메뉴
⚡ Debugging skill. Best for: 'why is X broken', 'find bug source', 'root cause analysis', 'trace error', 'debug issue'. Uses claudemem AST with context command for efficient call chain analysis.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/MadAppGang/claude-code --skill debugger-detective명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
| name | debugger-detective |
| description | ⚡ Debugging skill. Best for: 'why is X broken', 'find bug source', 'root cause analysis', 'trace error', 'debug issue'. Uses claudemem AST with context command for efficient call chain analysis. |
| allowed-tools | Bash, Task, Read, AskUserQuestion |
This skill uses claudemem's context command for debugging and root cause analysis.
| Task | claudemem | Native Tools |
|---|---|---|
| Trace backwards | callers shows call chain | Manual tracing |
| Trace forwards | callees shows dependencies | Manual tracing |
| Full context | context gives both directions | Multiple reads |
| Root cause | Call chain reveals origin | Trial and error |
Primary commands:
claudemem --agent context <name> - Full call chain (callers + callees)claudemem --agent callers <name> - Trace back to error sourceclaudemem --agent callees <name> - Trace forward from failure pointVersion: 3.3.0 Role: Debugger / Incident Responder Purpose: Bug investigation and root cause analysis using AST call chain tracing with blast radius impact analysis
You are investigating this codebase as a Debugger. Your focus is on:
context is Perfect for DebuggingThe context command shows you:
# Find the function mentioned in error
claudemem --agent symbol authenticate
# Get full context (callers + callees)
claudemem --agent context authenticate```
### Trace Back to Source (callers)
```bash
# Who called this function? (trace backwards)
claudemem --agent callers authenticate
# Follow the chain backwards
claudemem --agent callers LoginControllerclaudemem --agent callers handleRequest```
### Trace Forward to Effect (callees)
```bash
# What does this function call? (trace forward)
claudemem --agent callees authenticate
# Find where state changes happen
claudemem --agent callees updateSession```
### Blast Radius Analysis (v0.4.0+ Required)
```bash
# After finding the bug, check what else is affected
IMPACT=$(claudemem --agent impact buggyFunction)
if [ -z "$IMPACT" ] || echo "$IMPACT" | grep -q "No callers"; then
echo "No static callers - bug is isolated (or dynamically called)"
else
echo "$IMPACT"
echo ""
echo "This shows:"
echo "- Direct callers (immediately affected)"
echo "- Transitive callers (potentially affected)"
echo "- Complete list for testing after fix"
fi
Use for:
Limitations: Event-driven/callback architectures may have callers not visible to static analysis.
# Map error handling code
claudemem --agent map "throw error exception"
# Find specific error types
claudemem --agent symbol AuthenticationError
# Who throws this error?
claudemem --agent callers AuthenticationError```
### State Mutation Tracking
```bash
# Find where state changes
claudemem --agent map "set state update mutate"
# Find the mutation function
claudemem --agent symbol updateUserState
# Who calls this mutation?
claudemem --agent callers updateUserState```
## PHASE 0: MANDATORY SETUP
### Step 1: Verify claudemem v0.3.0
```bash
which claudemem && claudemem --version
# Must be 0.3.0+
Use AskUserQuestion (see ultrathink-detective for template)
# Check claudemem installation and index
claudemem --version && ls -la .claudemem/index.db 2>/dev/null
Before proceeding with investigation, verify the index is current:
# First check if index exists
if [ ! -d ".claudemem" ] || [ ! -f ".claudemem/index.db" ]; then
# Use AskUserQuestion to prompt for index creation
# Options: [1] Create index now (Recommended), [2] Cancel investigation
exit 1
fi
# Count files modified since last index
STALE_COUNT=$(find . -type f \( -name "*.ts" -o -name "*.tsx" -o -name "*.js" -o -name "*.jsx" -o -name "*.py" -o -name "*.go" -o -name "*.rs" \) \
-newer .claudemem/index.db 2>/dev/null | grep -v "node_modules" | grep -v ".git" | grep -v "dist" | grep -v "build" | wc -l)
STALE_COUNT=$((STALE_COUNT + 0)) # Normalize to integer
if [ "$STALE_COUNT" -gt 0 ]; then
# Get index time with explicit platform detection
if [[ "$OSTYPE" == "darwin"* ]]; then
INDEX_TIME=$(stat -f "%Sm" -t "%Y-%m-%d %H:%M" .claudemem/index.db 2>/dev/null)
else
INDEX_TIME=$(stat -c "%y" .claudemem/index.db 2>/dev/null | cut -d'.' -f1)
fi
INDEX_TIME=${INDEX_TIME:-"unknown time"}
# Get sample of stale files
STALE_SAMPLE=$(find . -type f \( -name "*.ts" -o -name "*.tsx" \) \
-newer .claudemem/index.db 2>/dev/null | grep -v "node_modules" | grep -v ".git" | head -5)
# Use AskUserQuestion (see template in ultrathink-detective)
fi
claudemem index
# Find where the error appears
claudemem --agent map "error message keywords"
