weekly-review 周回顾

The review that changes nothing is a diary. This one closes loops.

不改变任何事的回顾是日记。这个是闭环。

Why / 为什么

Weekly reviews fail by being vibes: "busy week, lots of progress." A real review reconciles what was committed against what happened — which is uncomfortable exactly as often as it is useful. The output is not feelings; it is three artifacts: a reconciliation, a carry-forward list with fresh reasons, and a small set of next-week commitments.

周回顾的失败方式是"感觉"："很忙，进展很多。"真正的回顾是把承诺过的和实际发生的对账——它让人不舒服的程度恰好等于它有用的程度。产出不是感受，是三样东西：对账、带新理由的顺延清单、下一周的少量承诺。

The five steps / 五步

1. Pull last week's commitments — from the last review log, not from memory.
2. Reconcile — for each commitment: done / partially done / not done, with evidence for each claim of completion.
3. Bank the wins — completed work gets recorded as evidence (shipped, sent, merged), not as a feeling. This is the section people skip and the section that pays at performance time.
4. Carry forward with a reason — every unfinished item either gets a fresh reason to continue or gets dropped explicitly. Silent carry-over is how dead tasks look alive.
5. Commit small — next week's commitments are few, specific, and checkable; a commitment list of fifteen items is next week's failure pre-written.

The log / 回顾日志

One dated file per review (a single reviews.md with newest on top works), so next week's step 1 is a read, not an excavation. The comparison across weeks is where drift becomes visible: the same item carried three weeks running is not a plan, it is a decision to make.

每周一条带日期的记录（一个 reviews.md 倒序放），下周的第一步是"读"而不是"考古"。跨周对比才让漂移显形：同一事项连顺三周，这就不是计划了，是一个待做的决定。

Install / 安装

npx skills add ChenneyZhuang/weekly-review

Per-agent paths: COMPATIBILITY.md. MIT. v0.1.0.

各 agent 安装路径见 COMPATIBILITY.md。MIT 许可，v0.1.0。