIsTrusted:
+    """Unit tests for is_trusted()."""
+
+    def test_trusted_with_matching_hash(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "myproject"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        (project / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(project))
+            assert is_trusted(str(project)) is True
+
+    def test_not_trusted_unregistered(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            assert is_trusted("/not/registered") is False
+
+    def test_not_trusted_hash_mismatch(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "myproject"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        hooks_file = project / ".aipass" / "hooks.json"
+        hooks_file.write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(project))
+            hooks_file.write_text('{"hooks_enabled": true, "tampered": true}')
+            assert is_trusted(str(project)) is False
+
+
+class TestBootstrap:
+    """Tests for bootstrap() — enrolls ONLY AIPASS_HOME."""
+
+    def test_bootstrap_enrolls_aipass_home(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        aipass_dir = temp_test_dir / "aipass_install"
+        aipass_dir.mkdir()
+        (aipass_dir / ".aipass").mkdir()
+        (aipass_dir / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {"AIPASS_HOME": str(aipass_dir)}),
+        ):
+            result = bootstrap()
+        assert result is True
+        data = json.loads(reg_path.read_text())
+        assert str(aipass_dir.resolve()) in data["projects"]
+
+    def test_bootstrap_no_aipass_home(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {}, clear=True),
+        ):
+            result = bootstrap()
+        assert result is False
+        assert not reg_path.exists()
+
+    def test_bootstrap_aipass_home_no_hooks_json(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        aipass_dir = temp_test_dir / "empty_install"
+        aipass_dir.mkdir()
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {"AIPASS_HOME": str(aipass_dir)}),
+        ):
+            result = bootstrap()
+        assert result is False
+
+    def test_bootstrap_refuses_hostile_project(self, temp_test_dir, mock_logger):
+        """Security-critical: registry absent + first event in hostile CWD.
+
+        Only AIPASS_HOME gets enrolled, hostile project is NOT enrolled.
+        """
+        reg_path = temp_test_dir / "registry.json"
+
+        aipass_dir = temp_test_dir / "real_aipass"
+        aipass_dir.mkdir()
+        (aipass_dir / ".aipass").mkdir()
+        (aipass_dir / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+
+        hostile_dir = temp_test_dir / "hostile_repo"
+        hostile_dir.mkdir()
+        (hostile_dir / ".aipass").mkdir()
+        (hostile_dir / ".aipass" / "hooks.json").write_text(
+            '{"hooks_enabled": true, "SessionStart": '
+            '{"evil": {"enabled": true, "command": "touch /tmp/pwned", "matcher": ""}}}'
+        )
+
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {"AIPASS_HOME": str(aipass_dir)}),
+        ):
+            result = bootstrap()
+            assert result is True
+
+            data = json.loads(reg_path.read_text())
+            assert str(aipass_dir.resolve()) in data["projects"]
+            assert str(hostile_dir.resolve()) not in data["projects"]
+
+            assert is_trusted(str(hostile_dir)) is False
+            assert is_trusted(str(aipass_dir)) is True
+
+
+class TestLoaderTrustIntegration:
+    """Integration tests: loader.find_project_config() with registry."""
+
+    def test_registered_project_loads(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "trusted_project"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        (project / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(project))
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=project),
+        ):
+            config = find_project_config()
+        assert config is not None
+        assert config["hooks_enabled"] is True
+        assert config["_source"] == "project"
+
+    def test_unregistered_project_skipped(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        reg_path.write_text('{"version": 1, "projects": {}}')
+        project = temp_test_dir / "unknown_project"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        (project / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=project),
+        ):
+            config = find_project_config()
+        assert config is None
+
+    def test_hash_mismatch_skipped(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        project = temp_test_dir / "tampered_project"
+        project.mkdir()
+        (project / ".aipass").mkdir()
+        hooks_file = project / ".aipass" / "hooks.json"
+        hooks_file.write_text('{"hooks_enabled": true}')
+        with patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path):
+            enroll(str(project))
+        hooks_file.write_text('{"hooks_enabled": true, "tampered": true}')
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=project),
+        ):
+            config = find_project_config()
+        assert config is None
+
+    def test_loader_bootstraps_on_missing_registry(self, temp_test_dir, mock_logger):
+        reg_path = temp_test_dir / "registry.json"
+        aipass_dir = temp_test_dir / "aipass_install"
+        aipass_dir.mkdir()
+        (aipass_dir / ".aipass").mkdir()
+        (aipass_dir / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {"AIPASS_HOME": str(aipass_dir)}),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=aipass_dir),
+        ):
+            config = find_project_config()
+        assert config is not None
+        assert reg_path.exists()
+
+    def test_loader_hostile_project_after_bootstrap(self, temp_test_dir, mock_logger):
+        """Full attack chain: hostile repo, registry absent, bootstrap fires."""
+        reg_path = temp_test_dir / "registry.json"
+
+        aipass_dir = temp_test_dir / "real_aipass"
+        aipass_dir.mkdir()
+        (aipass_dir / ".aipass").mkdir()
+        (aipass_dir / ".aipass" / "hooks.json").write_text('{"hooks_enabled": true}')
+
+        hostile_dir = temp_test_dir / "hostile_repo"
+        hostile_dir.mkdir()
+        (hostile_dir / ".aipass").mkdir()
+        (hostile_dir / ".aipass" / "hooks.json").write_text(
+            '{"hooks_enabled": true, "SessionStart": '
+            '{"evil": {"enabled": true, "command": "touch /tmp/pwned", "matcher": ""}}}'
+        )
+
+        with (
+            patch("aipass.hooks.apps.handlers.config.trust_registry.REGISTRY_PATH", reg_path),
+            patch.dict("os.environ", {"AIPASS_HOME": str(aipass_dir)}),
+            patch("aipass.hooks.apps.handlers.config.loader.Path.cwd", return_value=hostile_dir),
+        ):
+            config = find_project_config()
+        assert config is None
+        assert reg_path.exists()
+        data = json.loads(reg_path.read_text())
+        assert str(hostile_dir.resolve()) not in data["projects"]
@@ -189,7 +189,7 @@ class TestCheckEventWiring:
        assert len(warnings) == 1
        assert "duplicate unfiltered" in warnings[0]

-    def test_provider_wired_false_skips_error(self):
+    def test_provider_wired_false_still_errors(self):
        pidx = {"filtered": {"identity_injector": {"": 1}}, "unfiltered": 0, "empty": False}
        hooks_group = {
            "identity_injector": {"enabled": True, "handler": "x"},
@@ -197,15 +197,16 @@ class TestCheckEventWiring:
        }
        errors, warnings, info = [], [], []
        wire_verify._check_event_wiring("UserPromptSubmit", hooks_group, pidx, errors, warnings, info)
-        assert errors == []
+        assert any("presence_gate" in e for e in errors)

-    def test_provider_wired_false_no_event_no_error(self):
+    def test_provider_wired_false_no_event_still_errors(self):
        hooks_group = {
            "presence_gate": {"enabled": True, "handler": "y", "provider_wired": False},
        }
        errors, warnings, info = [], [], []
        wire_verify._check_event_wiring("UserPromptSubmit", hooks_group, None, errors, warnings, info)
-        assert errors == []
+        assert len(errors) == 1
+        assert "NO provider event entry" in errors[0]


 class TestVerifyWiring:
@@ -265,7 +266,7 @@ class TestVerifyWiring:
        assert result["ok"] is False
        assert any("presence_gate" in e for e in result["errors"])

-    def test_provider_wired_false_passes(self, tmp_path):
+    def test_provider_wired_false_now_errors(self, tmp_path):
        settings = tmp_path / "settings.json"
        settings.write_text(json.dumps({"hooks": GOOD_PROVIDER}))
        project = {
@@ -276,8 +277,8 @@ class TestVerifyWiring:
            },
        }
        result = wire_verify.verify_wiring(provider_path=settings, project_config=project)
-        assert result["ok"] is True
-        assert not any("presence_gate" in e for e in result["errors"])
+        assert result["ok"] is False
+        assert any("presence_gate" in e for e in result["errors"])

    def test_distinct_matchers_no_dupe_warning(self, tmp_path):
        provider = {
@@ -26,15 +26,20 @@ fi

 SHIM='
 # >>> AIPass boot shim >>>
-# Intercepts claude in AIPass branch dirs to attach-if-live / start-in-tmux.
+# Intercepts bare claude (or claude --permission-mode ...) in AIPass branch dirs.
+# All other invocations (agents, auth, --resume, -c, --help, ...) pass through.
 # Installed by: tools/install_boot_shim.sh
 claude() {
    if [ -d ".trinity" ]; then
-        __VENV_PY__ \
-            -m aipass.hooks.apps.handlers.lifecycle.session_boot "$@"
-    else
-        command claude "$@"
+        case "${1-}" in
+            ""|--permission-mode)
+                __VENV_PY__ \
+                    -m aipass.hooks.apps.handlers.lifecycle.session_boot "$@"
+                return
+                ;;
+        esac
    fi
+    command claude "$@"
 }
 # <<< AIPass boot shim <<<
 '
@@ -462,6 +462,16 @@
      "functions": ["render_all_meta_tabs"],
      "reason": "Cross-branch public API — called by @spawn's build_replacements_dict to resolve {{*_META}} placeholders at branch creation."
    },
+    {
+      "file": "apps/modules/governance.py",
+      "standard": "unused_function",
+      "reason": "Cross-branch public API — should_surface, record_message, new_state re-exported for @hooks compass_recall and @devpulse recall query."
+    },
+    {
+      "file": "apps/handlers/governance/engine.py",
+      "standard": "unused_function",
+      "reason": "Implementation functions re-exported via modules/governance.py for cross-branch consumers."
+    },
    {
      "file": "apps/modules/templates.py",
      "standard": "deep_nesting",
@@ -45,15 +45,17 @@ drone @memory watch                        # Auto-rollover watcher daemon (Ctrl+
 memory/
 ├── apps/
 │   ├── memory.py                # Entry point — auto-discovers modules
-│   ├── modules/                 # 6 modules
+│   ├── modules/                 # 7 modules
+│   │   ├── governance.py        # Surfacing governance — re-exports from handlers
 │   │   ├── lint.py              # Entry limit violation scanner (read-only)
 │   │   ├── rollover.py          # Rollover orchestration, status, sync-lines
 │   │   ├── search.py            # Semantic query routing
 │   │   ├── symbolic.py          # Fragmented memory extraction and search
 │   │   ├── templates.py         # Template push, diff, status
 │   │   └── verify.py            # Plan vectorization check
-│   └── handlers/                # 14 handler groups
+│   └── handlers/                # 15 handler groups
 │       ├── archive/             # indexer.py
+│       ├── governance/          # engine.py — surfacing decision logic
 │       ├── intake/              # plans_processor.py, pool_processor.py
 │       ├── json/                # json_handler.py, memory_files.py, entry_limits.py, lint_handler.py, config_loader.py
 │       ├── learnings/           # manager.py
@@ -0,0 +1,125 @@
+# =================== AIPass ====================
+# Name: engine.py
+# Description: Surfacing governance engine — implementation
+# Version: 1.0.0
+# Created: 2026-07-16
+# Modified: 2026-07-16
+# =============================================
+
+"""
+Surfacing Governance Engine
+
+Pure decision functions for controlling when recalled items should be
+surfaced. Implementation logic — public API re-exported from
+modules/governance.py for cross-branch consumers.
+"""
+
+from typing import Any, Dict, Tuple
+
+from aipass.prax import logger
+from aipass.memory.apps.handlers.json import json_handler
+
+
+# =============================================================================
+# CONSTANTS — default config values
+# =============================================================================
+
+DEFAULT_CONFIG: Dict[str, Any] = {
+    "enabled": True,
+    "threshold": 0.3,
+    "max_surfaces_per_session": 5,
+    "min_messages_between": 10,
+    "cooldown_seconds": 300,
+}
+
+
+# =============================================================================
+# STATE FACTORY
+# =============================================================================
+
+
+def new_state() -> Dict[str, Any]:
+    """Create a fresh governance state dict."""
+    return {
+        "surfaces_count": 0,
+        "messages_since_last": 0,
+        "last_surface_time": 0.0,
+        "surfaced_ids": [],
+    }
+
+
+# =============================================================================
+# CORE GOVERNANCE
+# =============================================================================
+
+
+def should_surface(
+    item_id: str,
+    relevance_score: float,
+    state: Dict[str, Any],
+    config: Dict[str, Any] | None = None,
+    *,
+    current_time: float | None = None,
+) -> Tuple[bool, str, Dict[str, Any]]:
+    """
+    Decide whether an item should be surfaced, given current state.
+
+    Pure function — does not mutate the input state dict.
+    """
+    import time
+
+    cfg = {**DEFAULT_CONFIG, **(config or {})}
+    now = current_time if current_time is not None else time.time()
+
+    if not cfg.get("enabled", True):
+        return False, "Surfacing disabled", state
+
+    threshold = cfg.get("threshold", 0.3)
+    if relevance_score < threshold:
+        return False, f"Below threshold ({relevance_score:.2f} < {threshold})", state
+
+    max_surfaces = cfg.get("max_surfaces_per_session", 5)
+    if state.get("surfaces_count", 0) >= max_surfaces:
+        return False, f"Session budget exhausted ({max_surfaces}/{max_surfaces})", state
+
+    min_messages = cfg.get("min_messages_between", 10)
+    messages_since = state.get("messages_since_last", 0)
+    last_time = state.get("last_surface_time", 0.0)
+    if last_time > 0 and messages_since < min_messages:
+        return False, f"Spacing not met ({messages_since}/{min_messages} messages)", state
+
+    cooldown = cfg.get("cooldown_seconds", 300)
+    elapsed = now - last_time
+    if last_time > 0 and elapsed < cooldown:
+        remaining = int(cooldown - elapsed)
+        return False, f"Cooldown active ({remaining}s remaining)", state
+
+    surfaced_ids = state.get("surfaced_ids", [])
+    if item_id in surfaced_ids:
+        return False, "Already surfaced this session", state
+
+    updated = {
+        "surfaces_count": state.get("surfaces_count", 0) + 1,
+        "messages_since_last": 0,
+        "last_surface_time": now,
+        "surfaced_ids": list(surfaced_ids) + [item_id],
+    }
+    logger.info(f"[governance] Surfacing {item_id} (score={relevance_score:.2f}, surfaces={updated['surfaces_count']})")
+    json_handler.log_operation(
+        "governance_surface",
+        {"item_id": item_id, "relevance_score": relevance_score, "surfaces_count": updated["surfaces_count"]},
+    )
+    return True, "Ready to surface", updated
+
+
+# =============================================================================
+# MESSAGE TRACKING
+# =============================================================================
+
+
+def record_message(state: Dict[str, Any]) -> Dict[str, Any]:
+    """Record that a message was processed. Pure — returns updated state."""
+    return {
+        **state,
+        "messages_since_last": state.get("messages_since_last", 0) + 1,
+    }
@@ -164,7 +164,7 @@ def _save_manifest(manifest: Dict[str, str]) -> None:
 # =============================================================================


