Notre avis
Diagnostique systématiquement les bugs avant de proposer des correctifs, en suivant un processus en six étapes.
Points forts
- Structure claire et reproductible pour le débogage.
- Encourage la vérification des hypothèses avec des preuves.
- Évite les correctifs prématurés grâce à une phase de diagnostic distincte.
Limites
- Peut être trop rigide pour des bugs triviaux ou très simples.
- Nécessite une compréhension préalable du contexte du projet.
Utilisez cette compétence lorsqu'un bug est signalé et que vous avez besoin d'une approche méthodique pour en trouver la cause racine.
Ne l'utilisez pas lorsque le correctif est évident et immédiat, ou pour des tâches ne relevant pas du débogage.
Analyse de sécurité
PrudenceThe skill instructs using Bash to gather diagnostic evidence, which is legitimate for debugging, but the tool is powerful and could be exploited. No explicit destructive instructions, but caution is warranted.
- •Bash tool is allowed, which can execute arbitrary commands; used for diagnostic purposes but could be risky if misused.
Exemples
The app crashes when I click 'Save'. Please debug this systematically.Login endpoint returns 500 error. Diagnose the root cause.The report generator produces incorrect totals. Debug the data flow.name: debug version: 1.0.0 changelog: Initial debug skill with systematic diagnosis description: Diagnose bugs systematically before proposing fixes. Use when something is broken, failing, erroring, or not working as expected. allowed-tools: Read, Grep, Glob, Bash, Task temperature: 0.2 # Mostly systematic, slight exploration
Debug Phase
Diagnose systematically before proposing fixes.
Task
$ARGUMENTS
Instructions
Output: ## Debug Phase
State: At phase start, update STATE.md:
- Set
task:from $ARGUMENTS (if STATE.md is idle/complete or task is "None") - Set
phase: debug - Set
status: in_progress
1. Reproduce
- Confirm the failure exists
- Identify exact error message/behavior
- Note environment conditions (versions, config, state)
2. Isolate
- Narrow to smallest reproducible case
- Identify which component fails
- Trace information flow to failure point
- Use targeted reads/greps, not broad exploration
3. Hypothesize
- List possible causes (most likely first)
- For each: what evidence would confirm/reject?
- Consider: recent changes, dependencies, environment
4. Test Hypotheses
- Gather evidence systematically
- Update hypothesis ranking as evidence arrives
- Stop when root cause is confirmed
5. Produce Diagnosis
Use the diagnosis template:
## Symptom
[What's failing — exact error or behavior]
## Root Cause
[Why it's failing — the actual problem]
## Evidence
- `file:line` — observation
- Command output — what it showed
## Fix Options
1. Option A: [approach] — tradeoffs
2. Option B: [approach] — tradeoffs
## Recommended
[Which option and why]
Context Budget
- Target: complete diagnosis under 30% context
- Use subagents for exploration if needed
- Keep evidence concise — excerpts, not full files
6. Checkpoint
Update STATE.md incrementally:
- Set
phase: debug - Add root cause and fix decision to
## Decisions - Add investigated files to
## Key Files - Update
## Next Stepswith recommended fix
Run /checkpoint if context is heavy or taking a break.
Constraints
- Do not fix yet — diagnosis first
- If fix is obvious (<5 lines, isolated), may proceed with approval
- Otherwise, get approval before implementing
- Don't guess — verify with evidence
Exit Criteria
Root cause identified with evidence. Fix approach clear and approved.
Gate: "Here's the diagnosis. Approve fix approach?"
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.