graft-ai-plan-governance

name	graft-ai-plan-governance
description	Repository-specific workflow for Graft ai-plan governance. Use when a docs/automation task changes ai-plan router documents, active-topic recovery materials, templates, catalog coverage, or bounded ai-plan validators without creating a second startup, validation, or recovery path.

Graft AI Plan Governance

Use this skill when a docs/automation task changes the governance shape of ai-plan/**.

Treat root AGENTS.md and ai-plan/AGENTS.md as the governance sources. This skill does not replace startup, validation, commit, closeout, or runtime authority.

Read First

1. Complete the root AGENTS.md startup preflight.
2. Read ai-plan/AGENTS.md.
3. Read .ai/environment/tools.ai.yaml.
4. Read ai-plan/README.md.
5. Read ai-plan/design/governance/ai/AI任务追踪与恢复设计.md.
6. Read ai-plan/design/governance/ai/AI工具与MCP接入治理规范.md.
7. Read ai-plan/public/README.md after startup preflight.
8. Read the touched topic's README.md, startup-prompt.md, tracking, and trace files.
9. If the same slice also changes .agents/skills/** or scripts/**, pair this skill with graft-ai-governance-audit.

When To Use

Use this skill when the task changes any of:

• ai-plan/AGENTS.md
• ai-plan/README.md
• ai-plan/catalog.json
• ai-plan/design/** documents that define ai-plan governance or ADR convergence
• ai-plan/public/README.md
• ai-plan/public/<topic>/** recovery materials
• ai-plan/templates/**
• Work Intake, Work Contract, or contract-driven bootstrap governance under ai-plan/**
• bounded docs/automation validators under scripts/** that guard ai-plan governance structure
• repo-local skills under .agents/skills/** only when they are part of the same ai-plan governance slice

When Not To Use

Do not use this skill for:

• server, web, OpenAPI, or runtime implementation work
• whole-repo ai-plan/** retrofit, frontmatter rollout, or projection-model expansion
• tasks whose real authority lives outside ai-plan/**
• broader AI tooling or MCP adoption work that does not materially change ai-plan governance; use graft-ai-governance-audit for that

Workflow

1. Authority And Router

• keep root AGENTS.md as the only startup-governance source
• choose the narrowest ai-plan owner: design/, design/decisions/, roadmap/, public/, public/archive/, lessons/, or templates/
• treat ai-plan/catalog.json as supplementary, single-file, and bounded
• treat Work Intake as the only approved entry workflow for new long-running work
• treat design, roadmap, topic, and ADR as artifacts selected by Work Contract, not peer intake paths
• escalate outside ai-plan/** only when authority discovery proves it is required

2. Topic Safety

• preserve concurrent active topics in ai-plan/public/README.md
• do not break another topic's startup entry while updating shared router docs
• keep startup-prompt.md aligned with root startup governance; do not invent a second receipt format
• when substantive work changes an active topic, update its tracking and trace files in the same change
• when a topic comes from Work Intake, keep the persisted Work Contract in tracking rather than inventing a new topic-level metadata file

3. Skill And Script Coupling

• add or modify repo-local skills only when the batch explicitly requires them
• keep new skills narrow and tied to existing authority docs
• if a validator or skill is added, sync only the minimum docs that must route future work to it
• keep validators as docs/automation guards, not runtime or CI substitutes

Validation

For ai-plan/** governance work, always run:

git diff --check

Add:

python3 scripts/validate_ai_plan_structure.py

when the slice touches ai-plan/catalog.json, ai-plan/public/README.md, ai-plan/templates/**, active-topic recovery structure, or the ai-plan structure guard itself.

Add:

python3 scripts/validate_ai_governance.py

when the slice touches .agents/skills/**, scripts/**, or ai-plan/design/governance/ai/AI工具与MCP接入治理规范.md.

Run:

python3 -m unittest scripts/test_validate_ai_plan_structure.py

only when scripts/validate_ai_plan_structure.py or its tests change.

Guardrails

• do not create a second startup path, second validation truth, or hidden recovery store
• do not let graft-work-intake own independent intake rules outside root AGENTS.md, ai-plan/AGENTS.md, and the approved ADRs
• do not start compose-project-management implementation from ai-plan governance work
• do not mass-retrofit existing topics to match template wording
• do not widen bounded validators into a whole-repo ai-plan linter
• do not treat owned scope as permission to ignore upstream authority drift

Closeout

Report:

ai_plan_governance:
- authority_owner:
- router_paths_changed:
- concurrent_topics_preserved:
- validators_run:
- scope_expanded: