Notre avis
Cette compétence aide un agent IA à traiter les commentaires de révision de Pull Request en vérifiant indépendamment chaque commentaire, en appliquant ou rejetant les modifications selon leur exactitude, et en demandant une nouvelle révision.
Points forts
- Vérifie indépendamment les retours avant de les appliquer, réduisant l'acceptation aveugle de suggestions incorrectes.
- Catégorise les retours par sévérité et suit chaque commentaire.
- Fournit une réponse structurée et un résumé pour la re-révision.
Limites
- Nécessite que l'agent ait accès au dépôt et puisse exécuter un serveur de développement local.
- Peut ne pas couvrir tous les cas extrêmes de tests visuels ou de logique de sécurité.
- Dépend de la capacité de l'agent à raisonner sur l'exactitude du code.
Utilisez lorsque vous devez répondre systématiquement à une Pull Request révisée et vous assurer que seuls les retours valides sont appliqués.
Évitez pour les PR triviales sans retour ou lorsqu'un humain doit prendre des décisions subjectives de conception.
Analyse de sécurité
SûrThe skill uses common development commands (npm, npx, gh) for legitimate PR review purposes. It emphasizes independent verification of feedback before applying changes, and does not instruct any destructive or exfiltrating actions. While powerful tools are used, the risk is minimal in a controlled development environment.
Aucun point d'attention détecté
Exemples
Address review comments on PR #15 with independent verification.Process and respond to feedback on pull request #42 using the address-review skill.Verify and apply review suggestions for PR #123, handling escalations.name: address-review description: Address PR review feedback, verify independently, and re-request review disable-model-invocation: true argument-hint: <pr-number>
Address Review Feedback
Address review comments on PR #$ARGUMENTS. Be adversarial about the feedback itself — verify that suggestions are correct before applying them.
PR Context
!gh pr view $ARGUMENTS
!gh api repos/{owner}/{repo}/pulls/$ARGUMENTS/comments --jq '.[] | "---\n\(.path):\(.line // .original_line)\n\(.body)\n"'
!gh pr view $ARGUMENTS --comments
Instructions
Step 1: Gather All Feedback
Read every review comment above. Categorize each piece of feedback:
- Action Required — Reviewer flagged as blocking merge
- Recommended — Reviewer suggested addressing before merge
- Minor — Nits, style suggestions
Step 2: Independently Verify Each Finding
Do not blindly apply suggestions. For each piece of feedback:
- Read the file in question — Use Glob/Grep/Read to find the relevant file and understand the full context, not just the diff snippet.
- Assess whether the feedback is correct — Reviewers (including agent reviewers) can be wrong. Check:
- Does the suggested change actually fix the issue identified?
- Could the suggestion introduce a new bug or visual regression?
- Is the reviewer missing context that makes the current code correct?
- Does the suggestion align with project conventions in
AGENTS.md?
- For visual feedback — Check out the branch, run
npm run dev, verify rendering at mobile and desktop widths, check both variants (/startand/build). - For security feedback — Take it seriously by default. Security suggestions should be applied unless you can clearly demonstrate they're wrong.
Step 3: Respond to Each Finding
For each piece of feedback, take one of these actions:
Apply — The feedback is correct. Make the change.
- Edit the file
- Run
npm run devand verify both variants render correctly - Note what was changed
Partially apply — The core insight is right but the suggested fix isn't quite right.
- Implement a better fix that addresses the underlying concern
- Explain why you deviated from the exact suggestion
Reject with justification — The feedback is incorrect or doesn't apply.
- Explain clearly why the current code is correct
- Reference
AGENTS.mdor project conventions to support your reasoning - Never reject feedback without a concrete justification
Escalate — You're unsure whether the feedback is valid.
- Flag it to the human with the evidence for and against
- Do not guess or silently skip
Step 4: Run Full Verification
After addressing all feedback:
- Run
npm run devand verify both variants render correctly at mobile and desktop widths - Check TypeScript compilation —
npx wrangler types && npx tsc --noEmit(Worker-side only) - Test form submission if any API or form changes were made
- Spot-check your changes — Read through your own diff. Did you introduce any new issues while fixing the review feedback?
- Check sizing — If the fixes significantly expanded the PR, flag whether it should be split.
Step 5: Commit and Push
- Commit fixes with clear messages linking to the review feedback:
fix: Address review — <description> - Keep fix commits separate from each other when they address unrelated feedback (easier to review the re-review)
- Push to the PR branch
Step 6: Post Summary and Re-request Review
Post a comment on the PR summarizing how each finding was addressed, then re-request review:
gh pr comment $ARGUMENTS --body "<summary>"
gh pr edit $ARGUMENTS --add-reviewer <reviewer>
Use this format for the summary:
## Review Feedback Addressed
| # | Finding | Action | Details |
|---|---------|--------|---------|
| 1 | <brief description> | Applied / Partially applied / Rejected | <what was done and why> |
| 2 | ... | ... | ... |
**Visual verification:** Both variants checked at mobile + desktop
**New commits:** <list of fix commits>
For any rejected findings, provide the full justification in the Details column so the reviewer can evaluate your reasoning.
Expert Next.js App Router
Developpement
Un skill qui transforme Claude en expert Next.js App Router.
Générateur de README
Developpement
Crée des README.md professionnels et complets pour vos projets.
Rédacteur de Documentation API
Developpement
Génère de la documentation API complète au format OpenAPI/Swagger.