AGENT SKILL

retrospective

Use at the end of a unit of work to capture durable, non-obvious knowledge into project memory (memory/) and flag decisions worth an ADR. Turns one-off discoveries into compounding, persistent context.

License 许可
Apache-2.0
最近更新
2026-09-13
Stars
★ 0
Forks
0

01 前言 Front Matter

name
retrospective
description
Use at the end of a unit of work to capture durable, non-obvious knowledge into project memory (memory/) and flag decisions worth an ADR. Turns one-off discoveries into compounding, persistent context.

03 SKILL.md 正文

Skill: retrospective

Purpose

Close out work by persisting what was learned so future sessions (human or agent) do not re-derive it. Keeps memory/ a living, trustworthy asset.

Required Input

  • The change or task just completed (diff, branch, or summary).
  • Notable surprises, gotchas, decisions, or new conventions encountered.

Output Format

  • Concise entries appended to the correct memory file:
    • memory/learnings.md — facts and gotchas (dated entry format).
    • memory/patterns.md — conventions others should follow.
    • memory/context.md — updated active focus / open threads.
  • A short summary of what was recorded.
  • A note on whether any decision warrants an ADR (docs/adr/).

Execution Rules

  1. Record only durable, non-obvious knowledge. Skip anything already clear from the code, the README, or commit messages.
  2. One fact per entry; keep entries short and self-contained.
  3. Use absolute dates (YYYY-MM-DD), not relative ones.
  4. Delete or correct memory entries that turned out to be wrong.
  5. If a significant, hard-to-reverse decision was made, recommend writing an ADR.
  6. Keep all artifacts in English.

04 同仓库更多 Skills