root-cause-review
Use when a task in code_review was sent back before (an earlier need_revision comment exists) - verify every prior point was fixed at its root
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 122
- Forks
- 18
01 前言 Front Matter
- name
- root-cause-review
- category
- quality
- description
- Use when a task in code_review was sent back before (an earlier need_revision comment exists) - verify every prior point was fixed at its root
03 SKILL.md 正文
Root-Cause Review
Overview
When a task returns to code_review after a need_revision cycle, the question is not "did something change" but "was the ROOT CAUSE addressed". Symptom patches come back as bugs; catching them here is cheaper than in QA or production.
The Process
- Find the prior feedback first. A re-submission enters code_review from
in_progress, so the earlier need_revision comment is NOT injected into this run automatically —list_task_commentson this task is how you find it (also checkget_task_pull_requestfor human reviewer comments on the PR itself). - Re-read the prior feedback: your previous need_revision comment, the QA failure report, or the pipeline failure — list every numbered point.
- Map each point to the new diff: for every point, find the exact change that addresses it. A point with no matching change means the revision is incomplete → back to need_revision citing the unaddressed point.
- Judge root cause vs symptom for each fix:
- Symptom patch: special-casing the failing input, catching-and-ignoring the error, widening a timeout, deleting the failing assertion.
- Root-cause fix: the change explains WHY the failure happened and removes that mechanism at its source.
- Check the guard test: a test now exists that reproduces the original failure and passes with the fix. No guard test → Important finding.
- Watch for regressions: revision diffs are written under pressure — re-check the surrounding code paths the fix touches.
Worked Example (symptom vs root cause)
Prior finding: "export returns empty CSV on DB error (swallowed error)."
- Resubmission A (symptom patch): the diff wraps the export call in the handler with
if len(csv)==0 { return 500 }. Rejected → back to need_revision: "Point 1: the swallowed error at service.go:24 is still there; you inferred failure from an empty result instead of propagating the error. An empty project also returns empty CSV, so this now 500s a valid empty project. Fix the source." - Resubmission B (root cause):
ListByProject's error is propagated and mapped to 500; a test injects a repo error and asserts 500; the empty-project test still asserts 200 + header-only CSV. Approved → ready_for_qa.
The tell was step 2: B's fix touched service.go:24 (where the mechanism lives); A's didn't.
Verdict
- Every prior point addressed at the root, guard tests present, pipeline green → ready_for_qa.
- Any symptom patch, unaddressed point, or missing guard test → need_revision, quoting the specific gap ("Point 3 from the previous review: ... — the new diff only catches the exception; the invalid state is still produced upstream in ...").
Red Flags
- The diff deletes or weakens a failing test instead of fixing the behavior.
- A fix touches none of the files where the failure mechanism lives.
- The fix's commit message or the revision's diff says "fixed" without showing what caused the failure — you never see the developer's closing message, so judge the code and the guard test, not a claim.
- Third revision cycle on the same task — question the design: comment on whether the approach itself is wrong and consider recommending a re-analysis instead of a fourth patch. The system backs this up: three moves into need_revision without a human touch parks the task automatically, so this is also where a real disagreement belongs, not a further round-trip.
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