agentic-qe 3.8.1 → 3.8.3

package/.claude/skills/qcsd-production-swarm/steps/07-learning-persistence.md

@@ -131,12 +131,12 @@ Save a JSON record to `${OUTPUT_FOLDER}/09-learning-persistence.json`:
 If MCP memory_store tool is unavailable, use CLI instead (STILL MANDATORY):
 
 ```bash
-npx @claude-flow/cli@latest memory store \
+npx --no-install ruflo memory store \
   --key "qcsd-production-${RELEASE_ID}-$(date +%s)" \
   --value '{"recommendation":"[VALUE]","doraScore":[N],"slaCompliance":[N],"incidentSeverity":"[P_]","defectTrend":"[VALUE]"}' \
   --namespace qcsd-production
 
-npx @claude-flow/cli@latest hooks post-task \
+npx --no-install ruflo hooks post-task \
   --task-id "qcsd-production-${RELEASE_ID}" \
   --success true
 ```

package/.claude/skills/qcsd-refinement-swarm/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: qcsd-refinement-swarm
-description: "QCSD Refinement phase swarm for Sprint Refinement sessions using SFDIPOT product factors, BDD scenario generation, and requirements validation."
+description: "Use when running Sprint Refinement sessions with SFDIPOT product factors, generating BDD scenarios, or validating requirements in the QCSD Refinement phase."
 category: qcsd-phases
 priority: critical
 version: 1.0.0
@@ -42,7 +42,6 @@ validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qcsd-refinement-swarm.yaml
-
 ---
 
 # QCSD Refinement Swarm v1.0

package/.claude/skills/qcsd-refinement-swarm/steps/01-flag-detection.md

@@ -26,7 +26,7 @@ mcp__agentic-qe__memory_query({
 **CLI Fallback:**
 
 ```bash
-npx @claude-flow/cli@latest memory search --query "qcsd-ideation" --namespace qcsd-ideation --limit 1
+npx --no-install ruflo memory search --query "qcsd-ideation" --namespace qcsd-ideation --limit 1
 ```
 
 ### Step 1: Scan Story Content and Detect Flags

package/.claude/skills/qe-chaos-resilience/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Chaos Resilience"
-description: "Chaos engineering and resilience testing including fault injection, load testing, and system recovery validation."
+description: "Use when injecting faults into distributed systems, validating system recovery, running load/stress tests, or building confidence in resilience through chaos engineering."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-chaos-resilience.yaml
-
 ---
 
 # QE Chaos Resilience
@@ -241,4 +240,4 @@ await resilienceTester.validateSLA({
 
 **Primary Agents**: qe-chaos-engineer, qe-load-tester, qe-resilience-tester
 **Coordinator**: qe-chaos-coordinator
-**Related Skills**: qe-performance, qe-security-compliance
+**Related Skills**: qe-performance, security-testing

package/.claude/skills/qe-code-intelligence/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Code Intelligence"
-description: "Knowledge graph-based code understanding with semantic search and 80% token reduction through intelligent context retrieval."
+description: "Use when understanding unfamiliar code, searching semantically across a codebase, mapping dependencies, or reducing context window usage through intelligent retrieval."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-code-intelligence.yaml
-
 ---
 
 # QE Code Intelligence
@@ -208,6 +207,14 @@ aqe kg export --format dot --output codebase.dot
 aqe kg stats
 ```
 
+## Gotchas
+
+- WARNING: code-intelligence domain has 18% success rate — prefer direct grep/glob over agent-based code search for simple queries
+- Knowledge graph construction fails on repos >50K LOC — scope to specific modules
+- Semantic search returns irrelevant results without domain-specific embeddings — always verify search results manually
+- Agent claims "80% token reduction" but may skip critical context — verify key files are included in results
+- Fleet must be initialized before using: run `npx ruflo doctor --fix` if you get initialization errors
+
 ## Coordination
 
 **Primary Agents**: qe-knowledge-graph, qe-semantic-searcher, qe-dependency-mapper

package/.claude/skills/qe-coverage-analysis/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Coverage Analysis"
-description: "O(log n) sublinear coverage gap detection with risk-weighted analysis and intelligent test prioritization."
+description: "Use when analyzing test coverage, identifying coverage gaps, prioritizing testing effort, or comparing coverage between branches."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-coverage-analysis.yaml
-
 ---
 
 # QE Coverage Analysis
@@ -186,6 +185,33 @@ quality_gates:
       - alert_on_decline: 3  # consecutive PRs
 ```
 