# Or find the specific function
claudemem --agent symbol failingFunction```
### Phase 2: Get Full Context
```bash
# Get callers + callees in one command
claudemem --agent context failingFunction```
### Phase 3: Trace Backwards (Find Root Cause)
```bash
# For each caller, check if it's the source
claudemem --agent callers caller1claudemem --agent callers caller2
# Keep tracing until you find the root
# Once you suspect a root cause, verify the path
claudemem --agent callees suspectedRoot
# Does it lead to the symptom?
# What else calls the buggy code?
claudemem --agent callers buggyFunction
# These are all potentially affected
┌─────────────────────────────────────────────────────────┐
│ BUG INVESTIGATION │
├─────────────────────────────────────────────────────────┤
│ Symptom: User sees "undefined" in profile name │
│ Location: src/components/Profile.tsx:45 │
│ Error Type: Data inconsistency / Null reference │
│ Search Method: claudemem v0.3.0 (AST call chain) │
└─────────────────────────────────────────────────────────┘
❌ SYMPTOM: undefined rendered
└── src/components/Profile.tsx:45
└── user.name is undefined
↑ CALLER CHAIN (trace backwards):
└── useUser hook (src/hooks/useUser.ts:23)
↑
└── fetchUser API (src/api/user.ts:67)
↑
└── userMapper (src/mappers/user.ts:12)
↑
🔍 ROOT CAUSE FOUND HERE
🔍 ROOT CAUSE IDENTIFIED:
Location: src/mappers/user.ts:12
Problem: Field name mismatch
API Response: { fullName: "John Doe" }
Mapper Expects: { full_name: "..." }
Result: name = undefined
Evidence:
- callees of fetchUser → userMapper
- callers of userMapper → useUser → Profile
- Complete chain verified via context command
⚠️ OTHER AFFECTED CODE:
claudemem --agent callers userMapper shows:
- useUser hook (main app)
- useAdmin hook (admin panel)
- tests/user.test.ts
All 3 locations may have the same bug!
# Step 1: Find where undefined is used
claudemem --agent map "undefined null"
# Step 2: Get context of the failing function
claudemem --agent context renderProfile
# Step 3: Trace backwards through callers
claudemem --agent callers getUserData
# Step 4: Find where null was introduced
claudemem --agent callees fetchUser```
### Scenario: Race Condition
```bash
# Step 1: Find async operations
claudemem --agent map "async await promise"
# Step 2: Find shared state
claudemem --agent symbol sharedState
# Step 3: Who reads it?
claudemem --agent callers sharedState
# Step 4: Who writes it?
claudemem --agent callees updateState```
### Scenario: Incorrect Behavior
```bash
# Step 1: Find the function with wrong behavior
claudemem --agent symbol calculateTotal
# Step 2: What does it depend on?
claudemem --agent callees calculateTotal
# Step 3: Who provides input?
claudemem --agent callers calculateTotal```
## Result Validation Pattern
After EVERY claudemem command, validate results:
### Context Validation for Debugging
When tracing call chains:
```bash
CONTEXT=$(claudemem --agent context failingFunction)
EXIT_CODE=$?
# Check for failure
if [ "$EXIT_CODE" -ne 0 ]; then
DIAGNOSIS=$(claudemem status 2>&1)
# Use AskUserQuestion
fi
# Validate all sections present
if ! echo "$CONTEXT" | grep -q "\[symbol\]"; then
# Missing symbol section - function not found
# Use AskUserQuestion: Reindex, Different name, or Cancel
fi
if ! echo "$CONTEXT" | grep -q "\[callers\]"; then
# Missing callers - may be entry point or index issue
# Entry points (API handlers, main) have 0 callers - this is expected
fi
if ! echo "$CONTEXT" | grep -q "\[callees\]"; then
# Missing callees - may be leaf function or index issue
# Leaf functions (console.log, throw) have 0 callees - this is expected
fi
CALLERS=$(claudemem --agent callers suspectedBug)
# 0 callers could mean:
# 1. Entry point (main, API handler) - expected
# 2. Dead code - use dead-code command (v0.4.0+)
# 3. Dynamically called - check for import(), eval, reflection
if echo "$CALLERS" | grep -qi "error\|not found"; then
# Actual error vs no callers
# Use AskUserQuestion
fi
CRITICAL: Never use grep/find/Glob without explicit user approval.
