deep-debug
|
Works with
---
name: deep-debug
description: |
license: MIT
---
# Deep Debug - Multi-Agent Investigation
**Status**: Production Ready
**Last Updated**: 2026-02-03
**Dependencies**: Chrome MCP tools (optional), debugger agent, code-reviewer agent
---
## When to Use
- **Going in circles** - You've tried multiple fixes but nothing works
- **Browser + API interaction bugs** - Need to see Network tab, console logs
- **Symptoms don't match expectations** - Something deeper is wrong
- **Complex state management bugs** - React hooks, closures, race conditions
## Commands
| Command | Purpose |
|---------|---------|
| `/debug` | Guided debugging workflow with evidence gathering and parallel investigation |
## Quick Start
```
/debug [description of the bug]
```
Or invoke naturally:
- "I'm stuck on this bug, can you do a deep investigation?"
- "This bug is resisting normal debugging"
- "I need you to really dig into this"
---
## The Process
### Phase 1: Gather Evidence (Before Hypothesizing)
**For browser-related bugs**, use Chrome MCP tools:
```typescript
// Get network requests (look for duplicates, failures, timing)
mcp__claude-in-chrome__read_network_requests
// Get console logs (errors, warnings, debug output)
mcp__claude-in-chrome__read_console_messages
// Get page state
mcp__claude-in-chrome__read_page
```
**For backend bugs**, gather:
- Error logs and stack traces
- Request/response payloads
- Database query logs
- Timing information
### Phase 2: Launch Parallel Investigation (3 Agents)
Launch these agents simultaneously with the gathered evidence:
#### Agent 1: Execution Tracer (debugger)
```
Task(subagent_type="debugger", prompt="""
EVIDENCE: [paste network/console evidence]
Trace the execution path that leads to this bug. Find:
1. Where the bug originates
2. What triggers it
3. The exact line/function causing the issue
Focus on TRACING, not guessing.
""")
```
#### Agent 2: Pattern Analyzer (code-reviewer)
```
Task(subagent_type="code-reviewer", prompt="""
EVIDENCE: [paste evidence]
Review the relevant code for common bug patterns:
- React useCallback/useMemo dependency issues
- Stale closures
- Race conditions
- State update ordering
- Missing error handling
Find patterns that EXPLAIN the evidence.
""")
```
#### Agent 3: Entry Point Mapper (Explore)
```
Task(subagent_type="Explore", prompt="""
EVIDENCE: [paste evidence]
Map all entry points that could trigger this behavior:
- All places [function] is called
- All event handlers involved
- All state updates that affect this
Find if something is being called MULTIPLE TIMES or from UNEXPECTED places.
""")
```
### Phase 3: Cross-Reference Findings
After agents complete, look for:
| Signal | Meaning |
|--------|---------|
| All 3 agree on root cause | High confidence - fix it |
| 2 agree, 1 different | Investigate the difference |
| All 3 different | Need more evidence |
### Phase 4: Verify Fix
After implementing the fix, re-gather the same evidence to confirm:
- Network tab shows expected behavior
- Console has no errors
- State updates correctly
---
## Real Example: Duplicate API Calls Bug
### Evidence Gathered
Network tab showed **2 fetch requests** for the same message.
### Parallel Investigation Results
| Agent | Finding |
|-------|---------|
| debugger | `state.messages` in useCallback deps causes callback recreation |
| code-reviewer | Same finding + explained React pattern causing it |
| Explore | Verified UI layer wasn't double-calling (ruled out) |
### Root Cause (Consensus)
`sendMessage` useCallback had `state.messages` in dependency array. Every state update recreated the callback, causing duplicate calls during React re-renders.
### Fix
Use `stateRef` to access current state without including in dependencies:
```typescript
const stateRef = useRef(state);
stateRef.current = state;
const sendMessage = useCallback(async (text) => {
// Use stateRef.current instead of state
const messages = stateRef.current.messages;
// ...
}, [/* state.messages removed */]);
```
---
## Common Bug Patterns This Catches
### React Hook Issues
- `state` in useCallback dependencies causing recreation
- Missing dependencies causing stale closures
- useEffect running multiple times
### API/Network Issues
- Duplicate requests (visible in Network tab)
- Race conditions between requests
- CORS failures
- Timeout handling
### State Management Issues
- State updates not batching correctly
- Optimistic updates conflicting with server response
- Multiple sources of truth
---
## Chrome Tools Quick Reference
| Tool | Use For |
|------|---------|
| `read_network_requests` | See all fetch/XHR calls, duplicates, failures |
| `read_console_messages` | Errors, warnings, debug logs |
| `read_page` | Current DOM state |
| `javascript_tool` | Execute debug code in page context |
---
## Tips for Success
1. **Evidence first, hypotheses second** - Don't guess until you have concrete data
2. **Network tab is gold** - Most frontend bugs show symptoms there
3. **Parallel agents save time** - 3 perspectives at once vs sequential guessing
4. **Cross-reference findings** - Agreement = confidence
5. **Verify with same evidence** - Confirm fix with same tools that found bugMore Debugging skills
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
explore-code
lllllllama/rigorpilot-skills
Rigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated branch or worktree to transplant modules, adapt a backbone, add LoRA or adapter layers, replace a head, or stitch together meaningful low-risk migration ideas with rollback-aware records in `explore_outputs/`. Do not use for end-to-end exploration orchestration on top of `current_research`, trusted baseline reproduction, conservative debugging, environment setup, verified contribution claims, or default repository analysis.
safe-debug
lllllllama/rigorpilot-skills
Rigor Debug / Rigor Audit skill for deep learning research work. Use when the user pastes a traceback, terminal error, CUDA OOM, checkpoint load failure, shape mismatch, NaN loss symptom, or training failure and wants conservative diagnosis before any patching, with debug fixes clearly separated from research contributions. Do not use for broad refactoring, speculative adaptation, automatic exploratory patching, or general repository familiarization.

