doraemonkeys/claude-code-debug-mode

88 stars · Last commit 2026-03-31

Cursor-style Debug Mode skill for coding agents. Hypothesis-driven debugging with runtime log instrumentation and human-in-the-loop verification.

README preview

# Debug Mode Skill (Claude Code + Codex + Gemini CLI)

A hypothesis-driven debugging skill for [Claude Code](https://docs.anthropic.com/en/docs/claude-code), Codex, and [Gemini CLI](https://github.com/google-gemini/gemini-cli). Instead of blindly guessing fixes, it instruments your code with runtime logs, generates multiple hypotheses, and iteratively narrows down the root cause — with you in the loop.

Inspired by [Cursor's Debug Mode](https://www.cursor.com/blog/introducing-debug-mode).

## How It Works

```
Bug Report → Hypotheses → Instrument Code → Reproduce → Analyze Logs → Fix → Verify → Clean Up
```

1. **Understand** — Gathers context about the bug (expected vs actual behavior, repro steps)
2. **Hypothesize** — Generates 3–5 testable hypotheses about root causes
3. **Instrument** — Adds targeted debug logging wrapped in `#region DEBUG` blocks
4. **Reproduce** — You trigger the bug while logs are collected to `.claude/debug.log`
5. **Diagnose** — Maps log output to hypotheses, confirms or rules out each one
6. **Fix** — Writes a minimal, targeted fix (not a refactor)
7. **Verify** — You confirm the fix works; if not, the cycle repeats
8. **Clean up** — Removes all instrumentation, leaving a clean diff

View full repository on GitHub →