feat(system): README Round 4 overhaul — apply all Goldfish findings (FPLAN-0165)

Co-Authored-By: @devpulse <devpulse@aipass>
This commit is contained in:
AIOSAI
2026-04-07 22:24:45 -07:00
co-authored by @devpulse
parent 035b8ad933
commit e0ecad665c
5 changed files with 84 additions and 147 deletions
+5
View File
@@ -42,6 +42,11 @@ Your identity and branch context are also injected via hooks on session start an
You are a citizen of AIPass. Your `.trinity/passport.json` defines who you are. Read it first — before writing anything, before making decisions. Your role, purpose, and principles are in that file.
## Security
- NEVER read, access, or reference files in `~/.secrets/` or `/home/patrick/.secrets/`. This directory contains API keys, tokens, and recovery codes. No agent needs to see this. Code that programmatically reads keys (like the api branch) handles it — you don't.
- NEVER output credentials, tokens, or API keys in responses.
## Key Principles
- Code is truth. Running code beats architecture.
+10 -8
View File
@@ -1,21 +1,23 @@
# AIPass
A multi-agent framework where autonomous citizens live in branches and deploy disposable agents to do work.
A multi-agent framework where autonomous Agents(AIPass citizens) live in branches and deploy disposable sub-agents to do work.
**User:** Name
# AIPass — Project Prompt
Project-level instructions. Persists in context for the entire conversation.
## Startup
# AIPass — Startup protocol
On any greeting, silently read these files from CWD and run the commands — no narration, no announcing steps. Just do it and respond with the status.
**Read:** `.trinity/passport.json`, `.trinity/local.json`, `.trinity/observations.json`, `README.md`, `STATUS.local.md`
**Check:** If `.ai_mail.local/inbox.json` exists, read it. Process any mail — don't ask.
**Check:** If `.ai_mail.local/inbox.json` exists, read it. Process any mail — don't ask,
**list:** `dropbox` files. Ignore README.md
**Run:** `git status`
## Security
- NEVER read, access, or reference files in `~/.secrets/` or `/home/patrick/.secrets/`. This directory contains API keys, tokens, and recovery codes. No agent needs to see this. Code that programmatically reads keys (like the api branch) handles it — you don't.
- NEVER output credentials, tokens, or API keys in responses.
## Memories
Update `.trinity/` at natural breakpoints, after milestones, and on `/memo`. If compaction hits before you save, it's gone.
Update `.trinity/` at natural breakpoints, after milestones, and on `/memo`.
+5
View File
@@ -42,6 +42,11 @@ Your identity and branch context are also injected via hooks on session start an
You are a citizen of AIPass. Your `.trinity/passport.json` defines who you are. Read it first — before writing anything, before making decisions. Your role, purpose, and principles are in that file.
## Security
- NEVER read, access, or reference files in `~/.secrets/` or `/home/patrick/.secrets/`. This directory contains API keys, tokens, and recovery codes. No agent needs to see this. Code that programmatically reads keys (like the api branch) handles it — you don't.
- NEVER output credentials, tokens, or API keys in responses.
## Key Principles
- Code is truth. Running code beats architecture.
+60 -137
View File
@@ -1,12 +1,10 @@
[![Status](https://img.shields.io/badge/status-beta-yellow)](HERALD.md)
[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue)](pyproject.toml)
[![License: MIT](https://img.shields.io/badge/license-MIT-green)](LICENSE)
[![PyPI](https://img.shields.io/pypi/v/aipass)](https://pypi.org/project/aipass/)
[![CLIs](https://img.shields.io/badge/CLIs-Claude%20%7C%20Codex%20%7C%20Gemini-purple)](#cli-support)
[![Give Feedback](https://img.shields.io/badge/Give-Feedback-brightgreen)](https://github.com/AIOSAI/AIPass/issues/new?template=feedback.yml)
<!-- TODO: Terminal GIF here — show a dispatch + mail + memory session -->
# AIPass
**Your AI agents remember yesterday.**
@@ -20,11 +18,9 @@ A local multi-agent framework where your AI assistants keep their memory between
- [The Problem](#the-problem)
- [What AIPass Does](#what-aipass-does)
- [Quick Start](#quick-start)
- [What You Can Do](#what-you-can-do)
- [How It Works](#how-it-works)
- [The 15 Agents](#the-15-agents)
- [CLI Support](#cli-support)
- [Platform Support](#platform-support)
- [Project Status](#project-status)
- [Requirements](#requirements)
- [Subscriptions & Compliance](#subscriptions--compliance)
@@ -45,14 +41,14 @@ What's missing isn't more agents — it's *presence*. Agents that have identity,
## What AIPass Does
AIPass is a local CLI framework that gives your AI agents **identity, memory, and teamwork**. Tested with Claude Code, Codex, and Gemini — built to work with any AI that can read a file and follow a prompt.