-def _embed_texts(texts: List[str]) -> dict:
+def _embed_texts(texts: List[str], timeout: int = 120) -> dict:
    """Encode texts via subprocess."""
    input_data = json.dumps({"texts": texts})
    try:
@@ -173,7 +173,7 @@ def _embed_texts(texts: List[str]) -> dict:
            input=input_data,
            capture_output=True,
            text=True,
-            timeout=120,
+            timeout=timeout,
        )
        if result.returncode != 0:
            return {"success": False, "error": result.stderr or "Embedding failed"}
@@ -219,24 +219,17 @@ def process_plans() -> Dict[str, Any]:
    """
    Process plan files from flow/processed_plans/ into vector storage.

-    Workflow:
-    1. Load config to find plans directory
-    2. Scan for unprocessed .md files
-    3. Chunk each file into sections
-    4. Embed all chunks via subprocess
-    5. Store vectors in ChromaDB
-    6. Update processed manifest
+    Processes each file independently so partial failure makes partial
+    progress — manifest is saved after every successful file.

    Returns:
        Dict with success, files_processed, total_chunks
    """
-    # Load config
    plans_config = config_loader.section("plans")

    if not plans_config.get("enabled", False):
        return {"success": True, "skipped": True, "reason": "plans disabled"}

-    # Resolve plans directory (relative to repo root)
    plans_dir = plans_config.get("path", ".backup/processed_plans")
    repo_root = _find_repo_root()
    plans_path = Path(plans_dir) if Path(plans_dir).is_absolute() else repo_root / plans_dir
@@ -246,7 +239,6 @@ def process_plans() -> Dict[str, Any]:
    if not plans_path.exists():
        return {"success": True, "files_processed": 0, "total_chunks": 0, "reason": "plans dir not found"}

-    # Get plan files
    files = []
    for ext in extensions:
        files.extend(plans_path.glob(f"*{ext}"))
@@ -254,7 +246,6 @@ def process_plans() -> Dict[str, Any]:
    if not files:
        return {"success": True, "files_processed": 0, "total_chunks": 0}

-    # Load manifest to skip already-processed files
    manifest = _load_manifest()
    unprocessed = [f for f in files if f.name not in manifest]

@@ -263,88 +254,73 @@ def process_plans() -> Dict[str, Any]:

    logger.info(f"[plans] Found {len(unprocessed)} unprocessed plan files")

-    errors = []
-
-    # -- Phase 1: Read all files, chunk them, collect texts + metadatas ----------
-    all_texts: List[str] = []
-    all_metadatas: List[Dict[str, str]] = []
-    # Track which files produced chunks (for manifest update)
-    files_with_chunks: List[Path] = []
-    # Files with 0 chunks still get marked in manifest (e.g. template content)
-    files_without_chunks: List[Path] = []
+    errors: List[str] = []
+    files_processed = 0
+    total_chunks = 0

    for plan_file in unprocessed:
        try:
            text = plan_file.read_text(encoding="utf-8")
        except Exception as e:
-            logger.warning(f"[plans_processor] Failed to read plan file {plan_file.name}: {e}")
+            logger.warning(f"[plans] Failed to read {plan_file.name}: {e}")
            errors.append(f"{plan_file.name}: read error: {e}")
            continue

        chunks = _chunk_plan_text(text, plan_file.name)
        if not chunks:
-            files_without_chunks.append(plan_file)
+            manifest[plan_file.name] = datetime.now().isoformat()
+            _save_manifest(manifest)
            continue

-        files_with_chunks.append(plan_file)
-        for c in chunks:
-            all_texts.append(c["text"])
-            all_metadatas.append(
-                {
-                    "source_file": plan_file.name,
-                    "section": c["section"],
-                    "processed_at": datetime.now().isoformat(),
-                    "type": "plan",
-                }
-            )
+        texts = [c["text"] for c in chunks]
+        metadatas = [
+            {
+                "source_file": plan_file.name,
+                "section": c["section"],
+                "processed_at": datetime.now().isoformat(),
+                "type": "plan",
+            }
+            for c in chunks
+        ]

-    total_chunks = len(all_texts)
-    files_processed = 0
-
-    # Mark empty-chunk files in manifest immediately (nothing to embed)
-    for plan_file in files_without_chunks:
-        manifest[plan_file.name] = datetime.now().isoformat()
-
-    # -- Phase 2: Batch embed + store (single subprocess each) ------------------
-    if all_texts:
-        logger.info(f"[plans] Batch embedding {total_chunks} chunks from {len(files_with_chunks)} files")
-
-        embed_result = _embed_texts(all_texts)
+        timeout = max(30, len(texts) * 3)
+        embed_result = _embed_texts(texts, timeout=timeout)
        if not embed_result.get("success"):
-            error_msg = f"batch embed error: {embed_result.get('error')}"
-            logger.error(f"[plans] {error_msg}")
-            errors.append(error_msg)
-        else:
-            embeddings = embed_result.get("embeddings", [])
-            if not embeddings:
-                errors.append("batch embed returned no embeddings")
-            else:
-                store_result = _store_vectors(embeddings, all_texts, all_metadatas, collection_name)
-                if not store_result.get("success"):
-                    error_msg = f"batch store error: {store_result.get('error')}"
-                    logger.error(f"[plans] {error_msg}")
-                    errors.append(error_msg)
-                else:
-                    # Success — mark all chunk-producing files in manifest
-                    for plan_file in files_with_chunks:
-                        manifest[plan_file.name] = datetime.now().isoformat()
-                    files_processed = len(files_with_chunks)
-                    logger.info(f"[plans] Batch complete: {files_processed} files, {total_chunks} chunks vectorized")
+            logger.warning(f"[plans] Embed failed for {plan_file.name}: {embed_result.get('error')}")
+            errors.append(f"{plan_file.name}: embed error: {embed_result.get('error')}")
+            continue

-    # Save manifest (includes empty-chunk files even if embedding failed)
-    _save_manifest(manifest)
+        embeddings = embed_result.get("embeddings", [])
+        if not embeddings:
+            errors.append(f"{plan_file.name}: embed returned no embeddings")
+            continue
+
+        store_result = _store_vectors(embeddings, texts, metadatas, collection_name)
+        if not store_result.get("success"):
+            logger.warning(f"[plans] Store failed for {plan_file.name}: {store_result.get('error')}")
+            errors.append(f"{plan_file.name}: store error: {store_result.get('error')}")
+            continue
+
+        manifest[plan_file.name] = datetime.now().isoformat()
+        _save_manifest(manifest)
+        files_processed += 1
+        total_chunks += len(texts)
+        logger.info(f"[plans] {plan_file.name}: {len(texts)} chunks vectorized")
+
+    if files_processed > 0:
+        logger.info(f"[plans] Complete: {files_processed} files, {total_chunks} chunks vectorized")

    result: Dict[str, Any] = {
-        "success": files_processed > 0 or (not errors and not files_with_chunks),
+        "success": files_processed > 0 or not errors,
        "files_processed": files_processed,
-        "total_chunks": total_chunks if files_processed > 0 else 0,
+        "total_chunks": total_chunks,
    }
    if errors:
        result["errors"] = errors

    json_handler.log_operation(
        "process_plans",
-        {"files_processed": files_processed, "total_chunks": result["total_chunks"], "success": result["success"]},
+        {"files_processed": files_processed, "total_chunks": total_chunks, "success": result["success"]},
    )

    return result
@@ -18,6 +18,7 @@ Purpose:
    layer to satisfy thin-module standard.
 """

+import re
 import subprocess
 import json
 import os
@@ -210,6 +211,77 @@ def _filter_results(results: list, n_results: int) -> list:
    return filtered


+# =============================================================================
+# PLAN-ID EXACT MATCHING
+# =============================================================================
+
+_PLAN_ID_RE = re.compile(
+    r"(?:^|\b)((?:d|f|p|td|a)plan)[\s\-_]*(\d{3,5})\b",
+    re.IGNORECASE,
+)
+
+
+def _extract_plan_id(query: str) -> str | None:
+    """Extract a normalized plan ID (e.g. 'FPLAN-0332') from a query string."""
+    m = _PLAN_ID_RE.search(query)
+    if not m:
+        return None
+    prefix = m.group(1).upper()
+    number = m.group(2)
+    return f"{prefix}-{number}"
+
+
+def _fetch_plan_by_metadata(plan_id: str, n_results: int) -> list:
+    """Fetch plan chunks directly from ChromaDB by source_file metadata."""
+    input_data = {
+        "operation": "get_by_source",
+        "collection_name": "flow_plans",
+        "source_pattern": plan_id,
+        "n_results": n_results,
+    }
+    try:
+        result = subprocess.run(
+            [str(MEMORY_PYTHON), str(CHROMA_SUBPROCESS_SCRIPT)],
+            input=json.dumps(input_data),
+            capture_output=True,
+            text=True,
+            timeout=30,
+        )
+        if result.returncode != 0:
+            return []
+        data = json.loads(result.stdout)
+        if not data.get("success"):
+            return []
+        return data.get("results", [])
+    except Exception as e:
+        logger.warning(f"[search] Plan metadata fetch failed: {e}")
+        return []
+
+
+def _pin_plan_id_matches(query: str, filtered: list, n_results: int) -> list:
+    """Pin exact plan-ID matches to the top of results.
+
+    If the query contains a plan-ID pattern, fetch matching chunks directly
+    from ChromaDB metadata (bypassing embedding similarity) and pin them.
+    """
+    plan_id = _extract_plan_id(query)
+    if not plan_id:
+        return filtered
+
+    exact = _fetch_plan_by_metadata(plan_id, n_results)
+    if not exact:
+        return filtered
+
+    for r in exact:
+        r["similarity"] = 1.0
+
+    logger.info(f"[search] Pinned {len(exact)} exact matches for {plan_id}")
+
+    seen_ids = {r.get("id") for r in exact}
+    rest = [r for r in filtered if r.get("id") not in seen_ids]
+    return (exact + rest)[:n_results]
+
+
 # =============================================================================
 # PUBLIC API
 # =============================================================================
@@ -273,6 +345,9 @@ def execute_search(
    # Step 3: Filter and score results
    filtered_results = _filter_results(raw_results, n_results)

+    # Step 4: Pin exact plan-ID matches to the top
+    filtered_results = _pin_plan_id_matches(query, filtered_results, n_results)
+
    logger.info(f"[search] Filtered to {len(filtered_results)} relevant results")

    json_handler.log_operation(
@@ -67,12 +67,30 @@ def _store_vectors(branch, memory_type, embeddings, documents, metadatas, db_pat
        embedding_function=None,
    )