+## Run History
+
+After each coverage analysis, append results to `run-history.json` in this skill directory:
+```bash
+# Read current history, append new entry, write back
+node -e "
+const fs = require('fs');
+const h = JSON.parse(fs.readFileSync('.claude/skills/qe-coverage-analysis/run-history.json'));
+h.runs.push({date: new Date().toISOString().split('T')[0], statements_pct: STATEMENTS, branches_pct: BRANCHES, gaps_found: GAPS});
+fs.writeFileSync('.claude/skills/qe-coverage-analysis/run-history.json', JSON.stringify(h, null, 2));
+"
+```
+Read `run-history.json` before each run to detect trends (e.g., "coverage dropped 3 consecutive times").
+
+## Skill Composition
+
+- **Coverage dropped?** → Use `/coverage-drop-investigator` to trace the cause
+- **Need more tests** → Use `/qe-test-generation` to fill gaps
+- **Validate quality** → Use `/mutation-testing` to ensure coverage means quality
+- **Ship decision** → Feed into `/qe-quality-assessment` for deployment readiness
+
+## Gotchas
+
+- High line coverage does NOT mean good tests — 100% coverage with 0% assertions is common agent output. Use mutation testing to verify
+- coverage-analysis domain has 86% success rate — 14% of runs fail on initialization. Always verify results and have fallback plan (e.g. manual coverage tools)
+- Self-learning pipeline may silently stop learning (statusline frozen for days) — only human inspection catches this
+
 ## Coordination
 
 **Primary Agents**: qe-coverage-specialist, qe-coverage-analyzer, qe-gap-detector

package/.claude/skills/qe-coverage-analysis/run-history.json

@@ -0,0 +1,6 @@
+{
+  "_description": "Coverage analysis run history. Append after each run. Claude reads this to detect trends.",
+  "_format": "Each entry: {date, scope, statements_pct, branches_pct, functions_pct, gaps_found, recommendation}",
+  "_instructions": "After running coverage analysis, append results here. Compare with previous entries to detect trends. Alert if coverage declines 3 consecutive times.",
+  "runs": []
+}

package/.claude/skills/qe-defect-intelligence/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Defect Intelligence"
-description: "AI-powered defect prediction, pattern learning, and root cause analysis for proactive quality management."
+description: "Use when predicting defects before they escape, analyzing root causes of failures, learning defect patterns, or implementing proactive quality management with AI."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-defect-intelligence.yaml
-
 ---
 
 # QE Defect Intelligence

package/.claude/skills/qe-iterative-loop/SKILL.md

@@ -5,7 +5,6 @@ trust_tier: 2
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
-
 ---
 
 # QE Iterative Loop

package/.claude/skills/qe-learning-optimization/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Learning Optimization"
-description: "Transfer learning, metrics optimization, and continuous improvement for AI-powered QE agents."
+description: "Use when optimizing QE agent performance, applying transfer learning across test domains, tuning quality metrics, or implementing continuous improvement loops for AI-powered testing."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-learning-optimization.yaml
-
 ---
 
 # QE Learning Optimization

package/.claude/skills/qe-quality-assessment/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Quality Assessment"
-description: "Comprehensive quality gates, metrics analysis, and deployment readiness assessment for continuous quality assurance."
+description: "Use when evaluating code quality, setting up quality gates, assessing deployment readiness, or generating quality reports."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-quality-assessment.yaml
-
 ---
 
