Agent项目必备文件管理器

Project Governance

中文：给 AI 长期项目建立一套「项目记忆 + 文件索引 + 工作规则 + 版本记录」的管理系统，让 AI 换会话、换模型后仍能正确接着项目做，而不是重新猜项目。

30-second overview

这个 Skill 给项目增加 8 个东西：

1. 项目规则 —— AI 应该怎么做
2. 文件地图 —— 文件在哪里
3. 项目状态 —— 现在做到哪里
4. 错误记录 —— 以前踩过什么坑
5. 版本索引 —— 哪个版本才是真的
6. 黑白名单 —— 什么能用、什么不能用
7. 会话交接 —— 上一个 AI 做到哪里
8. 变更记录 —— 为什么这么改

核心目标：让 AI 换会话、换模型、甚至换 Agent 后，仍然能正确接着项目做，而不是重新猜项目。

When to Use

Use this skill when:

• The user complains the project is messy, files are scattered, or the AI keeps misplacing files ("你怎么又乱放文件").
• The user expects the AI to find files itself instead of asking for paths ("你自己找").
• The AI keeps repeating mistakes or using wrong versions.
• The user says "你不是应该记得吗？" / "上次不是已经验证过了吗？" / "哪个才是最终版？" / "别重新做，之前已经跑通了" — Memory + Governance boundary scenarios where memory alone is not a reliable authority.
• Starting a new AI-assisted project and you want the agent to follow a stable protocol from day one.
• Onboarding an AI agent into an existing project that has no rules, error log, or parameter registry.
• A project has grown messy: files scattered, parameters changed without record, past mistakes repeated.
• You want to enforce durable rules such as "index-first file lookup", "plan before execute", "file existence ≠ file validity", or "registry-driven parameter selection".

Do NOT use this skill when:

• The task is a one-off question or small edit that does not need project-wide conventions.
• The project already has a mature governance system and you only need a small rule tweak — edit the existing files directly instead.

Memory & Governance Boundary

Platform memory (e.g. Trae user profile / project memory) is a context source, not an authoritative fact store. Governance files are the project execution protocol. The two complement each other:

Authority priority when they conflict:

1. Current project files / frozen versions
2. Project governance files (AGENTS.md, index.md, VERSIONS.md, registries)
3. Project memory
4. User long-term memory
5. AI inference

When memory and governance files disagree, governance wins. Durable conventions learned from memory must be settled into the governance files after human confirmation — memory alone never becomes the project's authority.

Instructions

Step 1 — Scaffold

python scripts/governance.py init --project-dir /path/to/project --project-name "My Project"

Creates 11 governance files from templates/ (never overwrites existing files unless --force).

Step 2 — Customize

Edit the generated AGENTS.md: directory permission zones, autonomy levels, artifact placement rules, and project-specific rules under "Project Customization". Keep the universal Core Governance Rules as-is.

Step 3 — Maintain (every session)

1. Session start: read index.md → session_handoff.md → LESSONS.md; before generating parameters, read blacklist.json / whitelist.json.
2. During work: find files via the index (never blind search); inherit whitelist entries with score > 0.85; never use permanent_ban: true; record new mistakes in LESSONS.md.
3. Session end: update session_handoff.md, index.md (file changes), CHANGELOG.md (decisions).

Step 4 — Validate, index, check

python scripts/governance.py validate --project-dir /path/to/project   # registries conform to schema
python scripts/governance.py index --project-dir /path/to/project      # rebuild index.md map (links + notes)
python scripts/governance.py check --project-dir /path/to/project      # health gate: files + registries + fresh index

Input / Output

• Input: a project directory (with or without existing governance files), a project name, and the user's governance pain points (messy files, wrong versions, repeated mistakes, "你自己找").
• Output: a governance workspace (AGENTS.md, index.md, VERSIONS.md, LESSONS.md, session_handoff.md, CHANGELOG.md, whitelist.json / blacklist.json) plus a validated, up-to-date index.

On Failure

• init fails (invalid path, permission): report the exact failing command and reason; do not partially scaffold or guess.
• validate reports schema errors: fix the registry entries; never bypass validation.
• check fails (missing files / stale index): run index, then re-run check; if still failing, report to the human.
• Never fabricate a "passed" result — report what was verified and what was not.

What It Produces

Details

• README.md — full package overview: Core / CLI / Skill-adapter structure, subcommands, limitations, Trae cold-start acceptance test
• templates/ — governance file templates
• scripts/governance.py — init / validate / index / check CLI

Limitations

This package is currently designed and tested primarily for Trae Skills. Other agents may use the generated governance files, but automatic Skill discovery/loading is not guaranteed outside Trae. The governance files and CLI are kept agent-neutral for future adapters.