Maestro Post-Mortem

Look back over the current Maestro session and find what could have gone better. The Maestro hooks record an objective timeline of every tool call, subagent dispatch, and handoff to .claude/maestro_session.log.jsonl; you also carry the conversation context of this session. Coupling the two lets you spot wasted work and bad calls — and turn them into concrete fixes.

This is read-and-reason first; it never changes anything until the user opts in.

Workflow

1. Generate the digest. The log embeds full subagent input/output, so don't read it raw — run the helper, which condenses it:

   node "${CLAUDE_SKILL_DIR}/../../scripts/maestro-post-mortem.js" "${CLAUDE_PROJECT_DIR:-.}"
   

   If it reports no session log found, tell the user the post-mortem must run during an active Maestro session — the log is ephemeral and is wiped at SessionEnd — then stop. (The digest also accepts --json if you'd rather parse it.)

2. Couple the digest with this session's context. The digest is the objective record (what ran, how often, which handoffs returned what). Your conversation context supplies the things the log can't: why you did something, which errors were real vs. spurious, where you went down a dead end, and which assumptions turned out wrong. Read the digest's flags as leads, not verdicts — confirm or dismiss each against what actually happened.