AIPass is a local CLI framework that gives your AI agents **identity, memory, and teamwork**. Verified with Claude Code, Codex, and Gemini CLI. Designed for terminal-native coding agents that support instruction files, hooks, and subprocess invocation.
**Start with one agent that remembers:**
Your AI reads `.trinity/` on startup and writes back what it learned before the session ends. That's the whole memory model — JSON files your AI can read and write. Next session, it picks up where it left off. No database, no API, no setup beyond one command.
```bash
aipass init ~/Projects/my-saas-app # coming soon — currently requires dev setup
aipass init ~/Projects/my-saas-app
```
Your project gets its own registry, its own identity, and persistent memory. Each project is isolated — its own agents, its own rules. No cross-contamination between projects.
@@ -67,56 +63,26 @@ When one agent isn't enough, use `spawn` to create specialists with the same inf
| A lightweight specialist | `spawn passport` | Identity + rich memory (no apps scaffold) |
| A full specialist | `spawn create` | All of the above + apps scaffold, mail, dashboard, tests |
**How AIPass itself is built:**
AIPass ships with 15 specialist agents that maintain and develop the framework. It looks like a lot — but every agent follows one pattern and needs one command:
```bash
drone @branch command [args] # Every agent, every task. Drone handles routing.
```
Drone resolves who you're talking to, routes the work, and handles errors. You never need to know where anything lives. Each agent has the same directory layout — learn the pattern once and you know every agent.
```
devpulse (orchestrator)
├── drone — command routing + @agent resolution
├── seedgo — 33 automated quality standards
├── prax — real-time monitoring across all agents
├── ai_mail — agent-to-agent communication + task dispatch
├── flow — plans, workflows, phased coordination
├── spawn — creates new agents anywhere on your filesystem
├── memory — vector search across archived context
├── backup — versioned backups + Google Drive sync
└── ...and 6 more specialists
```
These 15 agents work on the **same filesystem, same project, same time**. No sandboxes. No worktrees. No isolation. They see each other's work, coordinate through mail, and share a planning system that prevents conflicts.
**How they stay out of each other's way:**
- **Agent locks** — an agent can't run twice simultaneously. Dispatch checks for active locks before waking anyone.
- **PR locks** — only one agent creates a PR at a time. No merge conflicts from parallel commits.
- **File isolation** — branches don't touch each other's files. Hard rule, enforced every session.
- **Standards baked in** — quality checks are embedded in every workflow template. Agents follow them without being told.
- **Self-healing** — the monitoring system detects errors and can dispatch the offending agent to fix itself.
This is the pattern your projects inherit.
**What makes this different:**
- **Agents are persistent.** They have memories and expertise that develop over time. They're not disposable workers — they're specialists who remember.
- **Everything is local.** Your data stays on your machine. Memory is JSON files. Communication is local mailbox files. No cloud dependencies, no external APIs for core operations.
- **Projects are isolated by design.** Each project gets its own registry. Agents communicate within their project, not across projects. No external agent can accidentally break your system.
- **Standards are baked into the workflow.** Quality isn't an afterthought — workflow templates inject standards directly into every plan, ensuring agents follow best practices automatically.
- **One pattern for everything.** Every agent follows the same structure. One command (`drone @branch command`) reaches any agent. Learn it once, use it everywhere.
- **Projects are isolated by design.** Each project gets its own registry. Agents communicate within their project, not across projects.
- **The system protects itself.** Agent locks prevent double-dispatch. PR locks prevent merge conflicts. Branches don't touch each other's files. Quality standards are embedded in every workflow. Errors trigger self-healing.
**Say "hi" tomorrow and pick up exactly where you left off.** One agent or fifteen — the memory persists.
<p align="right"><a href="#contents">Back to contents</a></p>
---
## Quick Start
```bash
pip install aipass # Install the package
```
Or clone the full framework with all 15 agents:
```bash
git clone https://github.com/AIOSAI/AIPass.git
cd AIPass
@@ -131,74 +97,30 @@ cd src/aipass/devpulse
claude # devpulse reads its memory, knows who it is, picks up where it left off
```
Say "hi." devpulse reads its identity and memory files, tells you what's been happening, and is ready to dispatch work. Come back tomorrow — it remembers.
