addressing-review-feedback
How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 122
- Forks
- 18
01 前言 Front Matter
- name
- addressing-review-feedback
- source
- obra/superpowers (MIT), adapted
- category
- workflow
- description
- How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.
03 SKILL.md 正文
Addressing Review Feedback
Overview
A need_revision task is somebody else's evidence that something is wrong, not a request to improvise. Treat it the way you would a bug report: verify, fix at the root, prove it with a test, and answer every point — silence on one is how a task bounces a second time for something you could have agreed or disagreed with up front.
The Process
- Read everything before touching code. Every numbered point on the task, plus the PR review comments when the round happened there (
list_task_comments,get_task_pull_requestwithinclude_diff: falsewhen you don't need the diff itself). If a point is unclear, re-read the code it's about before acting on a guess. Points can be related — a root cause can explain two of them at once. - Verify each claim against the code and, where you can, the running product — a reviewer can be wrong about where a line lives or what a function does. Classify each point: fix (the claim holds), disagree (with evidence — a file:line or a run that shows the current behavior is correct), or needs a human decision (a product call neither of you can settle from the code).
- Fix in order: blocking → simple → complex. Blocking = crash, security hole, or an acceptance criterion left unmet. Each fix gets its own guard test that fails before the fix and passes after (root-cause-debugging) — a fix with no failing-first test is not verified, it's hoped.
- Push back technically, never performatively. A point you disagree with gets a reasoned answer with evidence, not silent compliance and not silent skipping. Reply in the PR thread (
comment_on_pull_requestwithreply_to_comment_id) when it came from a review comment — a new top-level comment leaves the reviewer's thread looking unanswered; otherwiseadd_task_commentfor a task-comment point. No "You're absolutely right," no thanks, no preamble — state the finding. - A YAGNI check when a reviewer asks you to "implement it properly" or "handle the general case":
grep_codefor actual callers first. Build the general version only if more than one caller needs it; otherwise say so and keep the fix scoped to what broke. - Closing message: a point map. One line per numbered point — point → root cause → fix (file:line) → guard test, or point → why you disagree, with evidence. The architect's review of your revision reads this map against the diff; a revision with no map is a second bounce waiting to happen.
Red Flags
- A partial fix — three of four points addressed and the fourth left unmentioned.
- A fix with no guard test that failed first.
- Weakening or deleting the test that caught the problem instead of fixing the code.
- Silently agreeing with a wrong claim because arguing felt slower — a wrong "fix" based on a wrong claim bounces again, further down the line.
- A closing message that says "addressed all feedback" without the per-point map.
04 同仓库更多 Skills
visual-craft
暂无简介
ux-copy
Use when you write any words a user reads in a mockup or a hand-off - content language and voice, buttons, labels, errors, empty states, confirmations, numbers and dates, and the longest-string check
screen-mockup-html
Use when you draw screens as the design task's HTML mockup - ONE canvas document holding every screen and every variant side by side, the sanitizer-safe recipe, phone and wide frames, the states grid, static control states, and the loopback self-check with screenshots at 375, 768 and 1440 attached to the task
performance-awareness
How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.
design-variants
Use when a screen design needs variants - how many, variants as side-by-side rows on the one canvas, what makes two variants genuinely different, recommending one, how the human's choice is read, and narrowing to the chosen one in revision
design-system-authoring
Use when you derive, change or extend a design system - where tokens live per stack, the DTCG token tree, role-based names for both themes, base vs layer vs drift, the DESIGN.md section order, the inventory format, the contrast check and the lint every proposal returns