-    # Content-hash IDs prevent duplicates across rollover runs
-    ids = [f"{branch}_{memory_type}_{hashlib.sha256(doc.encode()).hexdigest()[:16]}" for doc in documents]
+    # Content-hash IDs — idempotent across runs.  When metadata carries
+    # source_file (e.g. plans intake), salt the hash so identical boilerplate
+    # from different files gets distinct IDs and per-file provenance survives.
+    ids = []
+    for doc, meta in zip(documents, metadatas):
+        salt = meta.get("source_file", "") if isinstance(meta, dict) else ""
+        hash_input = f"{salt}:{doc}" if salt else doc
+        ids.append(f"{branch}_{memory_type}_{hashlib.sha256(hash_input.encode()).hexdigest()[:16]}")

    # Chroma expects lists, not numpy arrays
    embeddings_list = [emb.tolist() if hasattr(emb, "tolist") else emb for emb in embeddings]

+    # Safety net: deduplicate within batch — ChromaDB rejects non-unique IDs
+    # in a single upsert call.
+    seen = {}
+    for i, doc_id in enumerate(ids):
+        seen[doc_id] = i
+    if len(seen) < len(ids):
+        unique_indices = sorted(seen.values())
+        ids = [ids[i] for i in unique_indices]
+        embeddings_list = [embeddings_list[i] for i in unique_indices]
+        documents = [documents[i] for i in unique_indices]
+        metadatas = [metadatas[i] for i in unique_indices]
+
    # Upsert: idempotent — same content gets same ID, no duplicates
    collection.upsert(embeddings=embeddings_list, documents=documents, metadatas=metadatas, ids=ids)

@@ -134,6 +152,79 @@ def _check_plan(plan_label, db_path=None):
    return {"success": True, "found": match_count > 0, "count": match_count, "source_files": sorted(matching_files)}


+def _get_by_source(collection_name, source_pattern, n_results=5, db_path=None):
+    """Fetch documents whose source_file metadata contains a pattern.
+
+    Args:
+        collection_name: Name of the ChromaDB collection
+        source_pattern: Substring to match in source_file metadata
+        n_results: Maximum number of results to return
+        db_path: Optional path to Chroma database
+
+    Returns:
+        Dict with success, results list (document, metadata, id)
+    """
+    client = _get_client(db_path)
+
+    try:
+        collection = client.get_collection(collection_name, embedding_function=None)
+    except Exception as e:
+        logger.warning(f"[chroma_subprocess] Collection '{collection_name}' not found in get_by_source: {e}")
+        return {"success": False, "error": f"Collection '{collection_name}' not found: {e}"}
+
+    result = collection.get(include=["metadatas", "documents"])
+    matches = []
+    for i, meta in enumerate(result.get("metadatas", [])):
+        source = meta.get("source_file", "")
+        if source_pattern in source:
+            matches.append(
+                {
+                    "collection": collection_name,
+                    "document": result["documents"][i],
+                    "metadata": meta,
+                    "id": result["ids"][i],
+                    "distance": 0.0,
+                }
+            )
+            if len(matches) >= n_results:
+                break
+
+    return {"success": True, "results": matches, "count": len(matches)}
+
+
+def _delete_by_source(collection_name, source_pattern, db_path=None):
+    """Delete vectors whose source_file metadata contains a pattern.
+
+    Args:
+        collection_name: Name of the ChromaDB collection
+        source_pattern: Substring to match in source_file metadata
+        db_path: Optional path to Chroma database
+
+    Returns:
+        Dict with success, deleted count, and matched IDs
+    """
+    client = _get_client(db_path)
+
+    try:
+        collection = client.get_collection(collection_name, embedding_function=None)
+    except Exception as e:
+        logger.warning(f"[chroma_subprocess] Collection '{collection_name}' not found in delete_by_source: {e}")
+        return {"success": False, "error": f"Collection '{collection_name}' not found: {e}"}
+
+    result = collection.get(include=["metadatas"])
+    ids_to_delete = []
+    for i, meta in enumerate(result.get("metadatas", [])):
+        source = meta.get("source_file", "")
+        if source_pattern in source:
+            ids_to_delete.append(result["ids"][i])
+
+    if not ids_to_delete:
+        return {"success": True, "deleted": 0, "ids": [], "message": "No matching vectors found"}
+
+    collection.delete(ids=ids_to_delete)
+    return {"success": True, "deleted": len(ids_to_delete), "ids": ids_to_delete}
+
+
 def _search_vectors(query_embedding, branch=None, memory_type=None, n_results=5, db_path=None):
    """Search for similar vectors."""
    client = _get_client(db_path)
@@ -214,6 +305,19 @@ def main():
            )
        elif operation == "check_plan":
            result = _check_plan(plan_label=input_data.get("plan_label"), db_path=input_data.get("db_path"))
+        elif operation == "get_by_source":
+            result = _get_by_source(
+                collection_name=input_data.get("collection_name"),
+                source_pattern=input_data.get("source_pattern"),
+                n_results=input_data.get("n_results", 5),
+                db_path=input_data.get("db_path"),
+            )
+        elif operation == "delete_by_source":
+            result = _delete_by_source(
+                collection_name=input_data.get("collection_name"),
+                source_pattern=input_data.get("source_pattern"),
+                db_path=input_data.get("db_path"),
+            )
        else:
            result = {"success": False, "error": f"Unknown operation: {operation}"}

@@ -0,0 +1,80 @@
+# =================== AIPass ====================
+# Name: governance.py
+# Description: Surfacing governance module — public API
+# Version: 1.1.0
+# Created: 2026-07-16
+# Modified: 2026-07-16
+# =============================================
+
+"""
+Surfacing Governance Module — Public API
+
+Thin module re-exporting governance engine from handlers/governance/engine.py.
+Cross-branch consumers import from here:
+
+    from aipass.memory.apps.modules.governance import should_surface, record_message, new_state
+"""
+
+import os
+import sys
+
+if sys.platform == "win32":
+    os.environ.setdefault("PYTHONUTF8", "1")
+    for _stream in (sys.stdout, sys.stderr):
+        _reconfigure = getattr(_stream, "reconfigure", None)
+        if _reconfigure is not None:
+            _reconfigure(encoding="utf-8", errors="replace")
+
+from aipass.prax import logger  # noqa: F401
+from aipass.memory.apps.handlers.json import json_handler
+from aipass.memory.apps.handlers.governance.engine import (
+    DEFAULT_CONFIG,
+    new_state,
+    record_message,
+    should_surface,
+)
+
+__all__ = ["should_surface", "record_message", "new_state", "DEFAULT_CONFIG"]
+
+
+# =============================================================================
+# MODULE ROUTING (handle_command for drone auto-discovery)
+# =============================================================================
+
+
+def print_introspection() -> None:
+    """Display module introspection (seedgo standard)."""
+    from aipass.cli.apps.modules import console
+
+    console.print()
+    console.print("[bold cyan]governance Module[/bold cyan]")
+    console.print("Pure surfacing governance — state-in/state-out decision functions")
+    console.print()
+    console.print("[yellow]Public API:[/yellow]")
+    console.print("  should_surface(item_id, relevance_score, state, config)")
+    console.print("  record_message(state)")
+    console.print("  new_state()")
+    console.print()
+    console.print("[dim]Library module — import from: aipass.memory.apps.modules.governance[/dim]")
+
+
+def handle_command(command: str, args: list) -> bool:
+    """Entry point for drone module discovery — governance has no CLI surface."""
+    if command != "governance":
+        return False
+
+    json_handler.log_operation("governance_command", {"args": args})
+
+    if not args:
+        print_introspection()
+        return True
+
+    if args[0] in ("--help", "-h", "help"):
+        print_introspection()
+        return True
+
+    from aipass.cli.apps.modules import warning
+
+    warning(f"governance: unknown subcommand '{args[0]}'")
+    print_introspection()
+    return True
@@ -0,0 +1,252 @@
+"""Tests for the surfacing governance module (pure state-in/state-out API)."""
+
+import pytest
+
+from aipass.memory.apps.modules.governance import (
+    DEFAULT_CONFIG,
+    new_state,
+    record_message,
+    should_surface,
+)
+
+
+# =============================================================================
+# new_state
+# =============================================================================
+
+
+class TestNewState:
+    def test_returns_zeroed_state(self):
+        state = new_state()
+        assert state["surfaces_count"] == 0
+        assert state["messages_since_last"] == 0
+        assert state["last_surface_time"] == 0.0
+        assert state["surfaced_ids"] == []
+
+    def test_returns_independent_copies(self):
+        s1 = new_state()
+        s2 = new_state()
+        s1["surfaced_ids"].append("x")
+        assert s2["surfaced_ids"] == []
+
+
+# =============================================================================
+# should_surface — rejection paths
+# =============================================================================
+
+
+class TestShouldSurfaceRejections:
+    """Each rejection path tested independently with an otherwise-valid state."""
+
+    @pytest.fixture()
+    def ready_state(self):
+        """State that passes all checks when config is default."""
+        return {
+            "surfaces_count": 0,
+            "messages_since_last": 20,
+            "last_surface_time": 0.0,
+            "surfaced_ids": [],
+        }
+
+    def test_disabled(self, ready_state):
+        ok, reason, st = should_surface("item1", 0.8, ready_state, {"enabled": False}, current_time=1000.0)
+        assert ok is False
+        assert "disabled" in reason.lower()
+        assert st is ready_state
+
+    def test_below_threshold(self, ready_state):
+        ok, reason, st = should_surface("item1", 0.1, ready_state, {"threshold": 0.3}, current_time=1000.0)
+        assert ok is False
+        assert "threshold" in reason.lower()
+        assert st is ready_state
+
+    def test_budget_exhausted(self, ready_state):
+        ready_state["surfaces_count"] = 5
+        ok, reason, st = should_surface("item1", 0.8, ready_state, {"max_surfaces_per_session": 5}, current_time=1000.0)
+        assert ok is False
+        assert "budget" in reason.lower() or "exhausted" in reason.lower()
+        assert st is ready_state
+
+    def test_spacing_not_met(self, ready_state):
+        ready_state["messages_since_last"] = 3
+        ready_state["last_surface_time"] = 1.0
+        ok, reason, st = should_surface("item1", 0.8, ready_state, {"min_messages_between": 10}, current_time=1000.0)
+        assert ok is False
+        assert "spacing" in reason.lower()
+        assert st is ready_state
+
+    def test_cooldown_active(self, ready_state):
+        ready_state["last_surface_time"] = 900.0
+        ok, reason, st = should_surface("item1", 0.8, ready_state, {"cooldown_seconds": 300}, current_time=1000.0)
+        assert ok is False
+        assert "cooldown" in reason.lower()
+        assert st is ready_state
+
+    def test_already_surfaced(self, ready_state):
+        ready_state["surfaced_ids"] = ["item1"]
+        ok, reason, st = should_surface("item1", 0.8, ready_state, current_time=1000.0)
+        assert ok is False
+        assert "already" in reason.lower()
+        assert st is ready_state
+
+
+# =============================================================================
+# should_surface — happy path
+# =============================================================================
+
+
+class TestShouldSurfaceHappy:
+    def test_surfaces_and_returns_updated_state(self):
+        state = {
+            "surfaces_count": 0,
+            "messages_since_last": 15,
+            "last_surface_time": 0.0,
+            "surfaced_ids": [],
+        }
+        ok, reason, updated = should_surface("compass-42", 0.75, state, current_time=5000.0)
+        assert ok is True
+        assert "ready" in reason.lower()
+        assert updated["surfaces_count"] == 1
+        assert updated["messages_since_last"] == 0
+        assert updated["last_surface_time"] == 5000.0
+        assert "compass-42" in updated["surfaced_ids"]
+
+    def test_does_not_mutate_input_state(self):
+        state = {
+            "surfaces_count": 0,
+            "messages_since_last": 15,
+            "last_surface_time": 0.0,
+            "surfaced_ids": [],
+        }
+        original_ids = state["surfaced_ids"]
+        should_surface("item1", 0.8, state, current_time=5000.0)
+        assert state["surfaces_count"] == 0
+        assert state["surfaced_ids"] is original_ids
+        assert len(original_ids) == 0
+
+    def test_threshold_boundary_exact(self):
+        state = new_state()
+        state["messages_since_last"] = 10
+        ok, _, _ = should_surface("x", 0.3, state, {"threshold": 0.3}, current_time=1000.0)
+        assert ok is True
+
+    def test_cooldown_expired(self):
+        state = {
+            "surfaces_count": 0,
+            "messages_since_last": 15,
+            "last_surface_time": 500.0,
+            "surfaced_ids": [],
+        }
+        ok, _, _ = should_surface("x", 0.8, state, {"cooldown_seconds": 300}, current_time=801.0)
+        assert ok is True
+
+    def test_first_surface_ignores_spacing(self):
+        """Fresh session: first prompt with high relevance surfaces immediately."""
+        state = new_state()
+        state["messages_since_last"] = 1
+        ok, reason, updated = should_surface("compass-1", 0.8, state, current_time=100.0)
+        assert ok is True
+        assert "ready" in reason.lower()
+
+        for i in range(9):
+            updated = record_message(updated)
+            ok, reason, _ = should_surface("compass-2", 0.8, updated, current_time=100.0 + 400 + i)
+            assert ok is False
+            assert "spacing" in reason.lower()
+
+        updated = record_message(updated)
+        ok, _, _ = should_surface("compass-2", 0.8, updated, current_time=600.0)
+        assert ok is True
+
+
+# =============================================================================
+# State isolation — two independent states do not bleed
+# =============================================================================
+
+
+class TestStateIsolation:
+    def test_two_states_independent(self):
+        s1 = new_state()
+        s1["messages_since_last"] = 20
+        s2 = new_state()
+        s2["messages_since_last"] = 20
+
+        ok1, _, s1_updated = should_surface("a", 0.8, s1, current_time=1000.0)
+        ok2, _, s2_updated = should_surface("b", 0.9, s2, current_time=2000.0)
+
+        assert ok1 is True
+        assert ok2 is True
+        assert s1_updated["surfaced_ids"] == ["a"]
+        assert s2_updated["surfaced_ids"] == ["b"]
+        assert s1_updated["last_surface_time"] == 1000.0
+        assert s2_updated["last_surface_time"] == 2000.0
+
+    def test_chained_surfaces_accumulate(self):
+        state = new_state()
+        state["messages_since_last"] = 20
+
+        ok, _, state = should_surface("a", 0.8, state, current_time=1000.0)
+        assert ok is True
+        assert state["surfaces_count"] == 1
+
+        state["messages_since_last"] = 20
+        ok, _, state = should_surface("b", 0.7, state, current_time=2000.0)
+        assert ok is True
+        assert state["surfaces_count"] == 2
+        assert state["surfaced_ids"] == ["a", "b"]
+
+
+# =============================================================================
+# record_message
+# =============================================================================
+
+
+class TestRecordMessage:
+    def test_increments_counter(self):
+        state = new_state()
+        updated = record_message(state)
+        assert updated["messages_since_last"] == 1
+
+    def test_does_not_mutate_input(self):
+        state = new_state()
+        record_message(state)
+        assert state["messages_since_last"] == 0
+
+    def test_preserves_other_fields(self):
+        state = {
+            "surfaces_count": 3,
+            "messages_since_last": 5,
+            "last_surface_time": 100.0,
+            "surfaced_ids": ["x"],
+        }
+        updated = record_message(state)
+        assert updated["surfaces_count"] == 3
+        assert updated["messages_since_last"] == 6
+        assert updated["last_surface_time"] == 100.0
+        assert updated["surfaced_ids"] == ["x"]
+
+
+# =============================================================================
+# Config merging
+# =============================================================================
+
+
+class TestConfigMerging:
+    def test_none_config_uses_defaults(self):
+        state = new_state()
+        state["messages_since_last"] = 20
+        ok, _, _ = should_surface("x", 0.8, state, None, current_time=1000.0)
+        assert ok is True
+
+    def test_partial_config_merges_with_defaults(self):
+        state = new_state()
+        state["messages_since_last"] = 20
+        ok, reason, _ = should_surface("x", 0.25, state, {"threshold": 0.5}, current_time=1000.0)
+        assert ok is False
+        assert "threshold" in reason.lower()
+
+    def test_default_config_values_match(self):
+        assert DEFAULT_CONFIG["threshold"] == 0.3
+        assert DEFAULT_CONFIG["max_surfaces_per_session"] == 5
+        assert DEFAULT_CONFIG["min_messages_between"] == 10
+        assert DEFAULT_CONFIG["cooldown_seconds"] == 300
@@ -501,7 +501,7 @@ class TestProcessPlans:
        monkeypatch.setattr(
            mod,
            "_embed_texts",
-            lambda texts: {"success": True, "embeddings": [[0.1, 0.2]] * len(texts)},
+            lambda texts, timeout=120: {"success": True, "embeddings": [[0.1, 0.2]] * len(texts)},
        )
        monkeypatch.setattr(
            mod,
@@ -547,7 +547,7 @@ class TestProcessPlans:
        monkeypatch.setattr(
            mod,
            "_embed_texts",
-            lambda texts: {"success": False, "error": "GPU out of memory"},
+            lambda texts, timeout=120: {"success": False, "error": "GPU out of memory"},
        )

