| name | retro |
| description | Retrospective analysis after ultrawork or ralph-loop sessions — agent usage patterns, bottlenecks, and improvement suggestions. Activate when #re appears anywhere in the user message. |
| argument-hint | |
| disable-model-invocation | false |
| allowed-tools | Read, Glob, Grep, Bash, Task |
Run a retrospective analysis of the most recent work session.
Phase 1 — Gather evidence:
- Read
.claude/state/task-history.log for recent task events.
- Read
.claude/state/current-task.txt and the corresponding task note in .claude/state/tasks/.
- Read recent handoff notes in
.claude/state/handoffs/.
- Check if
.claude/state/ralph-loop.json exists (indicates ralph-loop was used).
- Read recent briefings from
.claude/state/briefings/ if they exist.
Phase 2 — Analyze patterns:
- Which agents were used and how often (estimate from task notes and briefings).
- Were there repeated failures or retries.
- How many iterations did ralph-loop take (if used).
- Were there scope changes or blockers.
- Did the task complete successfully or was it handed off.
Phase 3 — Generate retrospective:
Session Retrospective
=====================
Task: [task name]
Duration: [estimated from timestamps]
Agents used: [list]
What went well:
- [bullet points]
What could improve:
- [bullet points]
Bottlenecks:
- [bullet points]
Recommendations for next time:
- [bullet points]
Patterns to remember:
- [items worth saving to .claude/state/memory/ for future sessions]
Phase 4 — Memory update:
If .claude/state/memory/ exists, suggest specific entries to add based on the retrospective findings. Do not auto-write — present suggestions and ask before writing.