菜单
Post-task retrospective — capture lessons learned and store them for future sessions. Builds cumulative intelligence across sessions.
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。
基于 SOC 职业分类
| name | retro |
| description | Post-task retrospective — capture lessons learned and store them for future sessions. Builds cumulative intelligence across sessions. |
| argument-hint | [task description or 'this session'] |
| user_invocable | true |
| tools | Read, Glob, Grep, Bash |
| model | inherit |
After completing a task, capture what went well, what went wrong, and what to do differently. Store findings in context-store for cumulative intelligence.
$ARGUMENTS
Look at the work just completed:
git log --oneline -20
git diff --stat main..HEAD
Read MEMORY.md for task context and decisions made.
For each of these categories, note anything worth remembering:
What went well (reinforce):
What went wrong (prevent):
What was learned (remember):
For each lesson worth preserving, call mcp__context-store__store_context with:
project: the project namecategory: lessontitle: short, searchable titlecontent: the lesson with contexttags: relevant tags for future searchFor architectural decisions, also call mcp__context-store__record_decision.
## Retro: [Task]
### Lessons Stored
1. [lesson title] — [one-line summary]
2. [lesson title] — [one-line summary]
### Decisions Recorded
1. [decision] — [rationale in brief]
### Patterns to Watch
- [anything that might recur]
Commit all changes, push to remote, and open a pull request in one shot. The most common inner-loop workflow.
Clean start for a new task. Saves important decisions, clears context, and re-reads project config.
Resume work from a previous session. Reads MEMORY.md and continues autonomously.
Adversarial code review — stress-tests recent changes by hunting for bugs, edge cases, security holes, and design flaws. Spawns parallel attack agents, consolidates into a severity-rated report.
Debug and investigate issues — trace symptoms to root cause. Use when something is broken, a test fails mysteriously, or behavior is unexpected.
Prove that the current implementation works. Runs tests, diffs behavior, validates correctness. Use after implementing a feature or fix to verify before marking complete.