        mock_jh = MagicMock()
@@ -584,7 +584,7 @@ class TestProcessPlans:
        monkeypatch.setattr(
            mod,
            "_embed_texts",
-            lambda texts: {"success": True, "embeddings": []},
+            lambda texts, timeout=120: {"success": True, "embeddings": []},
        )

        mock_jh = MagicMock()
@@ -620,7 +620,7 @@ class TestProcessPlans:
        monkeypatch.setattr(
            mod,
            "_embed_texts",
-            lambda texts: {"success": True, "embeddings": [[0.1, 0.2]] * len(texts)},
+            lambda texts, timeout=120: {"success": True, "embeddings": [[0.1, 0.2]] * len(texts)},
        )
        monkeypatch.setattr(
            mod,
@@ -69,6 +69,34 @@
      "subject": "STATUS decommission — prax dashboard todo_count + engine dormant",
      "date_closed": "2026-06-08",
      "location": "prax"
+    },
+    {
+      "plan_id": "FPLAN-0321",
+      "type": "FPLAN",
+      "subject": "Pytest log routing: detect test context, route away from production logs",
+      "date_closed": "2026-07-14",
+      "location": "prax"
+    },
+    {
+      "plan_id": "FPLAN-0323",
+      "type": "FPLAN",
+      "subject": "Remove single-instance lock from monitor display path — concurrent viewers",
+      "date_closed": "2026-07-14",
+      "location": "prax"
+    },
+    {
+      "plan_id": "FPLAN-0325",
+      "type": "FPLAN",
+      "subject": "TG relay send backoff: offline mode + log-once on network failures",
+      "date_closed": "2026-07-14",
+      "location": "prax"
+    },
+    {
+      "plan_id": "TDPLAN-0013",
+      "type": "TDPLAN",
+      "subject": "DPLAN-0242 build: runaway-log detection and escalation",
+      "date_closed": "2026-07-15",
+      "location": "prax"
    }
  ],
  "document_metadata": {
@@ -5,7 +5,7 @@
 **Purpose:** System-wide logging, real-time monitoring, and dashboard infrastructure for AIPass.
 **Module:** `aipass.prax`
 **Version:** 2.0.0
-**Last Updated:** 2026-06-05
+**Last Updated:** 2026-07-14

 ---

@@ -51,11 +51,23 @@ Real-time unified console showing:
 - **Caller attribution** — `CALLER → TARGET` for drone commands
 - **Model tags** — `[BRANCH/model]` (e.g., `[DEVPULSE/opus]`, `[DEVPULSE/gpt-5.4]`)
 - **Multi-CLI** — Claude Code (JSONL), Codex (JSONL) session monitoring
+- **Rate tracking** — 4th background thread scans `system_logs/` for runaway log growth every 10s
 - **Polling fallback** — automatic fallback when inotify watches are exhausted
 - **Soft start** — only shows new activity after launch (seeks to EOF on startup)

 Interactive commands inside the monitor: `help`, `status`, `quit`/`exit`.

+### Log Health
+
+```bash
+drone @prax log-health                   # Show module info
+drone @prax log-health scan              # Scan all log files, show current growth rates
+drone @prax log-health snapshot          # Show last known rates (no new scan)
+drone @prax log-health --help            # Log health usage
+```
+
+Quick overview of log file growth rates across `system_logs/`. Powered by the rate tracker handler — `scan` runs a fresh measurement, `snapshot` reads the last persisted state without scanning.
+
 ### Status

 ```bash
@@ -123,12 +135,13 @@ prax/
 ├── __init__.py                        # Public API: exports `logger` (NullLogger fallback)
 ├── apps/
 │   ├── prax.py                        # Entry point — auto-discovers modules, routes commands
-│   ├── modules/                       # Business logic (5 command modules)
+│   ├── modules/                       # Business logic (6 command modules)
 │   │   ├── logger.py                  # SystemLogger — auto-routing, two-tier logging
-│   │   ├── monitor.py                 # Mission Control — 3-thread real-time monitoring
+│   │   ├── monitor.py                 # Mission Control — 4-thread real-time monitoring
 │   │   ├── dashboard.py               # Dashboard — template management, refresh, write-through
 │   │   ├── status.py                  # System status — health display (STATUS.md sync dormant)
-│   │   └── log_audit.py              # Log audit — scan, health summary, enforce limits
+│   │   ├── log_audit.py              # Log audit — scan, health summary, enforce limits
+│   │   └── log_health.py             # Log health — rate overview (scan/snapshot)
 │   └── handlers/                      # Implementation details (11 handler directories)
 │       ├── central/                   # Central file reader (.ai_central/*.central.json)
 │       ├── config/                    # Path resolution, log config, ignore patterns
@@ -137,13 +150,13 @@ prax/
 │       ├── json/                      # Auto-creating JSON handler (config/data/log per module)
 │       ├── json_templates/            # Default JSON templates for auto-creation
 │       ├── logging/                   # Setup, rotation, introspection, override, direct logger
-│       ├── monitoring/                # Event queue, branch detector, stream output, log watcher
+│       ├── monitoring/                # Event queue, branch detector, stream output, log watcher, rate tracker
 │       ├── registry/                  # Module registry load/save
 │       ├── status/                    # STATUS.md sync handler (dormant — TDPLAN-0007)
 │       └── watcher/                   # Background system watchers
 ├── prax_json/                         # Auto-created per-module config/data/log files
 ├── templates/                         # Dashboard template schema (DASHBOARD.template.json)
-└── tests/                             # 901 tests across 19 files
+└── tests/                             # 1028 tests across 20 files
 ```

 ### Design Pattern
@@ -164,14 +177,15 @@ drone @prax monitor run
 1. **Auto-routing** — `logger.info()` inspects the call stack to identify the caller's module, branch, and file path, then routes the log entry to the correct per-module log file.
 2. **Two-tier logging** — Each log entry goes to both `system_logs/` (central, all branches) and `<branch>/logs/` (branch-local), both with size-based rotation.
 3. **Self-healing** — Auto-creates missing log directories, falls back to `system_logs/external/` for unknown modules, provides NullLogger if prax itself fails to import.
-4. **Mission Control** — Three threads: display worker (pulls from event queue), file watcher (watchdog on branch `apps/` dirs), log watcher (tails `system_logs/*.log`). Falls back to polling when inotify is exhausted.
+4. **Mission Control** — Four threads: display worker (pulls from event queue), file watcher (watchdog on branch `apps/` dirs), log watcher (tails `system_logs/*.log`), rate tracker (scans `system_logs/` for runaway growth every 10s). Falls back to polling when inotify is exhausted.
 5. **Multi-CLI monitoring** — Watches Claude Code JSONL and Codex JSONL session files. Extracts agent activity (thinking, tool use, responses) with model detection and branch resolution.
-6. **Dashboard** — Template-based per-branch dashboard files. Refreshes from central files (`*.central.json`). Write-through API for services to update sections directly.
-7. **STATUS sync** — *(Dormant — TDPLAN-0007)* Previously scanned all branch `STATUS.local.md` files and built aggregated `STATUS.md`. Engine code intact but no longer triggered.
+6. **Runaway-log detection** — Rate tracker measures byte growth per log file, estimates lines/min from byte deltas. Sustained thresholds: WARNING (>100 lines/min for 2 min), CRITICAL (>10 lines/sec for 1 min). Fires `runaway_log_detected` on the trigger event bus. State persists to disk across process restarts. Per-file suppression available.
+7. **Dashboard** — Template-based per-branch dashboard files. Refreshes from central files (`*.central.json`). Write-through API for services to update sections directly.
+8. **STATUS sync** — *(Dormant — TDPLAN-0007)* Previously scanned all branch `STATUS.local.md` files and built aggregated `STATUS.md`. Engine code intact but no longer triggered.

