Flaky Detect

Skill access pattern (post-kernel-pivot, 2026.5+)

Skill names referenced in this document are AIWG skills, not slash commands. Most are not kernel-listed and cannot be invoked as /skill-name by the platform. Reach them via:

aiwg discover "<capability>"
aiwg show skill <name>

Only kernel-listed skills (aiwg-doctor, aiwg-refresh, aiwg-status, aiwg-help, use, steward) are directly invokable as slash commands. See skill-discovery rule.

Flaky Detect Skill

Purpose

Identify flaky tests (tests that pass and fail non-deterministically) by analyzing CI history, execution patterns, and test characteristics. Google research shows 4.56% of tests are flaky, costing millions in developer productivity.

Research Foundation

When This Skill Applies

• User reports "tests sometimes fail" or "intermittent failures"
• CI has been unstable or unreliable
• User wants to audit test suite reliability
• Pre-release quality assessment
• Debugging non-deterministic behavior

Trigger Phrases

Flaky Test Taxonomy (Google Research)

Detection Methods

1. CI History Analysis

Parse GitHub Actions / CI logs to find inconsistent results:

def analyze_ci_history(repo, days=30):
    """Analyze CI runs for flaky patterns"""
    runs = get_ci_runs(repo, days)
    test_results = {}

    for run in runs:
        for test in run.tests:
            if test.name not in test_results:
                test_results[test.name] = {"pass": 0, "fail": 0}

            if test.passed:
                test_results[test.name]["pass"] += 1
            else:
                test_results[test.name]["fail"] += 1

    # Identify flaky tests (pass rate between 5% and 95%)
    flaky = []
    for test, results in test_results.items():
        total = results["pass"] + results["fail"]
        if total >= 5:  # Enough data
            pass_rate = results["pass"] / total
            if 0.05 < pass_rate < 0.95:
                flaky.append({
                    "test": test,
                    "pass_rate": pass_rate,
                    "total_runs": total
                })

    return sorted(flaky, key=lambda x: x["pass_rate"])

2. Code Pattern Analysis

Scan test code for flaky patterns:

FLAKY_PATTERNS = [
    # Timing issues
    (r'setTimeout|sleep|delay', "timing", "Uses explicit delays"),
    (r'Date\.now\(\)|new Date\(\)', "timing", "Uses current time"),

    # Async issues
    (r'\.then\([^)]*\)(?!.*await)', "async", "Promise without await"),
    (r'async.*(?!await)', "async", "Async without await"),

    # Order dependencies
    (r'Math\.random\(\)', "random", "Uses random values"),
    (r'uuid|nanoid', "random", "Uses generated IDs"),

    # Environment
    (r'process\.env', "environment", "Environment-dependent"),
    (r'fs\.(read|write)', "environment", "File system access"),
    (r'fetch\(|axios\.|http\.', "network", "Network calls"),
]

def scan_for_flaky_patterns(test_file):
    """Scan test file for flaky patterns"""
    content = read_file(test_file)
    matches = []

    for pattern, category, description in FLAKY_PATTERNS:
        if re.search(pattern, content):
            matches.append({
                "category": category,
                "description": description,
                "pattern": pattern
            })

    return matches

3. Re-run Analysis

Run tests multiple times to detect flakiness:

# Run tests 10 times, track results
for i in {1..10}; do
  npm test -- --reporter=json >> test-results.jsonl
done

# Analyze for inconsistency
python analyze_reruns.py test-results.jsonl

Output Format

## Flaky Test Report

**Analysis Period**: Last 30 days
**Total Tests**: 450
**Flaky Tests Found**: 12 (2.7%)

### Critical Flaky Tests (< 50% pass rate)

#### 1. `test/api/login.test.ts:45`
**Pass Rate**: 42% (21/50 runs)
**Category**: Timing
**Pattern**: Uses `Date.now()` for token expiry

```typescript
// Flaky code
it('should expire token after 1 hour', () => {
  const token = createToken();
  const expiry = Date.now() + 3600000;  // Flaky!
  expect(token.expiresAt).toBe(expiry);
});

Root Cause: Test creates token and checks expiry in same millisecond sometimes, different millisecond other times.

Recommended Fix: Use mocked time

it('should expire token after 1 hour', () => {
  vi.setSystemTime(new Date('2024-01-01T00:00:00Z'));
  const token = createToken();
  expect(token.expiresAt).toBe(new Date('2024-01-01T01:00:00Z').getTime());
  vi.useRealTimers();
});

High Flaky Tests (50-80% pass rate)

2. test/db/connection.test.ts:23

Pass Rate: 68% (34/50 runs) Category: Resource Pattern: Connection pool exhaustion

[... more tests ...]

Summary by Category

Recommendations

1. Quick Win: Fix 5 timing tests with vi.setSystemTime() (+0.5% stability)
2. Medium Effort: Add proper async handling (+0.3% stability)
3. Infrastructure: Add test isolation for DB tests (+0.2% stability)

Quarantine Candidates

These tests should be skipped in CI until fixed:

// vitest.config.ts
export default {
  test: {
    exclude: [
      'test/api/login.test.ts',       // Timing flaky
      'test/db/connection.test.ts',   // Resource flaky
    ]
  }
}

Note: Track quarantined tests in .aiwg/testing/flaky-quarantine.md

## Quarantine Process

### 1. Identify

```bash
# Run flaky detection
python scripts/flaky_detect.py --ci-history 30 --threshold 95

2. Quarantine

// Mark test as flaky
describe.skip('flaky: login expiry', () => {
  // FLAKY: https://github.com/org/repo/issues/123
  // Root cause: timing-dependent
  // Fix in progress: PR #456
});

3. Track

Create tracking issue:

## Flaky Test: test/api/login.test.ts:45

- **Pass Rate**: 42%
- **Category**: Timing
- **Root Cause**: Uses real system time
- **Quarantined**: 2024-12-12
- **Fix PR**: #456
- **Target Unquarantine**: 2024-12-15

4. Fix and Unquarantine

After fix:

# Verify fix with multiple runs
for i in {1..20}; do npm test -- test/api/login.test.ts; done

# Remove from quarantine if all pass

Integration Points

• Works with flaky-fix skill for automated repairs
• Reports to CI dashboard
• Feeds into flow-gate-check for release decisions
• Tracks in .aiwg/testing/flaky-registry.md

Script Reference

flaky_detect.py

Analyze CI history for flaky tests:

python scripts/flaky_detect.py --repo owner/repo --days 30

flaky_scanner.py

Scan code for flaky patterns:

python scripts/flaky_scanner.py --target test/

References

• @$AIWG_ROOT/${CLAUDE_PLUGIN_ROOT}/README.md — Testing quality addon overview
• @$AIWG_ROOT/agentic/code/addons/aiwg-utils/rules/research-before-decision.md — Research-first approach for root cause analysis
• @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/README.md — SDLC framework context for quality gates
• @$AIWG_ROOT/docs/cli-reference.md — CLI reference