diff --git a/HERALD.md b/HERALD.md index 09394c54..d0f8550b 100644 --- a/HERALD.md +++ b/HERALD.md @@ -4,23 +4,29 @@ > The living record. What happened, what's changing, what matters. -**Last updated:** 2026-04-05 | **Session:** 74 | **PRs merged:** 181 +**Last updated:** 2026-04-05 | **Session:** 76 | **PRs merged:** 192 --- ## Current State - **15 branches** operational -- **4,900+ tests** system-wide (4,907 collected) -- **181 PRs** merged since inception -- **73 sessions** of development +- **4,900+ tests** system-wide +- **192 PRs** merged +- **76 sessions** of development +- **Multi-CLI support** — Claude Code, Codex (GPT-5.4), Gemini all integrated with hooks, identity, skills +- **Prax model tags** — monitor shows [BRANCH/MODEL] for all three CLIs - **Sentinel** — AIPass's first external project (Claude Code JSONL analyzer) -- **Dispatch safety net** — startup timeout, auto-retry, JSONL monitoring (FPLAN-0164) -- **README overhauled** — grouped branch tables, compliance section, navigation links - **Nexus vision** — personal AI companion architecture (co-founded with GPT-4o, March 2025) ## Recent Sessions +### S76 — Multi-CLI Integration Into Main (2026-04-05) +Ported Codex and Gemini from Docker prototype into the AIPass repo. Created AGENTS.md, GEMINI.md, .codex/ (hooks + skills), .gemini/ (hooks + skills). Updated setup.sh to detect CLIs and wire hooks automatically. Prax built Codex/Gemini log adapters, model tags ([DEVPULSE/OPUS], [DEVPULSE/GPT-5.4], [DEVPULSE/GEMINI-3-FLASH]), inotify errno-specific messages, and CWD branch detection. Fixed AGENTS.md startup behavior. PRs #189-192. + +### S75 — Multi-CLI Proven in Docker (2026-04-05) +Proved all three CLIs (Claude Code, Codex gpt-5.4, Gemini flash-preview) work in same Docker container. Hooks, identity, drone, ai_mail, seedgo, sub-agents all functional. Commons discussion with genuine cross-model responses. Git fixed (squash-merge to regular merge). PR #188 merged. + ### S74 — README Overhaul + Research Sprint (2026-04-05) Full README restructure: grouped branch tables with README links, collapsible setup sections, navigation (back-to-contents after every section, back-to-README in HERALD/STATUS). Compliance & Safety section added — researched Anthropic's April 4 OpenClaw crackdown, documented AIPass's full compliance (official CLI, hooks, no credential wrapping). Trademark research: Japanese hospitality company "AiPass" exists but different domain, low conflict. AIPL token optimization research: TOON/SNS achieve 30-85% savings, custom compressed language for AIPass internal comms is feasible (40-60% savings on structured content). 25 inbox messages processed from S73 night shift. Herald updated (was stale since S68). Subscription tier discussion: Max 20x to 5x testing. diff --git a/README.md b/README.md index cb66cd56..867ad1e1 100644 --- a/README.md +++ b/README.md @@ -34,7 +34,7 @@ A multi-agent operating system where AI agents live as citizens in a shared file ## What is AIPass -AIPass (**AI Passport**) is a multi-agent framework built on [Claude Code](https://docs.anthropic.com/en/docs/claude-code). Each agent is a **citizen** — it has an identity (passport), persistent memory, a mailbox, and the ability to communicate with other agents. Citizens live in **branches** (directories), each specializing in a domain. One orchestrator coordinates them all. +AIPass (**AI Passport**) is a multi-agent framework built on [Claude Code](https://docs.anthropic.com/en/docs/claude-code), with support for [OpenAI Codex](https://github.com/openai/codex) and [Google Gemini CLI](https://github.com/google-gemini/gemini-cli). Each agent is a **citizen** — it has an identity (passport), persistent memory, a mailbox, and the ability to communicate with other agents. Citizens live in **branches** (directories), each specializing in a domain. One orchestrator coordinates them all. The system is model-agnostic — hooks, identity, and commands work the same across all three CLIs. You talk to one agent. It dispatches work to specialists and brings results back. Memory persists across sessions — you never re-explain context.