 ## Tests

-901 tests across 19 files, covering all major components:
+1028 tests across 20 files, covering all major components:

 | Test File | Tests | Coverage |
 |-----------|-------|----------|
@@ -179,7 +193,7 @@ drone @prax monitor run
 | test_monitoring_handlers.py | 139 | Branch detector, stream output, event handling |
 | test_operations.py | 99 | Dashboard operations, write-through |
 | test_log_watcher.py | 82 | Log file tailing, agent activity parsing |
-| test_monitor_module.py | 73 | Monitor commands, thread lifecycle |
+| test_monitor_module.py | 73 | Monitor commands, thread lifecycle (4-thread) |
 | test_logging_handlers.py | 41 | Setup, rotation, introspection, direct logger |
 | test_logging.py | 41 | Core logging system |
 | test_logger_module.py | 40 | Logger init, routing, lifecycle |
@@ -193,6 +207,7 @@ drone @prax monitor run
 | test_central.py | 14 | Central reader |
 | test_devpulse_dashboard_plugin.py | 12 | Dashboard plugin (git, session, dispatch) |
 | test_log_audit.py | 10 | Log audit |
+| test_rate_tracker.py | 21 | Rate tracking, thresholds, persistence, suppression |
 | test_status.py | 8 | Status commands |

 ## Integration Points
@@ -216,7 +231,7 @@ drone @prax monitor run

 ---

-*Last Updated: 2026-06-05*
+*Last Updated: 2026-07-14*

 ---
 [← Back to AIPass](../../../README.md)
@@ -31,13 +31,14 @@ import inspect
 import json
 import logging
 import os
-
-logger = logging.getLogger(__name__)
+import tempfile
 from pathlib import Path
 from typing import Dict, Any, Optional

 from aipass.prax.apps.handlers.json import json_handler

+logger = logging.getLogger(__name__)
+
 # =============================================
 # CONFIGURATION
 # =============================================
@@ -75,6 +76,10 @@ def get_system_logs_dir() -> Path:
        p = Path(test_log_dir) / "system"
        p.mkdir(parents=True, exist_ok=True)
        return p
+    if os.environ.get("PYTEST_CURRENT_TEST"):
+        p = Path(tempfile.gettempdir()) / "aipass_test_logs" / "system"
+        p.mkdir(parents=True, exist_ok=True)
+        return p
    global _system_logs_dir_cache
    if _system_logs_dir_cache is None:
        repo_root = _find_repo_root()
@@ -133,6 +138,10 @@ def get_module_logs_dir(module_name: Optional[str] = None) -> Path:
        p = Path(test_log_dir) / module_name
        p.mkdir(parents=True, exist_ok=True)
        return p
+    if os.environ.get("PYTEST_CURRENT_TEST"):
+        p = Path(tempfile.gettempdir()) / "aipass_test_logs" / module_name
+        p.mkdir(parents=True, exist_ok=True)
+        return p

    # Standard: src/aipass/{module}/logs
    branch_dir = ECOSYSTEM_ROOT / module_name
@@ -6,13 +6,14 @@
 # Modified: 2026-07-10
 # =============================================

-"""Single-instance lock for the prax monitor.
+"""Relay-scoped lock for the prax monitor Telegram relay.

-Prevents duplicate monitor processes from running concurrently (and
-double-sending Telegram relay messages).  Uses a pidfile with liveness
-check — cross-platform (Linux / macOS / Windows).
+Prevents duplicate Telegram sends when multiple monitor viewers run
+concurrently. The display path is never blocked — only the TG relay
+acquires this lock, so interactive viewers always start.

-Lock file lives in prax_json/monitor.pid (outside system_logs/ to avoid
+Uses a pidfile with liveness check — cross-platform (Linux / macOS / Windows).
+Lock file lives in prax_json/relay.pid (outside system_logs/ to avoid
 the tailed-directory feedback loop).
 """

@@ -82,14 +83,14 @@ def _is_pid_alive(pid: int) -> bool:


 def get_lock_path() -> Path:
-    """Return the path for the monitor single-instance lock file."""
+    """Return the path for the relay lock file."""
    if _lock_path_override is not None:
        return _lock_path_override
-    return Path(__file__).resolve().parent.parent.parent / "prax_json" / "monitor.pid"
+    return Path(__file__).resolve().parent.parent.parent / "prax_json" / "relay.pid"


-def acquire(error_fn=None) -> None:
-    """Acquire single-instance lock. Raises SystemExit(1) if another live instance holds it."""
+def try_acquire() -> bool:
+    """Try to acquire the relay lock. Returns True if acquired, False if held by a live process."""
    global _held_lock
    lock_path = get_lock_path()
    json_handler.log_operation("instance_lock_acquire", {"pid": os.getpid()})
@@ -99,11 +100,8 @@ def acquire(error_fn=None) -> None:
            data = _json.loads(lock_path.read_text(encoding="utf-8"))
            existing_pid = data.get("pid", 0)
            if existing_pid and _is_pid_alive(existing_pid):
-                msg = f"Monitor already running (PID {existing_pid}). Kill the existing process or remove {lock_path}"
-                if error_fn:
-                    error_fn(msg)
-                logger.error("[instance_lock] %s", msg)
-                raise SystemExit(1)
+                logger.info("[instance_lock] Relay lock held by PID %d — skipping TG relay", existing_pid)
+                return False
            logger.info("[instance_lock] Reclaiming stale lock (PID %d is dead)", existing_pid)
        except (ValueError, OSError) as exc:
            logger.info("[instance_lock] Removing corrupt lock file: %s", exc)
@@ -111,7 +109,8 @@ def acquire(error_fn=None) -> None:
    lock_path.parent.mkdir(parents=True, exist_ok=True)
    lock_path.write_text(_json.dumps({"pid": os.getpid()}), encoding="utf-8")
    _held_lock = lock_path
-    logger.info("[instance_lock] Acquired (PID %d)", os.getpid())
+    logger.info("[instance_lock] Acquired relay lock (PID %d)", os.getpid())
+    return True


 def release() -> None:
@@ -0,0 +1,378 @@
+# =================== AIPass ====================
+# Name: rate_tracker.py
+# Description: Log file rate tracking for runaway detection
+# Version: 1.2.0
+# Created: 2026-07-14
+# Modified: 2026-07-15
+# =============================================
+
+"""
+Rate Tracker — volume-based runaway-log detection.
+
+Tracks byte growth rate per log file in system_logs/. When a file sustains
+abnormal growth (lines/min above threshold for consecutive intervals), fires
+a ``runaway_log_detected`` event on the trigger event bus.
+
+Orthogonal to medic's content-based ERROR/CRITICAL detection — this catches
+rate regardless of log level.
+
+State persists to ``prax_json/rate_tracker_data.json`` so that rates survive
+across process restarts and CLI invocations can display meaningful data.
+"""
+
+import time
+from collections import deque
+from pathlib import Path
+from typing import Dict, Optional
+
+from aipass.prax.apps.modules.logger import get_direct_logger
+from aipass.prax.apps.handlers.json import json_handler
+from aipass.prax.apps.handlers.monitoring.branch_detector import detect_branch_from_log
+
+logger = get_direct_logger()
+
+SCAN_INTERVAL = 10.0
+AVG_LINE_BYTES = 120
+
+WARNING_LINES_PER_MIN = 100
+WARNING_SUSTAINED_INTERVALS = 12  # 12 * 10s = 2 min
+
+CRITICAL_LINES_PER_MIN = 600  # 10/sec * 60
+CRITICAL_SUSTAINED_INTERVALS = 6  # 6 * 10s = 1 min
+
+_RATE_HISTORY_SIZE = 30
+_WINDOW_INTERVALS = 6
+
+_DATA_FILE = "rate_tracker"
+
+
+class FileRateState:
+    """Per-file tracking state."""
+
+    __slots__ = (
+        "last_offset",
+        "last_check",
+        "rates",
+        "warning_sustained",
+        "critical_sustained",
+        "fired_warning",
+        "fired_critical",
+    )
+
+    def __init__(self, offset: int, now: float) -> None:
+        self.last_offset: int = offset
+        self.last_check: float = now
+        self.rates: deque = deque(maxlen=_RATE_HISTORY_SIZE)
+        self.warning_sustained: int = 0
+        self.critical_sustained: int = 0
+        self.fired_warning: bool = False
+        self.fired_critical: bool = False
+
+    def to_dict(self) -> dict:
+        """Serialize to a dict for disk persistence."""
+        return {
+            "last_offset": self.last_offset,
+            "last_check": self.last_check,
+            "warning_sustained": self.warning_sustained,
+            "critical_sustained": self.critical_sustained,
+            "fired_warning": self.fired_warning,
+            "fired_critical": self.fired_critical,
+        }
+
+    @classmethod
+    def from_dict(cls, d: dict) -> "FileRateState":
+        """Restore from a persisted dict."""
+        state = cls(d.get("last_offset", 0), d.get("last_check", 0.0))
+        state.warning_sustained = d.get("warning_sustained", 0)
+        state.critical_sustained = d.get("critical_sustained", 0)
+        state.fired_warning = d.get("fired_warning", False)
+        state.fired_critical = d.get("fired_critical", False)
+        return state
+
+
+_tracked: Dict[str, FileRateState] = {}
+
+_suppressed_files: set = set()
+
+_logs_dir: Optional[Path] = None
+
+_EVENT_CALLBACK = None
+
+_state_loaded: bool = False
+
+
+def configure(
+    logs_dir: Optional[Path] = None,
+    event_callback=None,
+    suppressed_files: Optional[set] = None,
+) -> None:
+    """Inject dependencies from the module layer.
+
+    Args:
+        logs_dir: Path to system_logs/ directory to scan.
+        event_callback: Callable(event_name, **kwargs) for firing events.
+        suppressed_files: Set of log file names to skip during detection.
+    """
+    global _logs_dir, _EVENT_CALLBACK, _suppressed_files
+    if logs_dir is not None:
+        _logs_dir = logs_dir
+    if event_callback is not None:
+        _EVENT_CALLBACK = event_callback
+    if suppressed_files is not None:
+        _suppressed_files = suppressed_files
+
+
+def _load_state() -> None:
+    """Load persisted tracking state from disk on first scan."""
+    global _state_loaded
+    if _state_loaded:
+        return
+    _state_loaded = True
+
+    data = json_handler.load_json(_DATA_FILE, "data")
+    if data is None:
+        return
+
+    files = data.get("files", {})
+    for file_key, state_dict in files.items():
+        if not isinstance(state_dict, dict):
+            continue
+        _tracked[file_key] = FileRateState.from_dict(state_dict)
+
+    count = len(_tracked)
+    if count:
+        logger.info("[rate_tracker] Loaded %d file states from disk", count)
+
+
+def _save_state() -> None:
+    """Persist current tracking state to disk."""
+    files = {}
+    for file_key, state in _tracked.items():
+        files[file_key] = state.to_dict()
+
+    from datetime import date
+
+    today = date.today().isoformat()
+    data = {
+        "module_name": _DATA_FILE,
+        "created": today,
+        "last_updated": today,
+        "files": files,
+    }
+    json_handler.save_json(_DATA_FILE, "data", data)
+
+
+def scan_rates() -> list:
+    """Scan all .log files in system_logs/, update rates, fire events if thresholds met.
+
+    Returns a list of dicts describing each tracked file's current state,
+    suitable for display by the log-health command.
+    """
+    _load_state()
+
+    if _logs_dir is None or not _logs_dir.exists():
+        return []
+
+    now = time.time()
+    results = []
+
+    current_files = set()
+    for log_file in _logs_dir.glob("*.log"):
+        file_key = str(log_file)
+        current_files.add(file_key)
+
+        if log_file.name in _suppressed_files:
+            continue
+
+        try:
+            size = log_file.stat().st_size
+        except OSError as exc:
+            logger.info("[rate_tracker] Cannot stat %s: %s", log_file.name, exc)
+            continue
+
+        state = _tracked.get(file_key)
+        if state is None:
+            _tracked[file_key] = FileRateState(size, now)
+            continue
+
+        elapsed = now - state.last_check
+        if elapsed < 1.0:
+            continue
+
+        if size < state.last_offset:
+            state.last_offset = size
+            state.last_check = now
+            state.warning_sustained = 0
+            state.critical_sustained = 0
+            continue
+
+        bytes_added = size - state.last_offset
+        lines_estimate = bytes_added / AVG_LINE_BYTES if bytes_added > 0 else 0.0
+        lines_per_min = (lines_estimate / elapsed) * 60.0
+
+        state.rates.append((now, lines_per_min))
+        state.last_offset = size
+        state.last_check = now
+
+        severity = _evaluate_thresholds(state, lines_per_min, file_key, log_file)
+
+        results.append(
+            {
+                "file": log_file.name,
+                "path": file_key,
+                "size_kb": round(size / 1024, 1),
+                "rate_lines_per_min": round(lines_per_min, 1),
+                "warning_sustained": state.warning_sustained,
+                "critical_sustained": state.critical_sustained,
+                "severity": severity,
+                "branch": detect_branch_from_log(file_key),
+            }
+        )
+
+    stale = [k for k in _tracked if k not in current_files]
+    for k in stale:
+        del _tracked[k]
+
+    _save_state()
+    return results
+
+
+def _window_average(state: FileRateState) -> float:
+    """Average lines/min over the last _WINDOW_INTERVALS entries."""
+    if not state.rates:
+        return 0.0
+    window = list(state.rates)[-_WINDOW_INTERVALS:]
+    return sum(r for _, r in window) / len(window)
+
+
+def _evaluate_thresholds(
+    state: FileRateState,
+    lines_per_min: float,
+    file_key: str,
+    log_file: Path,
+) -> Optional[str]:
+    """Update sustained counters and fire events when thresholds are crossed."""
+    severity = None
+    effective_rate = max(lines_per_min, _window_average(state))
+
+    if effective_rate >= CRITICAL_LINES_PER_MIN:
+        state.critical_sustained += 1
+        state.warning_sustained += 1
+    elif effective_rate >= WARNING_LINES_PER_MIN:
+        state.critical_sustained = 0
+        state.warning_sustained += 1
+    else:
+        if state.fired_warning or state.fired_critical:
+            logger.info(
+                "[rate_tracker] %s rate subsided (%.0f lines/min avg)",
+                log_file.name,
+                effective_rate,
+            )
+        state.warning_sustained = 0
+        state.critical_sustained = 0
+        state.fired_warning = False
+        state.fired_critical = False
+        return None
+
+    if state.critical_sustained >= CRITICAL_SUSTAINED_INTERVALS and not state.fired_critical:
+        severity = "critical"
+        state.fired_critical = True
+        duration = state.critical_sustained * SCAN_INTERVAL
+        _fire_event(file_key, effective_rate, duration, "critical")
+    elif state.warning_sustained >= WARNING_SUSTAINED_INTERVALS and not state.fired_warning:
+        severity = "warning"
+        state.fired_warning = True
+        duration = state.warning_sustained * SCAN_INTERVAL
+        _fire_event(file_key, effective_rate, duration, "warning")
+    else:
+        if state.critical_sustained > 0:
+            severity = "rising_critical"
+        elif state.warning_sustained > 0:
+            severity = "rising_warning"
+
+    return severity
+
+
+def _fire_event(
+    file_path: str,
+    rate_lines_per_min: float,
+    sustained_duration_sec: float,
+    severity: str,
+) -> None:
+    """Fire runaway_log_detected on the trigger event bus."""
+    branch = detect_branch_from_log(file_path)
+    logger.warning(
+        "[rate_tracker] RUNAWAY %s: %s — %.0f lines/min sustained %.0fs (branch: %s)",
+        severity.upper(),
+        Path(file_path).name,
+        rate_lines_per_min,
+        sustained_duration_sec,
+        branch,
+    )
+    json_handler.log_operation(
+        "runaway_detected",
+        {
+            "file": file_path,
+            "rate": rate_lines_per_min,
+            "duration": sustained_duration_sec,
+            "severity": severity,
+            "branch": branch,
+        },
+    )
+
+    if _EVENT_CALLBACK is not None:
+        _EVENT_CALLBACK(
+            "runaway_log_detected",
+            file_path=file_path,
+            rate_lines_per_min=rate_lines_per_min,
+            sustained_duration_sec=sustained_duration_sec,
+            severity=severity,
+            branch=branch,
+        )
+
+
+def _resolve_severity(state: FileRateState) -> Optional[str]:
+    """Derive display severity from a file's current state."""
+    if state.fired_critical:
+        return "critical"
+    if state.fired_warning:
+        return "warning"
+    if state.critical_sustained > 0:
+        return "rising_critical"
+    if state.warning_sustained > 0:
+        return "rising_warning"
+    return None
+
+
+def _file_size(path: str) -> int:
+    """Read file size, returning 0 on any OS error."""
+    try:
+        return Path(path).stat().st_size
+    except OSError:
+        logger.info("[rate_tracker] Cannot stat %s for snapshot", Path(path).name)
+        return 0
+
+
+def get_snapshot() -> list:
+    """Return the current tracking state without scanning (for display only).
+
+    Loads persisted state from disk if not already loaded, so CLI
+    invocations can display rates collected by the monitor process.
+    """
+    _load_state()
+    results = []
+    for file_key, state in _tracked.items():
+        last_rate = state.rates[-1][1] if state.rates else 0.0
+        results.append(
+            {
+                "file": Path(file_key).name,
+                "path": file_key,
+                "size_kb": round(_file_size(file_key) / 1024, 1),
+                "rate_lines_per_min": round(last_rate, 1),
+                "warning_sustained": state.warning_sustained,
+                "critical_sustained": state.critical_sustained,
+                "severity": _resolve_severity(state),
+                "branch": detect_branch_from_log(file_key),
+            }
+        )
+    return results
@@ -17,7 +17,9 @@ bot config passed by the module layer (monitor.py loads from @api secrets).
 import json
 import os
 import threading
+import time
 from datetime import datetime
+from pathlib import Path
 from typing import Optional
 from urllib.error import URLError
 from urllib.request import Request
@@ -25,6 +27,7 @@ from urllib.request import urlopen as _http_fetch

