Claude Code Plugins

Community-maintained marketplace

Feedback

debugging

@nguyenthienthanh/aura-frog
1
0

Systematic debugging with root cause investigation. NO fixes without understanding cause first.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name debugging
description Systematic debugging with root cause investigation. NO fixes without understanding cause first.
autoInvoke true
priority high
model sonnet
triggers bug, error, not working, broken, crash, fix
allowed-tools Read, Grep, Glob, Bash

Debugging

Systematic debugging framework. Find root cause BEFORE fixing.

Core Principle

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST

Random fixes waste time and create new bugs.

When to Use

  • Test failures
  • Bugs and errors
  • Unexpected behavior
  • Performance issues
  • Build failures
  • Before claiming "fixed"

Four-Phase Process

Phase 1: Investigate

  1. Read the error - Full message, stack trace
  2. Reproduce - Confirm the bug exists
  3. Check recent changes - git diff, git log
  4. Gather evidence - Logs, console output

Phase 2: Analyze

  1. Find working example - Similar code that works
  2. Compare - What's different?
  3. Identify pattern - Where does it diverge?

Phase 3: Hypothesize

  1. Form theory - "It breaks because..."
  2. Test minimally - Smallest change to verify
  3. Confirm - Does fix work?

Phase 4: Fix

  1. Write test first - Reproduces bug
  2. Apply fix - Single focused change
  3. Verify - Test passes, no regressions

Root Cause Tracing

When error is deep in call stack:

1. Start at error location
2. Trace backward: where did bad data come from?
3. Continue tracing until source found
4. Fix at SOURCE, not symptom

Verification Protocol

Iron Law: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION

# Run verification
npm test  # or yarn test, pytest, etc.

# Check output
# THEN claim result

Wrong: "Should work now" Right: "Tests pass: [paste output]"

Red Flags (Stop & Follow Process)

  • "Quick fix for now"
  • "Just try changing X"
  • "It's probably X"
  • "Should work now"
  • "Seems fixed"

All mean: Return to Phase 1.

Quick Reference

Bug found
  → Phase 1: Read error, reproduce, check changes
  → Phase 2: Find working example, compare
  → Phase 3: Form hypothesis, test minimally
  → Phase 4: Write test, fix, verify

Deep stack error?
  → Trace backward to source
  → Fix at source, not symptom

About to claim fixed?
  → Run verification command
  → Read output
  → THEN claim result

References

For detailed techniques, see:

  • references/systematic-debugging.md - Full four-phase details
  • references/root-cause-tracing.md - Call stack tracing
  • references/verification.md - Verification protocols

Version: 2.0.0