agentskills.codes
BU

Investigate and fix software defects with reproducible evidence and minimal-risk patches. Use for crashes, regressions, flaky behavior, and logic faults in frontend or backend. Keywords: debug, root cause, repro, regression.

Install

mkdir -p .claude/skills/bug-hunter && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15079" && unzip -o skill.zip -d .claude/skills/bug-hunter && rm skill.zip

Installs to .claude/skills/bug-hunter

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Investigate and fix software defects with reproducible evidence and minimal-risk patches. Use for crashes, regressions, flaky behavior, and logic faults in frontend or backend. Keywords: debug, root cause, repro, regression.
224 chars✓ has a “when” trigger

About this skill

Bug Hunter

Mission

Find the true fault, apply the smallest durable fix, and prove it will not regress.

Use This Skill When

  • A feature is broken and the failure reason is unclear
  • A regression appeared after recent changes
  • The same issue is hard to reproduce consistently
  • You need a minimal fix with regression protection

Non-Goals

  • Do not perform broad refactors while debugging.
  • Do not mark root cause as confirmed without evidence.

Required Inputs

  • Expected behavior
  • Actual behavior
  • Reproduction steps and environment details
  • Logs, stack traces, or failing test output if available

Procedure

  1. Define failure contract: expected vs actual, scope, impact.
  2. Reproduce minimally; if not reproducible, list variability factors and rank likely causes.
  3. Trace code path and state transitions around the fault boundary.
  4. Build 1-3 hypotheses and eliminate alternatives using direct evidence.
  5. Implement smallest durable fix at the causal layer.
  6. Add or update regression test for the failed behavior.
  7. Re-run targeted validations and nearby critical flows.

Edge Cases To Check

  • Null, empty, malformed, and extreme input values
  • Race conditions and async ordering
  • Permission and role-dependent paths
  • Retry/timeout behavior and partial failures
  • Environment-specific config drift

Exit Criteria

  • Reproduction is documented or uncertainty is explicitly bounded.
  • Root cause is evidenced and fix is applied/proposed.
  • Validation evidence is provided for fix and adjacent flow safety.

Output Format

  1. Reproduction summary
  2. Confirmed root cause
  3. Applied or proposed fix
  4. Validation evidence
  5. Regression-prevention checklist

Search skills

Search the agent skills registry