 from aipass.prax.apps.modules.logger import get_direct_logger
 from aipass.prax.apps.handlers.json import json_handler
+from aipass.prax.apps.handlers.monitoring import instance_lock

 logger = get_direct_logger()

@@ -32,6 +35,9 @@ BATCH_INTERVAL = 5.0
 TELEGRAM_MAX_LENGTH = 4000
 FLOOD_CAP = 150

+CONTROL_FILE = Path.home() / ".aipass" / "telegram_bots" / "prax_monitor_control.json"
+_ERROR_MARKERS = ("WARNING", "ERROR", "CRITICAL")
+
 _lock = threading.Lock()
 _buffer: list[str] = []
 _thread: Optional[threading.Thread] = None
@@ -39,6 +45,19 @@ _stop_event = threading.Event()
 _bot_token: Optional[str] = None
 _chat_id: Optional[int] = None
 _RELAY_ACTIVE = False
+_control_mtime: float = 0.0
+_control_cache: dict = {}
+
+_BACKOFF_INITIAL = 1.0
+_BACKOFF_CAP = 60.0
+_SUMMARY_INTERVAL = 300.0
+
+_OFFLINE = False
+_CURRENT_BACKOFF: float = _BACKOFF_INITIAL
+_NEXT_RETRY: float = 0.0
+_OFFLINE_SINCE: float = 0.0
+_SUPPRESSED_COUNT = 0
+_LAST_SUMMARY: float = 0.0


 def init_relay(enabled: bool, config: Optional[dict] = None) -> None:
@@ -66,6 +85,10 @@ def init_relay(enabled: bool, config: Optional[dict] = None) -> None:
        logger.info("[telegram_relay] Incomplete config (missing bot_token or chat_id) — relay inactive")
        return

+    if not instance_lock.try_acquire():
+        logger.info("[telegram_relay] Another process owns the TG relay — viewer-only mode")
+        return
+
    _bot_token = token
    _chat_id = int(chat)
    _RELAY_ACTIVE = True
@@ -105,6 +128,7 @@ def stop_relay() -> None:
        _thread.join(timeout=BATCH_INTERVAL + 2)
        _thread = None

+    instance_lock.release()
    json_handler.log_operation("relay_stopped", {})
    logger.info("[telegram_relay] Relay stopped")

@@ -143,8 +167,41 @@ def _format_event(event) -> Optional[str]:
    return f"[{ts}] [{branch_label}] {event.message}"


+def _read_control() -> dict:
+    """Read the control file, caching by mtime. Returns defaults on missing/invalid file."""
+    global _control_mtime, _control_cache
+
+    try:
+        stat = CONTROL_FILE.stat()
+    except OSError:
+        logger.info("[telegram_relay] Control file not found, using defaults")
+        return {}
+
+    if stat.st_mtime == _control_mtime:
+        return _control_cache
+
+    try:
+        data = json.loads(CONTROL_FILE.read_text(encoding="utf-8"))
+        if not isinstance(data, dict):
+            raise ValueError("not a dict")
+        _control_mtime = stat.st_mtime
+        _control_cache = data
+        return data
+    except (json.JSONDecodeError, ValueError, OSError) as exc:
+        logger.warning("[telegram_relay] Control file parse error, using defaults: %s", exc)
+        _control_mtime = stat.st_mtime
+        _control_cache = {}
+        return {}
+
+
 def _flush_buffer() -> None:
-    """Drain buffer and send to Telegram."""
+    """Drain buffer, apply control-file pause/filter, and send to Telegram.
+
+    Control semantics (written by the @skills TG bot):
+    - paused=true: buffer is discarded, nothing sent.
+    - level="errors": only lines containing WARNING/ERROR/CRITICAL are sent.
+    - level="all" (default): everything is sent.
+    """
    with _lock:
        if not _buffer:
            return
@@ -154,11 +211,26 @@ def _flush_buffer() -> None:
    if not _bot_token or not _chat_id:
        return

+    ctrl = _read_control()
+
+    if ctrl.get("paused", False):
+        return
+
+    level = ctrl.get("level", "all")
+    if level == "errors":
+        lines = [ln for ln in lines if any(m in ln for m in _ERROR_MARKERS)]
+        if not lines:
+            return
+
    if len(lines) > FLOOD_CAP:
        suppressed = len(lines) - FLOOD_CAP
        lines = lines[:FLOOD_CAP]
        lines.append(f"…({suppressed} more suppressed)")

+    if _OFFLINE and time.monotonic() < _NEXT_RETRY:
+        _count_suppressed(len(lines))
+        return
+
    _send_batched(lines)


@@ -191,8 +263,20 @@ def _send_batched(lines: list[str]) -> None:
        _send_message("\n".join(batch))


+def _count_suppressed(count: int) -> None:
+    """Track suppressed events during offline and log a summary at most every 5 minutes."""
+    global _SUPPRESSED_COUNT, _LAST_SUMMARY
+    _SUPPRESSED_COUNT += count
+    now = time.monotonic()
+    if now - _LAST_SUMMARY >= _SUMMARY_INTERVAL:
+        logger.info("[telegram_relay] Still offline — %d events suppressed so far", _SUPPRESSED_COUNT)
+        _LAST_SUMMARY = now
+
+
 def _send_message(text: str) -> bool:
    """POST a single message to the Telegram Bot API."""
+    global _OFFLINE, _CURRENT_BACKOFF, _NEXT_RETRY, _OFFLINE_SINCE, _SUPPRESSED_COUNT, _LAST_SUMMARY
+
    url = f"https://api.telegram.org/bot{_bot_token}/sendMessage"
    payload = json.dumps(
        {
@@ -206,9 +290,29 @@ def _send_message(text: str) -> bool:
    try:
        with _http_fetch(req, timeout=10) as resp:
            result = json.loads(resp.read().decode("utf-8"))
+            if _OFFLINE:
+                duration = time.monotonic() - _OFFLINE_SINCE
+                logger.info(
+                    "[telegram_relay] Recovered (was offline %.0fs, %d events dropped from TG feed)",
+                    duration,
+                    _SUPPRESSED_COUNT,
+                )
+                _OFFLINE = False
+                _CURRENT_BACKOFF = _BACKOFF_INITIAL
+                _SUPPRESSED_COUNT = 0
            return result.get("ok", False)
-    except (URLError, Exception) as e:
-        logger.warning("[telegram_relay] Send failed: %s", e)
+    except (URLError, OSError) as e:
+        now = time.monotonic()
+        if not _OFFLINE:
+            logger.warning("[telegram_relay] TG relay offline: %s", e)
+            _OFFLINE = True
+            _OFFLINE_SINCE = now
+            _CURRENT_BACKOFF = _BACKOFF_INITIAL
+            _SUPPRESSED_COUNT = 0
+            _LAST_SUMMARY = now
+        else:
+            _CURRENT_BACKOFF = min(_CURRENT_BACKOFF * 2, _BACKOFF_CAP)
+        _NEXT_RETRY = now + _CURRENT_BACKOFF
        return False


