v3/plugins/agentic-qe/README.md
AI-powered quality engineering that writes tests, finds bugs, and breaks things (safely) so your users don't have to.
This plugin adds 58 AI agents to Claude Flow that handle all aspects of software quality:
Think of it as having a team of QA engineers who never sleep, never miss edge cases, and learn from every bug they find.
Via Claude Flow CLI (recommended):
npx claude-flow plugins install --name @claude-flow/plugin-agentic-qe
Via npm:
npm install @claude-flow/plugin-agentic-qe
Verify installation:
npx claude-flow plugins list
The simplest use case - point it at a file and get tests:
npx claude-flow@v3alpha mcp call aqe/generate-tests \
--targetPath ./src/utils/calculator.ts \
--testType unit \
--framework vitest
What you get:
// Generated: calculator.test.ts
describe('Calculator', () => {
it('should add two numbers', () => {
expect(add(2, 3)).toBe(5);
});
it('should handle negative numbers', () => {
expect(add(-1, 5)).toBe(4);
});
it('should handle decimal precision', () => {
expect(add(0.1, 0.2)).toBeCloseTo(0.3);
});
});
Give it a requirement, and it runs the full red-green-refactor cycle:
npx claude-flow@v3alpha mcp call aqe/tdd-cycle \
--requirement "Users can reset their password via email" \
--targetPath ./src/auth \
--style london
What happens:
Scan your code for vulnerabilities:
npx claude-flow@v3alpha mcp call aqe/security-scan \
--targetPath ./src \
--scanType sast \
--compliance owasp-top-10
Output:
{
"vulnerabilities": [
{
"severity": "high",
"type": "SQL Injection",
"file": "src/db/queries.ts",
"line": 42,
"fix": "Use parameterized queries instead of string concatenation"
}
],
"compliance": {
"owasp-top-10": { "passed": 8, "failed": 2 }
}
}
Block releases that don't meet quality standards:
const evaluation = await mcp.call('aqe/evaluate-quality-gate', {
gates: [
{ metric: 'line_coverage', operator: '>=', threshold: 80 },
{ metric: 'test_pass_rate', operator: '==', threshold: 100 },
{ metric: 'security_vulnerabilities', operator: '==', threshold: 0 },
{ metric: 'accessibility_violations', operator: '<=', threshold: 5 }
]
});
if (!evaluation.passed) {
console.log('Release blocked:', evaluation.failedCriteria);
process.exit(1);
}
Use ML to find likely defects:
npx claude-flow@v3alpha mcp call aqe/predict-defects \
--targetPath ./src/checkout \
--includeRootCause true
Output:
{
"predictions": [
{
"file": "src/checkout/payment.ts",
"probability": 0.78,
"reason": "High cyclomatic complexity + recent churn + no error handling for network failures",
"suggestedTests": ["network timeout", "partial payment failure", "currency conversion edge cases"]
}
]
}
Test how your system handles failures. Always use dryRun first!
# Step 1: Preview what would happen (safe)
npx claude-flow@v3alpha mcp call aqe/chaos-inject \
--target payment-service \
--failureType network-latency \
--duration 30 \
--intensity 0.5 \
--dryRun true
# Step 2: Run the actual experiment
npx claude-flow@v3alpha mcp call aqe/chaos-inject \
--target payment-service \
--failureType network-latency \
--duration 30 \
--intensity 0.5 \
--dryRun false
Failure types available:
network-latency - Add delays to network callsnetwork-partition - Isolate services from each othercpu-stress - Simulate high CPU loadmemory-pressure - Simulate memory exhaustiondisk-failure - Simulate storage issuesprocess-kill - Randomly kill processesdns-failure - Break DNS resolutionCatch UI changes automatically:
// Compare against baseline
const result = await mcp.call('aqe/visual-regression', {
targetUrl: 'http://localhost:3000',
viewports: [
{ width: 1920, height: 1080 }, // Desktop
{ width: 768, height: 1024 }, // Tablet
{ width: 375, height: 812 } // Mobile
],
threshold: 0.1 // 10% difference allowed
});
if (result.hasRegressions) {
console.log('Visual changes detected:', result.diffs);
}
Combine everything for comprehensive quality assurance:
// 1. Generate tests for uncovered code
const tests = await mcp.call('aqe/generate-tests', {
targetPath: './src',
coverage: { target: 90, focusGaps: true }
});
// 2. Run security scan
const security = await mcp.call('aqe/security-scan', {
targetPath: './src',
scanType: 'sast',
compliance: ['owasp-top-10', 'sans-25']
});
// 3. Check accessibility
const a11y = await mcp.call('aqe/check-accessibility', {
targetUrl: 'http://localhost:3000',
standard: 'WCAG21-AA'
});
// 4. Predict defects
const defects = await mcp.call('aqe/predict-defects', {
targetPath: './src'
});
// 5. Assess release readiness
const readiness = await mcp.call('aqe/assess-readiness', {
criteria: [
{ name: 'coverage', required: true },
{ name: 'security', required: true },
{ name: 'accessibility', required: false }
]
});
console.log('Ready to ship:', readiness.approved);
The plugin learns from your codebase and improves over time:
// The plugin stores patterns in memory
// After running on your codebase, it learns:
// - Your testing style and conventions
// - Common edge cases in your domain
// - Patterns that historically caused bugs
// Query learned patterns
const patterns = await mcp.call('aqe/suggest-tests', {
targetPath: './src/new-feature.ts',
useLearned: true // Use patterns learned from your codebase
});
// Patterns are stored in:
// - aqe/v3/test-patterns (test generation)
// - aqe/v3/defect-patterns (bug prediction)
// - aqe/v3/learning-trajectories (improvement over time)
| Category | Tools | What They Do |
|---|---|---|
| Test Generation | generate-tests, tdd-cycle, suggest-tests | Write tests automatically |
| Coverage | analyze-coverage, prioritize-gaps, track-trends | Find untested code |
| Quality | evaluate-quality-gate, assess-readiness, calculate-risk | Release decisions |
| Defects | predict-defects, analyze-root-cause, find-similar-defects | Bug prediction |
| Security | security-scan, audit-compliance, detect-secrets | Vulnerability scanning |
| Contracts | validate-contract, compare-contracts | API validation |
| Visual | visual-regression, check-accessibility | UI testing |
| Chaos | chaos-inject, assess-resilience, load-test | Resilience testing |
# claude-flow.config.yaml
plugins:
agentic-qe:
enabled: true
config:
defaultFramework: vitest
coverageTarget: 80
tddStyle: london
complianceStandards:
- owasp-top-10
- sans-25
MIT