regression-testing（回归测试）

Regression Testing

<default_to_action> When verifying changes don't break existing functionality:

1. ANALYZE what changed (git diff, impact analysis)
2. SELECT tests based on change + risk (not everything)
3. RUN in priority order (smoke → selective → full)
4. OPTIMIZE execution (parallel, sharding)
5. MONITOR suite health (flakiness, execution time)

Quick Regression Strategy:

• Per-commit: Smoke + changed code tests (5-10 min)
• Nightly: Extended regression (30-60 min)
• Pre-release: Full regression (2-4 hours)

Critical Success Factors:

• Smart selection catches 90% of regressions in 10% of time
• Flaky tests waste more time than they save
• Every production bug becomes a regression test </default_to_action>

Quick Reference Card

When to Use

• After any code change
• Before release
• After dependency updates
• After environment changes

Test Selection Strategies

Change-Based Test Selection

// Analyze changed files and select impacted tests
function selectTests(changedFiles: string[]): string[] {
  const testsToRun = new Set<string>();

  for (const file of changedFiles) {
    // Direct tests
    testsToRun.add(`${file.replace('.ts', '.test.ts')}`);

    // Dependent tests (via coverage mapping)
    const dependentTests = testCoverage[file] || [];
    dependentTests.forEach(t => testsToRun.add(t));
  }

  return Array.from(testsToRun);
}

// Example: payment.ts changed
// Runs: payment.test.ts, checkout.integration.test.ts, e2e/purchase.test.ts

CI/CD Integration

# .github/workflows/regression.yml
jobs:
  quick-regression:
    runs-on: ubuntu-latest
    timeout-minutes: 15
    steps:
      - name: Analyze changes
        id: changes
        uses: dorny/paths-filter@v2
        with:
          filters: |
            payment:
              - 'src/payment/**'
            auth:
              - 'src/auth/**'

      - name: Run affected tests
        run: npm run test:affected

      - name: Smoke tests (always)
        run: npm run test:smoke

  nightly-regression:
    if: github.event_name == 'schedule'
    timeout-minutes: 120
    steps:
      - run: npm test -- --coverage

Agent-Driven Regression

// Smart test selection
await Task("Regression Analysis", {
  pr: 1234,
  strategy: 'change-based-with-risk',
  timeBudget: '15min'
}, "qe-regression-risk-analyzer");

// Returns:
// {
//   mustRun: ['payment.test.ts', 'checkout.integration.test.ts'],
//   shouldRun: ['order.test.ts'],
//   canSkip: ['profile.test.ts', 'search.test.ts'],
//   estimatedTime: '12 min',
//   riskCoverage: 0.94
// }

// Generate regression test from production bug
await Task("Bug Regression Test", {
  bug: { id: 'BUG-567', description: 'Checkout fails > 100 items' },
  preventRecurrence: true
}, "qe-test-generator");

Agent Coordination Hints

Memory Namespace

aqe/regression-testing/
├── test-selection/*     - Impact analysis results
├── suite-health/*       - Flakiness, timing trends
├── coverage-maps/*      - Test-to-code mapping
└── bug-regressions/*    - Tests from production bugs

Fleet Coordination

const regressionFleet = await FleetManager.coordinate({
  strategy: 'comprehensive-regression',
  agents: [
    'qe-regression-risk-analyzer',  // Analyze changes, select tests
    'qe-test-executor',             // Execute selected tests
    'qe-coverage-analyzer',         // Analyze coverage gaps
    'qe-quality-gate'               // Go/no-go decision
  ],
  topology: 'sequential'
});

Related Skills

• risk-based-testing - Risk-based prioritization
• test-automation-strategy - Automation pyramid
• continuous-testing-shift-left - CI/CD integration

Remember

With Agents: qe-regression-risk-analyzer provides intelligent test selection achieving 90% defect detection in 10% of execution time. Agents generate regression tests from production bugs automatically.

Skill Composition

• Test failing? → Use /test-failure-investigator to diagnose root cause
• File a bug → Use /bug-reporting-excellence for proper bug reporting
• Test selection → Use /risk-based-testing for risk-based prioritization

Gotchas

• Agent defaults to "run everything" despite being told to select — explicitly constrain with --affected or file list
• Change-based selection misses transitive dependencies — a model change can break a controller test 3 hops away
• Flaky tests in regression suites erode trust faster than missing tests — quarantine immediately, don't skip
• Agent may report "0 regressions" when tests simply weren't run — verify test count in output, not just pass/fail
• Running full regression in containers often OOMs — use --workers=2 and --shard for CI environments