@@ -0,0 +1,190 @@
+# =================== AIPass ====================
+# Name: log_health.py
+# Description: PRAX Log Health Command — rate overview
+# Version: 1.0.0
+# Created: 2026-07-14
+# Modified: 2026-07-14
+# =============================================
+
+"""
+PRAX Log Health Module
+
+Implements the 'log-health' command showing current log file growth rates
+across system_logs/. Powered by the rate_tracker handler.
+"""
+
+import os
+import sys
+from typing import List
+
+if sys.platform == "win32":
+    os.environ.setdefault("PYTHONUTF8", "1")
+    for _stream in (sys.stdout, sys.stderr):
+        _reconfigure = getattr(_stream, "reconfigure", None)
+        if _reconfigure is not None:
+            _reconfigure(encoding="utf-8", errors="replace")
+
+from aipass.prax.apps.modules.logger import system_logger as logger
+from aipass.cli.apps.modules import console, error
+from aipass.prax.apps.handlers.json import json_handler
+
+
+def print_introspection():
+    """Display module introspection."""
+    console.print()
+    console.print("[bold cyan]PRAX Log Health Module[/bold cyan]")
+    console.print()
+    console.print("[yellow]Purpose:[/yellow]")
+    console.print("  Show log file growth rates and detect runaway logs")
+    console.print()
+    console.print("[yellow]Connected Handlers:[/yellow]")
+    console.print()
+    console.print("  [cyan]prax/handlers/monitoring/[/cyan]")
+    console.print("    [dim]- rate_tracker.py (scan_rates, get_snapshot)[/dim]")
+    console.print()
+    console.print("[dim]Run 'drone @prax log-health --help' for usage[/dim]")
+    console.print()
+
+
+def print_help():
+    """Drone-compliant help output."""
+    console.print()
+    console.print("[bold cyan]PRAX Log Health[/bold cyan]")
+    console.print()
+    console.print("[yellow]Purpose:[/yellow]")
+    console.print("  Monitor log file growth rates across system_logs/")
+    console.print()
+    console.print("[yellow]Subcommands:[/yellow]")
+    console.print()
+    console.print("  [cyan]scan[/cyan]      Scan all log files and show current rates")
+    console.print("  [cyan]snapshot[/cyan]   Show last known rates (no new scan)")
+    console.print()
+    console.print("[yellow]Usage:[/yellow]")
+    console.print()
+    console.print("  [dim]# Scan and show current rates[/dim]")
+    console.print("  $ drone @prax log-health scan")
+    console.print()
+    console.print("  [dim]# Show last known rates without scanning[/dim]")
+    console.print("  $ drone @prax log-health snapshot")
+    console.print()
+
+
+def _display_rates(results: list, is_scan: bool) -> None:
+    """Display rate results in a formatted table."""
+    label = "Scan" if is_scan else "Snapshot"
+    console.print()
+    console.print(f"[bold cyan]Log Health {label}[/bold cyan] [dim](system_logs/)[/dim]")
+
+    if not results:
+        console.print("  [dim]No log files tracked yet[/dim]")
+        console.print()
+        return
+
+    active = [r for r in results if r["rate_lines_per_min"] > 0]
+    idle = [r for r in results if r["rate_lines_per_min"] == 0]
+    flagged = [r for r in results if r.get("severity")]
+
+    console.print(f"  Files tracked: {len(results)}")
+    console.print(f"  Active: {len(active)}, Idle: {len(idle)}")
+    if flagged:
+        console.print(f"  [yellow]Flagged: {len(flagged)}[/yellow]")
+
+    if flagged:
+        console.print()
+        console.print("[yellow]Flagged files:[/yellow]")
+        for r in sorted(flagged, key=lambda x: x["rate_lines_per_min"], reverse=True):
+            sev = r["severity"] or ""
+            color = "red" if "critical" in sev else "yellow"
+            console.print(
+                f"  [{color}]{sev.upper()}[/{color}] "
+                f"{r['file']}: {r['rate_lines_per_min']} lines/min "
+                f"({r['size_kb']} KB) [{r['branch']}]"
+            )
+
+    if active:
+        console.print()
+        console.print("[cyan]Active files:[/cyan]")
+        for r in sorted(active, key=lambda x: x["rate_lines_per_min"], reverse=True):
+            if r.get("severity"):
+                continue
+            console.print(f"  {r['file']}: {r['rate_lines_per_min']} lines/min ({r['size_kb']} KB) [{r['branch']}]")
+
+    if idle and len(idle) <= 10:
+        console.print()
+        console.print("[dim]Idle files:[/dim]")
+        for r in sorted(idle, key=lambda x: x["file"]):
+            console.print(f"  [dim]{r['file']}: {r['size_kb']} KB [{r['branch']}][/dim]")
+    elif idle:
+        console.print()
+        console.print(f"  [dim]{len(idle)} idle files (0 lines/min)[/dim]")
+
+    console.print()
+
+
+def _get_event_callback():
+    """Return trigger.fire if available, else None."""
+    try:
+        from aipass.trigger.apps.modules.core import trigger
+
+        return trigger.fire
+    except ImportError as exc:
+        logger.info("[log-health] trigger not available: %s", exc)
+        return None
+
+
+def handle_command(command: str, args: List[str]) -> bool:
+    """Handle log-health command.
+
+    Args:
+        command: Command name
+        args: Command arguments
+
+    Returns:
+        True if command was handled
+    """
+    if command != "log-health":
+        return False
+
+    if not args:
+        print_introspection()
+        return True
+
+    if args[0] in ("--help", "-h", "help"):
+        print_help()
+        return True
+
+    from aipass.prax.apps.handlers.monitoring.rate_tracker import scan_rates, get_snapshot, configure
+    from aipass.prax.apps.handlers.config.load import get_system_logs_dir
+
+    configure(logs_dir=get_system_logs_dir(), event_callback=_get_event_callback())
+
+    subcmd = args[0]
+    logger.info("[log-health] %s", subcmd)
+    json_handler.log_operation("log_health_executed", {"mode": subcmd})
+
+    if subcmd == "scan":
+        results = scan_rates()
+        _display_rates(results, is_scan=True)
+        return True
+
+    if subcmd == "snapshot":
+        results = get_snapshot()
+        _display_rates(results, is_scan=False)
+        return True
+
+    error(f"Unknown log-health subcommand: {subcmd}")
+    print_help()
+    return True
+
+
+if __name__ == "__main__":
+    if len(sys.argv) == 1:
+        print_introspection()
+        sys.exit(0)
+
+    if "--help" in sys.argv:
+        print_help()
+        sys.exit(0)
+
+    args = [arg for arg in sys.argv[1:] if not arg.startswith("--")]
+    handle_command("log-health", args)
@@ -48,7 +48,6 @@ from aipass.prax.apps.handlers.monitoring.telegram_relay import (
    is_relay_enabled_by_env,
 )
 from aipass.prax.apps.handlers.monitoring.pid_cache import get_pid_for_branch as _get_pid_for_branch
-from aipass.prax.apps.handlers.monitoring import instance_lock

 import json as _json

@@ -64,6 +63,7 @@ _module_tracker: Optional[ModuleTracker] = None
 _display_thread: Optional[threading.Thread] = None
 _file_watcher_thread: Optional[threading.Thread] = None
 _log_watcher_thread: Optional[threading.Thread] = None
+_rate_tracker_thread: Optional[threading.Thread] = None


 def print_introspection():
@@ -175,9 +175,7 @@ def _load_relay_config() -> Optional[dict]:
 def _run_monitor(args: List[str]) -> bool:
    """Launch Mission Control live monitoring."""
    global _event_queue, _module_tracker
-    global _display_thread, _file_watcher_thread, _log_watcher_thread
-
-    instance_lock.acquire(error_fn=error)
+    global _display_thread, _file_watcher_thread, _log_watcher_thread, _rate_tracker_thread

    json_handler.log_operation("monitor_started", {"args": args})
    logger.info(f"Starting unified monitoring (args: {args})")
@@ -226,20 +224,20 @@ def _run_monitor(args: List[str]) -> bool:

 def _start_threads():
    """Start all monitoring threads"""
-    global _display_thread, _file_watcher_thread, _log_watcher_thread
+    global _display_thread, _file_watcher_thread, _log_watcher_thread, _rate_tracker_thread

-    # Display thread - pulls from event queue and displays
    _display_thread = threading.Thread(target=_display_worker, daemon=True)
    _display_thread.start()

-    # File watcher thread - watches filesystem changes
    _file_watcher_thread = threading.Thread(target=_file_watcher_worker, daemon=True)
    _file_watcher_thread.start()

-    # Log watcher thread - watches log files
    _log_watcher_thread = threading.Thread(target=_log_watcher_worker, daemon=True)
    _log_watcher_thread.start()

+    _rate_tracker_thread = threading.Thread(target=_rate_tracker_worker, daemon=True)
+    _rate_tracker_thread.start()
+
    logger.info("All monitoring threads started")


@@ -254,11 +252,10 @@ def _stop_threads():
        _event_queue.stop()

    # Join all daemon threads with timeout
-    for t in (_display_thread, _file_watcher_thread, _log_watcher_thread):
+    for t in (_display_thread, _file_watcher_thread, _log_watcher_thread, _rate_tracker_thread):
        if t is not None and t.is_alive():
            t.join(timeout=2.0)

-    instance_lock.release()
    logger.info("All monitoring threads stopped")


@@ -491,6 +488,29 @@ def _log_watcher_worker():
        stop_log_watcher()


