Agent Rules

Universal AI collaboration rules extracted from production projects. Technology-agnostic, reusable across any codebase.

Quick Start

One-liner (remote, no clone needed)

cd your-project && curl -fsSL https://raw.githubusercontent.com/CtriXin/agent-rules/main/setup.sh | bash

Local (if you already cloned)

cd your-project && bash /path/to/agent-rules/setup.sh

The script will:

• Create .ai/ state management structure
• Add rules as CLAUDE.md (or AGENT_RULES.md if CLAUDE.md exists)
• Add AGENTS.md for Codex startup
• Add AI_BOOTSTRAP.md for cross-AI startup
• Never overwrite existing CLAUDE.md, AGENT.md, AGENTS.md, or AI_BOOTSTRAP.md
• Update .gitignore for session-specific files
• Show warnings for any conflicts

Structure

agent-rules/
├── AGENT_RULES.md          # Complete rule set (single file, all-in-one)
├── rules/                  # Individual rule files (granular)
│   ├── startup.md          # Session startup protocol (mandatory 3-read)
│   ├── iteration.md        # 6-step iteration cycle
│   ├── todo-management.md  # Four-quadrant TODO + archive lifecycle
│   ├── plan-quality.md     # Plan quality gates (preflight + challenge)
│   ├── code-redlines.md    # Code quality hard limits + security
│   ├── error-handling.md   # Error escalation protocol (max 2 retries)
│   ├── documentation.md    # Documentation sync rules
│   ├── sensitive-ops.md    # Sensitive operation guardrails
│   ├── worktree.md         # Git worktree parallel development
│   ├── cross-agent-review.md  # a2a review trigger rules
│   ├── multi-ai-handoff.md    # Multi-AI task handoff protocol
│   └── rule-sync.md        # Rules for modifying rules
├── templates/              # Copy to your project
│   ├── AGENTS.md           # Codex bootstrap entry
│   ├── AI_BOOTSTRAP.md     # Cross-AI startup guide
│   └── .ai/                # State management infrastructure
│       ├── manifest.json   # Project state (single source of truth)
│       ├── workflow.md     # Execution flow reference
│       ├── plan/
│       │   ├── current.md  # Current task plan template
│       │   └── TODO.md     # Four-quadrant TODO template
│       ├── startup/
│       │   └── STARTUP_BRIEF.md  # Session recovery brief
│       ├── summaries/      # Review outputs go here
│       └── modules/
│           └── _template/
│               └── doc.md  # Module documentation template
├── docs/
│   ├── agent-rules-overview.md
│   └── diagrams/
│       ├── README.md
│       └── agent-rules-overview.elements.json
└── LICENSE

Two Layers

Layer	Purpose	Where
Rules	Behavioral constraints — what AI must/must not do	AGENT_RULES.md + rules/
Templates	State infrastructure and startup entry files	templates/

Rules tell the AI HOW to behave. Templates give it WHERE to persist state.

Rule Categories

Rule	What It Enforces
Startup	Read project state before doing anything
Iteration	6-step cycle, plans must be landed not just suggested
Plan Quality	5-dimension preflight + challenge review
Code Red Lines	File ≤800 lines, function ≤30 lines, no eval/innerHTML
Error Handling	Max 2 retries, then 3 options for user
Documentation	Every code change syncs docs, never delete (archive)
Sensitive Ops	Confirm before destructive/irreversible actions
Worktree	Git worktree for parallel, don't auto-delete
Cross-Agent Review	When a2a review is mandatory vs skippable
Multi-AI Handoff	Breakpoint protocol for switching models
Rule Sync	How to modify rules without breaking things
TODO Management	Four-quadrant TODOs, checkbox lifecycle, monthly archive

中文导览

• 稳定入口: docs/agent-rules-overview.md
• 最新中文流程图: docs/diagrams/agent-rules-system-map-20260313-110304.elements.json

License

MIT