LLM Compact Logger Analysis

Quick Start

When analyzing llm-compact-logger output:

1. Read debug-compact.json first (~18 tokens vs ~88 for full report)
2. Check for meta.rootCauses - patterns found across failures (confidence >0.85 is actionable)
3. Review topFails - most common error types
4. Examine code field - failing line with context
5. Look for history.flaky - flakiness indicators

Use jq for queries (not grep): See EXAMPLES.md for jq query patterns.

For detailed format reference and error patterns, see REFERENCE.md.

Analyzing Test Output

Root Cause First

If meta.rootCauses exists, lead with the highest confidence pattern:

{
  "meta": {
    "rootCauses": [{
      "pattern": "Accessing property 'name' on undefined",
      "confidence": 0.95,
      "suggestion": "Check if object exists before accessing"
    }]
  }
}

Confidence >0.85 means actionable insight found.

Error Pattern Analysis

Group failures by topFails:

{"topFails": [{"type": "TypeError", "count": 5}]}

Common fixes:

• TypeError undefined → Add null checks
• Type mismatch → API structure changed
• Array length → Data filtering issue

Code Context

Display failing line from code field:

{
  "code": {
    "fail": "expect(result).toBe(5)",
    "ctx": ["const result = calculate();", "expect(result).toBe(5);"]
  }
}

Flakiness Detection

Flag tests with history.flaky.isFlaky: true.

Configuring Enhancements

Ask user:

1. Need history tracking? → Enable persistentIndex (requires better-sqlite3)
2. Using coverage? → Enable Vitest coverage with provider: 'v8', reporter: ['json']
3. Have flaky tests? → Recommend persistentIndex for historical analysis

Minimal setup: new VitestReporter({ outputDir: './debug' })

See REFERENCE.md for complete configuration options and examples.

Response Format

Structure your analysis as:

📊 Summary: X tests (X passed, X failed) - X% pass rate

🎯 Root Cause: [Pattern] (X% confidence)
Suggestion: [Fix]

❌ Key Failures:
1. [Test] (file:line)
   Fix: [Solution]

💡 Next Steps: [Actions]

When to Use This Skill

Activate when:

• User shares files ending in debug-compact.json or debug-report.json
• User asks to analyze test failures
• User mentions llm-compact-logger configuration
• User asks about test flakiness

Troubleshooting

Common issues:

• "better-sqlite3 not installed" → Install or disable persistentIndex
• "Coverage file not found" → Enable Vitest coverage with reporter: ['json']
• Missing enhancements → Requires v0.2.0+

See REFERENCE.md for complete troubleshooting guide.