This wraps up completed work with a six-phase process: runs a council verdict to validate what shipped, extracts learnings from the session, scores and deduplicates the backlog, promotes high-value insights to MEMORY.md, retires stale knowledge, and surfaces next work for the flywheel. It's move 7 in the operating loop, designed to capture evidence against beads and ratchet learnings forward only when they repeat or change behavior. Most observations die at handoff, which is intentional. Use it after finishing an epic or batch of work. The council integration means you get multi-model validation on whether you actually shipped what you intended before deciding what to remember.
npx skills add https://github.com/boshu2/agentops --skill post-mortem