If claudemem fails or returns irrelevant results:
claudemem status// Fallback options (in order of preference)
AskUserQuestion({
questions: [{
question: "claudemem bug investigation failed or found no call chain. How should I proceed?",
header: "Debugging Issue",
multiSelect: false,
options: [
{ label: "Reindex codebase", description: "Run claudemem index (~1-2 min)" },
{ label: "Try different function name", description: "Search for related functions" },
{ label: "Use grep (not recommended)", description: "Traditional search - loses call chain tracing" },
{ label: "Cancel", description: "Stop investigation" }
]
}]
})
See ultrathink-detective skill for complete Fallback Protocol documentation.
| Anti-Pattern | Why Wrong | Correct Approach |
|---|---|---|
grep "error" | No call relationships | claudemem --agent context func |
| Read random files | No direction | Trace callers/callees systematically |
| Fix symptom only | Bug returns | Trace to root cause with callers |
| Skip impact check | Miss related bugs | ALWAYS check all callers |
cmd | head/tail | Hides call chain context | Use full output or --max-depth |
╔══════════════════════════════════════════════════════════════════════════════╗ ║ ║ ║ ❌ Anti-Pattern 7: Truncating Claudemem Output ║ ║ ║ ║ FORBIDDEN (any form of output truncation): ║ ║ → BAD: claudemem --agent map "query" | head -80 ║ ║ → BAD: claudemem --agent callers X | tail -50 ║ ║ → BAD: claudemem --agent search "x" | grep -m 10 "y" ║ ║ → BAD: claudemem --agent map "q" | awk 'NR <= 50' ║ ║ → BAD: claudemem --agent callers X | sed '50q' ║ ║ → BAD: claudemem --agent search "x" | sort | head -20 ║ ║ → BAD: claudemem --agent map "q" | grep "pattern" | head -20 ║ ║ ║ ║ CORRECT (use full output or built-in limits): ║ ║ → GOOD: claudemem --agent map "query" ║ ║ → GOOD: claudemem --agent search "x" -n 10 ║ ║ → GOOD: claudemem --agent map "q" --tokens 2000 ║ ║ → GOOD: claudemem --agent search "x" --page-size 20 --page 1 ║ ║ → GOOD: claudemem --agent context Func --max-depth 3 ║ ║ ║ ║ WHY: Output is pre-optimized; truncation hides critical results ║ ║ ║ ╚══════════════════════════════════════════════════════════════════════════════╝
After completing investigation, report search feedback to improve future results.
Report feedback ONLY if you used the search command during investigation:
| Result Type | Mark As | Reason |
|---|---|---|
| Read and used | Helpful | Contributed to investigation |
| Read but irrelevant | Unhelpful | False positive |
| Skipped after preview | Unhelpful | Not relevant to query |
| Never read | (Don't track) | Can't evaluate |
# Track during investigation
SEARCH_QUERY="your original query"
HELPFUL_IDS=""
UNHELPFUL_IDS=""
# When reading a helpful result
HELPFUL_IDS="$HELPFUL_IDS,$result_id"
# When reading an unhelpful result
UNHELPFUL_IDS="$UNHELPFUL_IDS,$result_id"
# Report at end of investigation (v0.8.0+ only)
if claudemem feedback --help 2>&1 | grep -qi "feedback"; then
timeout 5 claudemem feedback \
--query "$SEARCH_QUERY" \
--helpful "${HELPFUL_IDS#,}" \
--unhelpful "${UNHELPFUL_IDS#,}" 2>/dev/null || true
fi
Include in investigation report:
Search Feedback: [X helpful, Y unhelpful] - Submitted (v0.8.0+)
symbol to find where error appearscontext for callers + callees togethercallers chain to root causecallees to confirm the pathcallers of buggy code may be affectedcontext is your primary tool - Shows full call chaincallers - Find root cause, not just symptomcallees - Confirm the execution pathMaintained by: MadAppGang Plugin: code-analysis v2.7.0 Last Updated: December 2025 (v3.3.0 - Cross-platform compatibility, inline templates, improved validation)
SOC 직업 분류 기준
Use when starting isolated feature work or before executing implementation plans. Manages full worktree lifecycle from creation through cleanup with safety checks and error recovery.
Configuration reference and troubleshooting for the statusline plugin — sections, themes, bar widths, and script architecture
Common agent patterns and templates for Claude Code. Use when implementing agents to follow proven patterns for Tasks integration, quality checks, and external model invocation via claudish CLI.