Say "hi." Here's what that looks like:
```
You: hi
devpulse: Hey. Picking up where we left off.
Status:
- Branch: main, up to date
- Inbox: 1 email from drone — routing fix applied
- Git: 3 files modified, not committed
- Dropbox: 1 item from @api
Ready when you are.
```
Come back tomorrow — it remembers.
```bash
# See the system in action:
# More things you can do:
drone @seedgo audit aipass # Run 33 quality checks across all agents
drone @flow create . "Add user auth" # Create a work plan
drone systems # List every agent and what it does
```
<details>
<summary>Linux (fully tested)</summary>
Works out of the box. This is the primary development platform.
```bash
./setup.sh
```
</details>
<details>
<summary>macOS (untested, should work)</summary>
setup.sh should work on macOS. Known issue: Apple Silicon Macs may need Homebrew path adjustment for symlinks.
```bash
brew install python@3.10
./setup.sh
```
</details>
<details>
<summary>Windows</summary>
**WSL2 (recommended):** setup.sh runs with zero changes inside WSL2.
**Native Windows:** Has been tested on Windows 10 with most functionality working. No setup.ps1 yet — manual setup required.
</details>
<details>
<summary>Docker</summary>
```bash
docker build -t aipass .
docker run -d -p 8080:8080 aipass
```
Opens a code-server IDE with Python, Node.js, and Claude Code pre-installed.
</details>
---
## What You Can Do
- **Start any project with memory.** `aipass init` gives your AI persistent context — it picks up where you left off, every session, no re-explaining.
- **Build your own agents.** Use `spawn` to create agents with the same infrastructure AIPass runs on. Full scaffold or lightweight — your call.
- **Dispatch work, don't do it yourself.** Send a task to the right agent — it investigates, builds, tests, and reports back. You keep working on something else.
- **Or work with an agent directly.** `cd src/aipass/memory && claude` — sit down with a specialist one-on-one for complex problems. Direct access is a first-class workflow, not a fallback. The agent has its own memory, its own expertise, and picks up where you left off.
- **Agents work as a team.** Agents within a project share one filesystem, communicate through mail, and coordinate through plans. No sandboxes isolating them.
- **Enforce quality automatically.** Standards are embedded in every workflow template. Agents follow them without being told. Code stays consistent at scale.
- **Use any AI CLI.** Tested with Claude Code, Codex, and Gemini. Same hooks, same identity, same commands.
- **Scale your way.** One agent with memory, or fifty agents with full infrastructure. The framework grows with your project.
<p align="right"><a href="#contents">Back to contents</a></p>
---
## How It Works
@@ -215,7 +137,11 @@ src/aipass/<agent>/
└── README.md # Domain knowledge (the agent reads this on startup)
```
Identical layout everywhere. If you know one agent, you know all of them. Communication happens through mail, coordination through plans, and routing through one command:
Identical layout everywhere. If you know one agent, you know all of them. One command reaches anyone:
```bash
drone @branch command [args] # Every agent, every task. Drone handles routing.
```
```bash
drone @seedgo audit aipass # Run quality checks on everything
@@ -228,7 +154,22 @@ drone @ai_mail dispatch @memory "Archive old sessions" "Find sessions older than
- **Team mode (most of the time):** Talk to `devpulse`, dispatch work across the team. Agents work in parallel and report back.
- **Direct mode (for deeper work):** `cd src/aipass/memory && claude` — work one-on-one with a specialist when the problem needs focused domain expertise.
<p align="right"><a href="#contents">Back to contents</a></p>
**AIPass ships with 15 specialist agents** that maintain and develop the framework — the reference implementation proving the architecture works at scale:
```
devpulse (orchestrator)
├── drone — command routing + @agent resolution
├── seedgo — 33 automated quality standards
├── prax — real-time monitoring across all agents
├── ai_mail — agent-to-agent communication + task dispatch
├── flow — plans, workflows, phased coordination
├── spawn — creates new agents anywhere on your filesystem
├── memory — vector search across archived context
├── backup — versioned backups + Google Drive sync
└── ...and 6 more specialists
```
These agents work on the **same filesystem, same project, same time** — no sandboxes, no worktrees. This is the pattern your projects inherit.
