English | 简体中文

A public playbook for turning prompts into a governed AI development system with charters, workflows, reusable skills, Copilot routing, evaluation gates, and knowledge sinks.

这是一个把“几段 Prompt”升级成“团队可治理 AI 协作体系”的公开样板仓库，包含章程、工作流、Skill 源码、Copilot 适配、评测门禁和知识沉淀方法。

Teams often hit the same ceiling when using AI for development:

• prompts keep growing, but behavior still drifts
• implementation happens before planning is confirmed
• code search degrades into noisy full-repo scanning
• results stay in chat history instead of becoming reusable knowledge
• “done” is declared before tests and evaluation gates are actually closed

This repo packages one practical answer:

charter -> workflow -> skill -> tool entry -> docs sink

flowchart LR
    A["Charter"] --> B["Workflow"]
    B --> C["Skill Source"]
    C --> D["Runtime Artifacts"]
    D --> E["Copilot / Claude / Prompt Entry"]
    E --> F["Task Delivery"]
    F --> G["Evaluation Gate"]
    G --> H["Docs Sink / Knowledge Base"]

• It separates governance from execution.
• It treats skills as reusable methods, not just longer prompts.
• It keeps prompt files thin and pushes stable logic into source-controlled skills.
• It routes risky work through explicit planning and evaluation gates.
• It assumes results should land in docs, not disappear into chat history.

• A governance layer for AI-assisted engineering
• A prompt-to-skill migration pattern
• A reusable docs/skills-src/ structure
• Generated runtime targets for .claude/skills/ and .github/
• A plan-gate -> auto-dev -> evaluation-gate delivery path
• A docs/tasks/ and docs/knowledge-base/ sink pattern

git clone https://github.com/liutiger/ai-collab-skill-playbook.git
cd ai-collab-skill-playbook/template

python3 docs/skills-src/tools/validate_skills.py
python3 docs/skills-src/tools/validate_copilot_assets.py
python3 docs/skills-src/tools/acceptance_check.py

Then open these in order:

1. template/docs/guides/AI协作试运行说明.md
2. template/docs/guides/AI协作研发章程.md
3. template/docs/skills-src/README.md

• docs/ Public-facing notes about structure, publishing, and sanitization.
• template/ A copyable project template with:
  • template/AGENTS.md
  • template/docs/guides/
  • template/docs/prompts/
  • template/docs/skills-src/
  • template/.github/
  • template/.claude/skills/

1. template/docs/guides/AI协作试运行说明.md
2. template/docs/guides/AI协作研发章程.md
3. template/docs/guides/ai-workflow.md
4. template/docs/prompts/README.md
5. template/docs/skills-src/README.md

• Engineering teams moving from prompt collections to reusable AI workflows
• Teams using GitHub Copilot but wanting stronger governance and repeatability
• Teams experimenting with Claude-style local skills and runtime generation
• Multi-module or chain-heavy systems that need planning gates and knowledge sinks

1. Copy template/ into your own repo.
2. Rewrite AGENTS.md with your real project context.
3. Rewrite the charter and 00-department-standards.md.
4. Update docs/skills-src/manifest.yaml.
5. Regenerate .github/ and .claude/skills/.
6. Pilot with plan-gate first, then add auto-dev and evaluation-gate.

• Start with docs/public-sharing-checklist.md
• Then read docs/repo-structure.md
• Before pushing your own fork, read docs/publish-to-github.md

• README.zh-CN.md
• README.en.md

This is a public-first template, not a turnkey SaaS product.

The goal of v1 is simple:

• make the system understandable
• make the template copyable
• make the generated runtime artifacts reproducible
• make planning, delivery, evaluation, and docs sink visibly connected