菜单
Retrospective analysis after ultrawork or ralph-loop sessions — agent usage patterns, bottlenecks, and improvement suggestions. Activate when #re appears anywhere in the user message.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Review the current diff, run the narrowest useful verification, and summarize remaining release risk. Activate when #sc appears anywhere in the user message.
Show unified omo status dashboard — boulder, ralph-loop, tasks, teams, and memory state at a glance. Activate when #ss appears anywhere in the user message.
Check if README.md and user guide reflect current project state — philosophy, features, agent models, skill list. Fix any drift. Activate when #sync appears anywhere in the user message.
Full release pipeline — commit, push, tag, GitHub release, marketplace version sync. Activate when #rel appears anywhere in the user message.
Triage a failure by capturing symptoms, narrowing reproduction, isolating likely files, and choosing the smallest useful verification loop. Activate when #bh appears anywhere in the user message.
Multi-perspective code review on the current diff or specified files. Examines correctness, security, performance, maintainability, and scope creep. Activate when #dr appears anywhere in the user message.
| 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:
.claude/state/task-history.log for recent task events..claude/state/current-task.txt and the corresponding task note in .claude/state/tasks/..claude/state/handoffs/..claude/state/ralph-loop.json exists (indicates ralph-loop was used)..claude/state/briefings/ if they exist.Phase 2 — Analyze patterns:
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.