Mode d’emploi
- Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
- Consultez le document original pour les outils, identifiants et environnements requis.
- Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/krishnakanthb13/antigravity_global_skills --skill code_reviewAvant de commencer
- L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
- Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.
Code Review Checklist
Use this skill to conduct a thorough, senior-level code review.
Inputs
- Target: The file or directory to review.
- Focus: (Optional) Specific area to check (e.g., security, performance).
Tooling Strategy
- Use
view_fileto read the code. - Use
grep_searchif looking for specific patterns across multiple files.
Workflow
1. Functionality & Logic (The "Does it Work?" Check)
- Requirements: Does this solving the stated problem?
- Edge Cases: Nulls, empty states, limits (min/max).
- Error Handling: Are errors caught, logged, and handled gracefully (no silent failures)?
2. Security (The OWASP Check)
- Injection: Check for raw SQL,
innerHTML, oreval(). - ❌ Bad:
db.query("SELECT * FROM users WHERE name = " + input) - ✅ Good:
db.query("SELECT * FROM users WHERE name = $1", [input]) - Secrets: Ensure NO API keys or passwords are committed. (Use
.env). - Auth: Are permission checks present on sensitive routes?
3. Performance (The Scalability Check)
- Loops: No expensive operations inside loops (e.g., DB calls).
- N+1 Problems: Check for fetching data in a loop instead of batches.
- Memory: Are listeners/connections cleaned up?
- Code Quality (The Maintainability Check)
- Naming: Variables should be intentional (
isValidvsv). - DRY (Don't Repeat Yourself): Can distinct logic be extracted to a helper?
- SOLID Principles:
- S: Does the class/function do only ONE thing?
- O: Can we add features without changing this code?
- L: Can subclasses be used interchangeably?
- I: Are interfaces too bloated?
- D: Is it hard-coded to a specific implementation? (Use dependency injection).
- Comments: Explain WHY, not WHAT.
5. Git & Hygiene
- Commits: Are they atomic and descriptive?
- Leftovers: No
console.log(debug only), commented-out code, or TODOs without an issue tracking them.
Output Format
Structure your review like this:
Summary: [High level opinion - Ready to Merge / Changes Requested]
| Category | Status | Notes |
|---|---|---|
| Functionality | ✅ / ⚠️ | ... |
| Security | ✅ / 🔴 | ... |
| Performance | ✅ / ⚠️ | ... |
Detailed Comments:
- [File/Line]: [Issue description]
- Suggestion: [Code snippet check]
Comprehensive Review Guidelines
1. The Review Mindset
- Goals: Catch bugs, ensure maintainability, share knowledge, improve architecture.
- Not Goals: Nitpicking formatting, showing off, rewriting to personal preference.
- Provide Effective Feedback: Be specific, actionable, educational, and focused on the code, not the person. Suggest, don't command.
2. Differentiate Severity (The Label System)
Use explicit labels to set expectations:
- 🔴
[blocking]: Must fix before merge (e.g., security flaws, breaking bugs). - 🟡
[important]: Should fix, discuss if disagree (e.g., architectural issues, N+1 queries). - 🟢
[nit]: Nice to have, not blocking (e.g., minor naming improvements). - 💡
[suggestion]: Alternative approach to consider. - 📚
[learning]: Educational comment, no action needed. - 🎉
[praise]: Acknowledge good work and solid patterns!
3. Review Techniques
- The Question Approach: Ask questions to encourage thinking (e.g., "What happens if
itemsis empty?" instead of "This fails if empty"). - Collaborative Language: Use suggestions (e.g., "Would it make sense to extract this?" instead of "Extract this").
- Avoid Implementation Detail Reviews: Ensure tests describe behavior, not internal state (e.g., test what happens on click, not that an internal counter incremented).
4. Handling Disagreements
If there's disagreement: Seek to understand the author's approach, acknowledge valid points, request data (like a benchmark) if needed, and know when to let go if it's working and not critical.