description: Use for AI collaboration postmortem / 复盘 when a user or project admin asks to review missed issues, mistaken approval, repeated rework, unclear responsibility boundaries, broken message handoff, or failed AI collaboration workflow, and produce a result that lands in workflow, docs, code, data, evidence, or closure.

fp - AI协作复盘

Use this skill when the user explicitly asks for fp、复盘、协作复盘、流程复盘，或追问为什么问题漏发现、为什么审核误放行、为什么反复修不好、为什么消息或角色协作断链。

Before acting, read AI协作复盘规范.md in this skill directory completely.

Core rules:

Formal postmortem starts only when the user or project admin triggers it.
AI may suggest a postmortem, but must not upgrade ordinary issues into formal postmortem by itself.
The postmortem must have an outcome: workflow, docs, code, data, evidence, closure, downgrade, rerun, or assigned fix.
Check executor side, reviewer side, governance side, and collaboration/message side.
Do not over-escalate or add unnecessary process burden.
Ask the user only when a real blocking decision cannot be resolved from documents, evidence, logs, code, permissions, or project configuration.
Keep large evidence in files; show only summaries, file paths, message IDs, and key proof entries in the Codex window.