---
@@ -247,6 +188,9 @@ You don't need to memorize this list. Start with `devpulse`, use `drone` to reac
| [**memory**](src/aipass/memory/README.md) | Long-term vector search across all agent knowledge |
| [**spawn**](src/aipass/spawn/README.md) | Creates new agents from templates |
<details>
<summary>See all 15 agents</summary>
**Quality and operations** — how the system stays healthy:
| Agent | Role |
@@ -267,6 +211,8 @@ You don't need to memorize this list. Start with `devpulse`, use `drone` to reac
| [**commons**](src/commons/README.md) | Community space for agent updates and discussion |
| [**skills**](src/skills/README.md) | Reusable capabilities agents can invoke |
</details>
---
## CLI Support
@@ -281,19 +227,6 @@ AIPass works with three AI coding CLIs. Claude Code is the most tested.
setup.sh auto-detects which CLIs are installed and configures hooks for each.
<p align="right"><a href="#contents">Back to contents</a></p>
---
## Platform Support
| Platform | Status |
|----------|--------|
| Linux | Fully tested |
| Windows (WSL2) | Expected to work, zero changes needed |
| Windows (native) | Partial testing on Windows 10 |
| macOS | Untested, should work |
---
## Project Status
@@ -310,19 +243,15 @@ setup.sh auto-detects which CLIs are installed and configures hooks for each.
For detailed session history, see [HERALD.md](HERALD.md).
<p align="right"><a href="#contents">Back to contents</a></p>
---
## Requirements
- Python 3.10+
- Linux recommended (macOS should work; Windows via WSL2)
- At least one AI CLI: [Claude Code](https://docs.anthropic.com/en/docs/claude-code) (recommended), [Codex](https://github.com/openai/codex), or [Gemini CLI](https://github.com/google-gemini/gemini-cli)
- At least one AI CLI: Claude Code (recommended), Codex, or Gemini CLI
- `sudo` access (for global CLI symlinks)
- API keys optional (only for the `api` branch — OpenRouter/OpenAI)
<p align="right"><a href="#contents">Back to contents</a></p>
- API keys optional (only for the `api` agent — OpenRouter/OpenAI)
- **Platforms:** Linux (fully tested), macOS (untested, should work), Windows via WSL2
---
@@ -331,12 +260,10 @@ For detailed session history, see [HERALD.md](HERALD.md).
### Use your existing subscription
AIPass runs on your **existing CLI subscription** — Claude Pro/Max, Codex, or Gemini. No API keys required for core functionality. No extra costs. Your subscription covers everything.
AIPass runs on your **existing CLI subscription** — Claude Pro/Max, Codex, or Gemini. No API keys required for core functionality. No extra costs beyond your existing subscription.
This works because AIPass runs each CLI as an **official subprocess** — the same binary you'd run yourself in a terminal. It doesn't extract credentials, proxy API calls, or intercept tokens. Your subscription stays within the provider's infrastructure at all times.
This is different from tools like OpenClaw that were [restricted by Anthropic](https://venturebeat.com/technology/anthropic-cracks-down-on-unauthorized-claude-usage-by-third-party-harnesses) for extracting subscription OAuth tokens and routing workloads outside the official CLI. AIPass doesn't do that — it enhances the CLI through officially supported extension points (hooks, CLAUDE.md, AGENTS.md, GEMINI.md).
### What AIPass does NOT do
- Extract or redirect subscription OAuth tokens
@@ -344,12 +271,8 @@ This is different from tools like OpenClaw that were [restricted by Anthropic](h
- Bypass rate limits or prompt caching
- Impersonate official CLI clients
Claude Code is proprietary but officially supports hooks and subprocess usage. Codex and Gemini CLI are open source (Apache 2.0). No provider forbids this usage pattern.
Claude Code is proprietary but officially supports hooks and subprocess usage. Codex and Gemini CLI are open source (Apache 2.0).
> API keys are only needed for the optional `api` agent (OpenRouter/OpenAI). For server/automated deployments, API key authentication is recommended per [Anthropic's guidance](https://code.claude.com/docs/en/legal-and-compliance).
</details>
---
<p align="center"><a href="#aipass">Back to top</a></p>
+4 -2
View File
@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
[project]
name = "aipass"
version = "2.0.0"
description = "Orchestration framework for autonomous AI agent ecosystems"
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"
requires-python = ">=3.10"
@@ -29,7 +29,6 @@ dependencies = [
"rich>=13.0",
"watchdog>=3.0",
"requests>=2.28",
"openai>=1.0",
]
[project.urls]
@@ -38,6 +37,9 @@ Repository = "https://github.com/AIOSAI/AIPass"
"Bug Tracker" = "https://github.com/AIOSAI/AIPass/issues"
[project.optional-dependencies]
llm = [
"openai>=1.0",
]
trinity = [
"trinity-pattern>=1.0.0",
]