Review Pattern Mining
SkillDev toolsLets your agent group code review comments into broader fixes across the whole diff instead of one line at a time.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Review Pattern Mining skill
About this capability
Mine review comments for broader diff-wide fixes instead of handling each line in isolation
What this skill tells your AI
The instructions your AI receives, as published by udecode/plate in .agents/skills/review-sweep/SKILL.md and read by ahel’s review.
Use this when handling PR review feedback and you want to reduce repeated follow-up review on the same branch.
Goal
Treat each review comment as a chance to infer a broader rule, then apply that rule across the rest of the current diff where it clearly fits.
Core Rules
- Extract the underlying rule, not just the literal line edit.
- Fix the same pattern everywhere in the current diff when the rule is objective and low-ambiguity.
- Do not cargo-cult a reviewer suggestion across unrelated code.
- If the feedback is architectural or taste-heavy, decide once and either:
- refactor intentionally across the relevant seam, or
- prepare a concise pushback draft with reasoning.
- Prefer one coherent follow-up refactor over many tiny copy-paste fixes.
- Keep the blast radius proportional to confidence.
- If a suggested sweep depends on another PR or pending upstream change, call that out explicitly instead of guessing.
Good Sweep Targets
- Safety rules:
- input validation
- index or key guards
- dangerous property filtering
- null or undefined handling
- Repeated docs issues:
- incomplete explanations
- drift between API docs and concept docs
- missing tradeoff or usage guidance
- Repeated code-shape issues:
- redundant casts
- awkward conditionals
- repeated helper patterns
- duplicated comments or missing comments at the same seam
- Repeated test issues:
- brittle setup
- over-specific selectors
- restored assertions that should exist in similar tests
Bad Sweep Targets
- Naming preferences without a strong readability win
- Broad architecture changes without repo evidence
- Subjective style choices that only weakly generalize
- Changes that widen public API just to satisfy local convenience
- Anything you cannot explain in one sentence
Workflow
- Read the review comment and classify it:
- safety
- docs clarity
- code-shape cleanup
- test robustness
- architecture or taste
- Write down the inferred rule in one sentence.
- Search the rest of the current diff for the same pattern.
- Split findings into:
- clear applies
- maybe applies
- does not apply
- Apply only the clear set automatically.
- For the maybe set, make one explicit decision with reasoning.
- Verify the changed seam directly.
- In your handoff, separate:
- the original comment you handled
- the additional diff-wide fixes you inferred from it
Output Rule
When summarizing the work, make it obvious that you did not just answer the comment literally. Say what broader rule you inferred and where else in the diff you applied it.
Signals
- GitHub stars
- 17k
- Forks
- 996
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
review-sweep- Source
- github.com/udecode/plate