πDebugger
Tracks down the root cause of your bug instead of patching the symptom
What it does
What it does: Diagnoses and fixes bugs by isolating the actual root cause, not just the surface symptom, then leaves behind a test so it can't come back.
- Runs a six-step decision tree β reproduce, confirm observed vs. expected, rank 2-3 hypotheses, falsify the most likely one with a cheap experiment, fix with a regression test, document the root cause
- For production incidents, starts with the three observability pillars: distributed traces (first failing span), correlated logs (Β±2 min around the error), and change correlation (deploys/config/flags in the prior 30 min) before touching code
- Handles the hard classes of bugs: memory leaks and corruption (use-after-free, double-free, heap analysis), concurrency (race conditions, deadlocks, lock ordering), and performance bottlenecks (CPU/memory/IO profiling, slow queries)
- Applies divide-and-conquer, git bisect, differential and statistical debugging, plus stack-trace and core-dump analysis
- Ships with Read, Write, Edit, Bash, Glob, and Grep so it can grep logs, trace code paths, and apply the fix directly
Installation
- Make sure Claude is on your device and in your terminal.
Agents load from
~/.claude/agents/when Claude Code starts. If you don't have Claude Code yet, install it once with the command below, then runclaudein any terminal to verify.One-time setupnpm i -g @anthropic-ai/claude-codeAlready have it? Skip ahead.
- Paste into your terminal.
Downloads the agent into
~/.claude/agents/debugger-wshobson.md. Safe to re-run; it just overwrites. - Restart Claude Code.
Quit and reopen Claude Code. New agents are picked up on startup.
- Use it.
Claude delegates to the agent when your ask matches its description β phrases like "review this," "plan this," "audit this." You can also invoke directly: "Use the debugger-wshobson agent toβ¦"
Prefer to read the source first? Open on GitHub.