+def _rate_tracker_worker():
+    """Rate tracker thread — scans system_logs/ for runaway growth every SCAN_INTERVAL."""
+    from aipass.prax.apps.handlers.monitoring.rate_tracker import scan_rates, configure, SCAN_INTERVAL
+    from aipass.prax.apps.handlers.config.load import get_system_logs_dir
+
+    try:
+        from aipass.trigger.apps.modules.core import trigger
+
+        event_cb = trigger.fire
+    except ImportError as exc:
+        logger.info("[monitor] trigger not available for rate tracker: %s", exc)
+        event_cb = None
+
+    configure(logs_dir=get_system_logs_dir(), event_callback=event_cb)
+
+    while not _stop_event.is_set():
+        try:
+            scan_rates()
+        except Exception as exc:
+            logger.info("[monitor] Rate tracker scan error: %s", exc)
+        _stop_event.wait(SCAN_INTERVAL)
+
+
 def _handle_interactive_cmd(cmd: str, get_help_text) -> None:
    """Dispatch an interactive monitor command."""
    if cmd == "help":
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
AIPass	53a695580f	Merge pull request #698 from AIOSAI/dev Merge playbook SOP fixes from live run PPLAN-0010	2026-07-16 23:43:55 -07:00
AIOSAI	6163202a93	release: bump 2.7.2 — compass v2 + ambient recall window. New cadence ruling: PATCH bump + tag every merge, PyPI tracks main	2026-07-16 23:29:26 -07:00
AIOSAI	4912d96f58	docs(flow): merge playbook SOP fixes from live run PPLAN-0010 — gated fetch step replaced with drone @git sync, create-syntax hint at template top	2026-07-16 23:19:50 -07:00
AIPass	87bfccd55b	Merge pull request #697 from AIOSAI/dev Startup protocol: announce current PID on greeting	2026-07-16 23:04:50 -07:00
AIOSAI	a89ddb30bd	fix(ci): seedgo gate back to 100% — hooks handler logging + silent catches, memory governance modules/handlers split (import path frozen, bridge E2E green), devpulse README test count. All 17 branches 100%	2026-07-16 22:44:01 -07:00
AIOSAI	e412cfed14	fix(drone): namespace subprocess timeout to --drone-timeout — plain --timeout passes through to modules (watchdog 600s regression, live repro x2). Regression test, 879 green, CHANGELOG both fixes	2026-07-16 22:04:03 -07:00
AIOSAI	b8f6fb8dad	fix(memory): plan-ID searches pin the exact plan via metadata lookup (Patrick ruling) + purge 193 scratchpad junk vectors. Live-verified 100% top-hit, 1011 green	2026-07-16 22:03:04 -07:00
AIOSAI	6b0dcdccef	fix(memory): governance module ate all @memory commands — standard handle_command signature, declines non-governance commands (Track 2 follow-up). Search verified live, 1011 green	2026-07-16 21:31:10 -07:00
AIOSAI	1902775812	feat(compass): Track 2 ambient recall — compass_recall hook + pure governance (memory) + recall API w/ rare-token scoring (devpulse), verbatim tidbit injection, live acceptance matrix green (DPLAN-0246/FPLAN-0332). 446+1011+1129 tests, seedgo 100%	2026-07-16 20:43:49 -07:00
AIOSAI	03dfce20b4	feat(devpulse): compass curation v2 Track 1 — supersedes links + atomic archive, write-time FTS conflict advisory, note command, --include-archived, score code-removal, review-per-prep (DPLAN-0246/FPLAN-0331). 435 green, seedgo 31/31	2026-07-16 19:45:21 -07:00
AIOSAI	b3bb529a6a	feat(drone): 3-layer timeout policy — per-command overrides + --timeout flag + 30s default, override hint in error (DPLAN-0245). 878 tests green	2026-07-16 00:09:28 -07:00
AIOSAI	9a61d237fa	feat(flow): close pipeline self-completing — auto-vectorization from post_close_runner + locked atomic registry writes (DPLAN-0245). E2E proven, 730 green	2026-07-16 00:09:14 -07:00
AIOSAI	702e335cb8	fix(skills): TG routine read-timeouts silenced on OSError path + outage episode-start demoted ERROR→WARNING per Patrick ruling — ends medic wake-loop. 825 TG tests green	2026-07-15 23:21:41 -07:00
AIOSAI	73e9ededd4	fix(flow): CLOSED_PLANS append race — O_EXCL lockfile with retry/backoff, surface silent append failures (S314 sweep lost 18/21 entries). 730 tests green	2026-07-15 23:21:26 -07:00
AIOSAI	bad08e9b03	fix(memory): unwedge plan vectorization — salt vector IDs with source file, per-file batches with incremental manifest (DPLAN-0245). Backlog drained 229/229, 990 tests green	2026-07-15 23:21:11 -07:00
AIOSAIandClaude Fable 5	851988abbc	fix(seedgo): DPLAN-0244 trust files to 100% standards — json_handler + justified bypasses CI seedgo gate is strict 100%. trust_registry.py now uses json_handler for registry I/O (log_operation on writes only — no per-hook-event flood); unused_function bypassed (cross-branch public API, static analysis can't see callers). trust.py gained print_introspection + --help/no-args gates; genuinely-N/A standards (json_structure delegated to trust_registry, frozen cross-branch import) bypassed with justification. Both branches 100%, all tests green, live acceptance re-verified (attack still blocked both gates). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YEAyLFCuo4uD934fwFxocz	2026-07-15 19:17:45 -07:00
AIOSAIandClaude Fable 5	222a9c8382	docs(navmap): open-source status is fleet-wide identity, not a coding footnote (Patrick ruling) Every agent's tier1 navmap now states AIPass is open source in the intro and reframes the house rule as write-as-if-it-ships. External scans are contributions, not intrusions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YEAyLFCuo4uD934fwFxocz	2026-07-15 18:23:13 -07:00
AIOSAIandClaude Fable 5	a8b1ce6658	feat(hooks): DPLAN-0244 hooks.json trust model hardening — Layer A engine gates + Layer B registry/CLI Closes a zero-interaction RCE where a hostile repo's .aipass/hooks.json (discovered via loader CWD walk-up, bridge wired globally) could run an arbitrary command-type hook on SessionStart. Defense-in-depth: Layer A (engine): refuse command-type hooks from per-project configs via unconditional _source clobber; gate handler paths to aipass.* namespace. Layer B (loader+CLI): trusted-project registry (path+sha256), fail-closed trust-check, $AIPASS_HOME-only bootstrap (no TOFU), aipass init auto-enroll + new aipass trust/revoke commands. Live acceptance test (real bridge, real payload) proves both gates block independently. 1105 hooks + 133 aipass tests green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YEAyLFCuo4uD934fwFxocz	2026-07-15 18:23:02 -07:00
AIOSAI	807924241f	docs: startup protocol — announce current PID on greeting (Patrick)	2026-07-15 14:12:55 -07:00
AIPass	91f8cc6c07	Merge pull request #696 from AIOSAI/dev Post-v2.7.0 fixes: #692 legacy builder-class migration + birth-cert template, #694 order-dependent test pollution. Both verified live: Vera Studio dry-run plans exactly the 2 passport migrations (zero writes), original repro pair passes, prax+skills+spawn 1576 tests green, skills_json litter eliminated.	2026-07-15 13:37:45 -07:00
AIOSAI	989d19020d	chore(release): bump 2.7.0 -> 2.7.1 — supply-chain hardening (DPLAN-0243: commit signing, hash-pinned CI tooling, release provenance attestation), TG streaming v2 polish, rate_tracker burst-evasion fix, CI green pass	2026-07-15 13:24:41 -07:00
AIOSAI	b4f66ce84d	docs(flow): merge playbook template — DPLAN-0243 new-setup notes (auto-SSH-signing, hash-pinned CI advisory mode, provenance attestation step). Built by @flow, 730/730 green, seedgo 100%	2026-07-15 13:19:29 -07:00
AIOSAI	294d25cea3	docs: CHANGELOG entry for DPLAN-0243 supply-chain hardening (signing, hash-pinned CI, provenance, hvtracker#186)	2026-07-15 12:36:19 -07:00
AIOSAI	9048666c65	ci: OSSF scorecard hardening (DPLAN-0243) — hash-pin all standalone workflow pip installs via .github/requirements/ locks (pip/lint/build/e2e/audit, pip-compile --generate-hashes, 11/11 target-env closure verified incl. Windows colorama marker fix) + provenance attestation on publish (attest-build-provenance v4.1.1 SHA-pinned, id-token+attestations perms) + dependabot pip ecosystem for the new locks. Editable -e . installs untouched byte-identical. 5/5 fresh-venv --require-hashes installs green, 5/5 YAML parse, pinned ruff matches repo lint. First SSH-signed commit (repo config wired this session).	2026-07-15 12:21:22 -07:00
AIOSAI	25fc02d07a	feat: TG streaming v2 polish (DPLAN-0229) — logs_were_active + multi-chunk (>4096) finalize now honor the streaming flag: logs-active reconcile-edits the streamed message instead of Done.+fresh, multi-chunk edits chunk 1 in place + sends [2/N] continuations, edit-fail falls back safely. Batch path verified zero-change via regression tests. 6 new tests, 1077 hooks green + seedgo 100% devpulse-verified. Live streamed-turn proof pending Patrick's next streaming session.	2026-07-15 11:14:05 -07:00
AIOSAI	9dc2ecd604	feat: rate_tracker v1.2.0 burst-evasion fix — severity evaluates max(instant_rate, 60s window avg): bursty runaways (20 lines/6s retry-loop shape, 200/min avg, previously 4min undetected live) now sustain through gap windows; continuous unchanged, subsidence clears. 4 new burst tests, 1032 green + seedgo 100% devpulse-verified, live-proven from running service: RUNAWAY WARNING prax_burst_storm_test.log 191 lines/min sustained 120s.	2026-07-15 10:45:38 -07:00
AIOSAI	13ae64cbae	fix: unpark the parked CI reds — Patrick ruling: red CI is never parked. @prax: relay mtime-cache flake root-caused (test relied on two writes sharing one mtime-granularity window — true locally, false on CI) — os.utime pins mtime so the cache contract tests deterministically, 50/50+20/20 loops green. @hooks: 4 Windows session_boot reds — _tmux_session_exists() real subprocess spawn (no tmux on Windows, WinError 2) mocked per `ca096295` convention, execvp already mocked = zero real spawns left. 1028 prax + 102 session_boot green, devpulse-verified.	2026-07-15 10:26:09 -07:00
AIOSAI	024cf5a104	fix: pin sys.platform=linux in test_is_pid_alive_dead — Windows runners take the OpenProcess path so the os.kill mock never fires; test reds whenever PID 1234 is alive on the runner (first hit today, 6/6 sibling tests were already pinned by `ca096295`). 38 presence tests green.	2026-07-15 09:03:01 -07:00
AIOSAI	be68d23d6a	fix: CI green pass on PR#696 — lint (ruff format on trigger runaway tests), seedgo 100% both red branches (@hooks: new json_handler + persistent_alert/alert_dismiss wired through it, cc_sessions introspection gate, _menu_live nesting extraction, 1071 tests; @prax: rate_tracker DI refactor — module layer injects logs_dir + trigger.fire via configure(), 1028 tests), navmap trim 9.3k→7.9k under injection cap. All owner-fixed via dispatch, devpulse-verified: both audits 100% independently re-run, full runaway chain re-proven live post-refactor (332 lines/min storm → WARNING at 130s → trigger → @aipass triage → alert banner rendered in-session → dismiss clears). Burst-evasion design finding (pre-existing, not regression) filed to @prax by mail.	2026-07-15 08:49:21 -07:00
AIOSAI	81658ce0ea	feat: runaway-log detection + escalation (DPLAN-0242, agent-designed) + citizen wake-back. Prax-led three-branch build via TDPLAN-0013: prax rate_tracker (disk-persisted volume detection, WARNING >100 l/min 2min / CRITICAL >10 l/s 1min, 4th monitor thread) + drone @prax log-health; trigger runaway_log_detected event + handler (per-file 30min cooldown independent of medic breaker, UNKNOWN->prax, writes .aipass/alerts.json); hooks persistent_alert banner + drone @hooks dismiss (devpulse fixed nearest-.aipass path bug in both + wired settings.json - registration is not deployment). Live-fire proven: planted 240 l/min storm -> detect 257 l/min -> event -> dispatch -> @aipass autonomous no-action triage -> banner -> dismiss. ai_mail wake ruling: owner-gate removed (citizen wake-back live-proven), devpulse structurally unwakeable (manager check all paths), self-wake loop found+fixed same night (guard + senderless wake-back sessions). Navmap comms section, 4 branch READMEs + root README + CHANGELOG. ~77 new tests, suites green: prax 1028, trigger 619, hooks 1071, ai_mail 765	2026-07-15 00:05:02 -07:00
AIOSAI	de109846bf	fix: prax TG relay offline backoff — network-class send failures enter offline mode (1s-60s doubling, flush gate skips sends, monitor loop never blocks, viewers keep rendering), log-once (enter + 5min summary + recovery w/ drop count), full reset on first success. Found live in Patrick's plug-pull: bots went quiet right, relay spun Send failed every 5s (89 lines) + self-fed via log watcher re-ingest. 11 new tests, 1007 prax green devpulse-verified	2026-07-14 17:01:23 -07:00
AIOSAI	5744073c11	fix: TG bot offline hot-spin — network-class poll errors (DNS/connection/socket) back off exponentially 1s-60s cap, reset on first successful poll; log-once semantics (1 unreachable line + 5min summaries + 1 recovery line) instead of 13 err/sec; routine long-poll read-timeouts silent (863/day medic noise class gone). Found live: Patrick's tether outage spun all 5 bots for an hour. 25 new tests, 822 TG + 252 skills green devpulse-verified	2026-07-14 16:29:57 -07:00
AIOSAI	b791af2372	feat: medic revival + concurrent monitor viewers — pytest logs route to tmp (PYTEST_CURRENT_TEST, fixture storms cant pollute prod logs), breaker self-heals (half-open on read, close on probe, cooldown decay), TTL mutes (mute/off auto-expire 24h, --for/--forever, temp off keeps detection), footer+navmap mute breadcrumbs; prax monitor lock scoped to TG relay role (relay.pid) so viewers always start — Patrick ruling: processes are not agents. Loop proven live: planted commons bug fixed by medic dispatch in 105s byte-identical. 993 prax + 603 trigger green	2026-07-14 14:57:40 -07:00
AIOSAI	af12158cbc	fix: TG bot heartbeat race — generation counter kills resurrected heartbeat threads (shared stop Event cleared by next start let a >5s-stuck thread overwrite delivered replies with Processing...), delivered re-check before every edit in batch+streaming loops, superseded pending placeholders finalized in message+file paths (rapid-fire single-slot strand). Root-caused live from Patrick's frozen bubble; 6 new heartbeat tests, full TG suite 797 green devpulse-verified.	2026-07-14 11:00:48 -07:00
AIOSAI	62d047cac1	fix: TG mirror live-test round — agent_type filter unblocked main chats (daemon-backed sessions carry agent_type=claude; subagents never fire UserPromptSubmit; skip is now agent_id-based) + TG-echo gate (bot stores injected_prompt in pending file, relay skips fresh undelivered text-match; raw injections carry no marker). Found live by Patrick's morning door-tests; mirror proven both directions. 791 TG tests green (incl. cross-file mock fix @skills missed).	2026-07-14 09:16:02 -07:00
AIOSAI	5c82db6729	feat: TG user-comment mirror — user messages from ALL doors (terminal/remote) now mirror to the branch TG chat with origin tag. UserPromptSubmit relay handler (self-contained in telegram skill, crash-isolated last entry in hooks.json), human-only noise fences (system/task notifications, slash-command output, dispatch wakes, subagents, TG-echo, dupes), inbound hardening (stale-pending clean before write, undelivered-overwrite warning). 47 new TG tests, execution proven via engine.jsonl, positive path live-verified to real TG chat.	2026-07-14 02:42:41 -07:00
AIOSAI	116c4e9d69	fix: DPLAN-0241 round 4 — R6 extra_args threaded through ALL launch paths (user flags survive resume/takeover/continue/dead-window/headless), R7 auto-namer stamps --name branch-shortid on every launch (flag live-verified 2.1.209, user -n/--name wins), new-over-all aborts on failed daemon stop, honest close-all hint, exit/q/quit in all menus. op:kill per-job stop found in daemon socket protocol — documented, not shipped (undocumented internal). 1048 hooks tests green (102 session_boot, 11 CLI contract).	2026-07-14 02:10:44 -07:00
AIOSAI	0b739ac525	fix: DPLAN-0241 rounds 2-3 — Enter IS the takeover. Phantom claude-agents-stop removed (bg close honest, never SIGTERM), bg resume = daemon stop --any (returncode-checked, blast-radius y/N confirm) + --resume in tmux with bypass, ALL interactive launches tmux-wrapped, multi-session menu shows real names + explicit pick + honest new/close, real-binary CLI contract test tier (20 tests, phantom-subcommand class unshippable). 1025 hooks tests green, all facts live-verified vs claude 2.1.208. Plus DPLAN north-star: one conversation per branch, surfaces are views, agents bind to machine not interface.	2026-07-13 23:49:57 -07:00
AIOSAI	364cefa5fd	fix: DPLAN-0241 session-mgmt overhaul — boot shim passthrough (only bare/permission-mode intercepted), session_boot 3-option boot menu (resume/new-closes-old/close, per-kind proper stops, never kill for bg), presence_gate repaired (session-file PID resolver, bg sessions gated) + wired OBSERVE-ONLY, wire_verify flags unwired security hooks as ERROR, new drone @hooks sessions + reclaim, PID-first session naming. Plus S304 discovery report + DPLAN-0241. Verified live: gate's first production run logged correct would-block, no self-block; 987 hooks tests green.	2026-07-13 22:50:27 -07:00
AIOSAI	e9b86c3ebb	feat: TG log-stream control — /logs on branch bots (persisted pref, honored by auto-start) + prax_monitor receiver bot (/pause /resume /errors /all /status, menu registered) + relay honors shared control file each flush. 84 new tests, all suites green; live-verified end-to-end from Telegram Web (errors filter kicked in within one flush).	2026-07-12 11:22:19 -07:00
AIOSAI	e0811fcf93	fix: #692 builder->aipass_framework legacy migration + birth-cert template; #694 test-order pollution (prax fixture scope + skills hermetic tests). Verified: repro pair passes, 1576 tests green, Vera dry-run plans exactly 2 migrations zero writes.	2026-07-12 00:02:32 -07:00