 # QE Quality Assessment
@@ -212,8 +211,35 @@ quality_check:
       - 1  # Warnings only
 ```
 
+## Run History
+
+After each quality assessment, append results to `run-history.json` in this skill directory:
+```bash
+node -e "
+const fs = require('fs');
+const h = JSON.parse(fs.readFileSync('.claude/skills/qe-quality-assessment/run-history.json'));
+h.runs.push({date: new Date().toISOString().split('T')[0], gate_result: 'PASS_OR_FAIL', failed_checks: []});
+fs.writeFileSync('.claude/skills/qe-quality-assessment/run-history.json', JSON.stringify(h, null, 2));
+"
+```
+Read `run-history.json` before each run — alert if quality gate failed 3 of last 5 runs.
+
+## Skill Composition
+
+- **Before assessment** → Run `/qe-coverage-analysis` and `/mutation-testing` first
+- **If issues found** → Use `/test-failure-investigator` to diagnose failures
+- **For PR review** → Combine with `/code-review-quality` for comprehensive review
+
+## Gotchas
+
+- NEVER trust agent-reported pass/fail status — 12 test failures were caught that agents claimed were passing (Nagual pattern, reward 0.92)
+- Completion theater: agent hardcoded version '3.0.0' instead of reading from package.json — verify actual values in output
+- Fix issues in priority waves (P0 → P1 → P2) with verification between each wave — don't fix everything in parallel
+- quality-assessment domain has 53.7% success rate — expect failures and have fallback
+- If HybridMemoryBackend initialization fails, run `npx ruflo doctor --fix` first
+
 ## Coordination
 
 **Primary Agents**: qe-quality-analyzer, qe-deployment-advisor, qe-metrics-collector
 **Coordinator**: qe-quality-coordinator
-**Related Skills**: qe-coverage-analysis, qe-security-compliance
+**Related Skills**: qe-coverage-analysis, security-testing

package/.claude/skills/qe-quality-assessment/run-history.json

@@ -0,0 +1,6 @@
+{
+  "_description": "Quality assessment run history. Append after each run. Claude reads this to track quality gate pass/fail trends.",
+  "_format": "Each entry: {date, scope, gate_result, scores, failed_checks, recommendation}",
+  "_instructions": "After running quality assessment, append results here. Alert if quality gate fails 3 of last 5 runs. Track which checks fail most often.",
+  "runs": []
+}

package/.claude/skills/qe-requirements-validation/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Requirements Validation"
-description: "Requirements traceability, acceptance criteria validation, and BDD scenario management for complete requirements coverage."
+description: "Use when validating acceptance criteria, tracing requirements to tests, managing BDD scenarios, or ensuring complete requirements coverage before development."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-requirements-validation.yaml
-
 ---
 
 # QE Requirements Validation

package/.claude/skills/qe-test-execution/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Test Execution"
-description: "Parallel test execution orchestration with intelligent scheduling, retry logic, and comprehensive result aggregation."
+description: "Use when running test suites, optimizing test execution time, handling flaky tests, or setting up CI/CD test pipelines."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-test-execution.yaml
-
 ---
 
 # QE Test Execution
@@ -181,6 +180,14 @@ interface ExecutionResults {
 }
 ```
 
+## Gotchas
+
+- Full test suites may OOM in containers — the rule "don't run full suite" was violated 20x despite being in CLAUDE.md. Fix: make suite lightweight, don't just add more rules
+- Fewer focused agents (3-4) outperform many vague ones (6-8) — always include verification command in each agent prompt
+- New model releases can shift agent behavior mid-sprint — rules followed yesterday may be ignored today after model update
+- Running all tests in parallel can mask flaky tests — use `--workers=1` for initial diagnosis
+- Session crashes lose all context — save intermediate results to disk, not just memory
+
 ## Coordination
 
 **Primary Agents**: qe-test-executor, qe-test-selector, qe-flaky-detector

package/.claude/skills/qe-test-generation/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Test Generation"
-description: "AI-powered test generation using pattern recognition, code analysis, and intelligent test synthesis for comprehensive test coverage."
+description: "Use when generating tests for new/changed code, improving test coverage, or migrating between Jest/Vitest/Playwright frameworks."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-test-generation.yaml
-
 ---
 
 # QE Test Generation
@@ -140,6 +139,20 @@ quality_checks:
     statements: 85
 ```
 
+## Skill Composition
+
+- **After generating tests** → Run `/mutation-testing` to verify test quality
+- **Before generating** → Use `/test-automation-strategy` to choose framework and patterns
+- **Related** → `/qe-coverage-analysis` to find where tests are needed most
+
+## Gotchas
+
+- Agent truncates output on files >3000 lines — scope generation to individual modules, not entire directories
+- Components that pass unit tests individually may have zero integration wiring — always generate at least one integration test per module boundary
+- When generating tests for a new codebase, check which framework is installed (jest vs vitest vs mocha) — they have different mock APIs and Claude will use the wrong one
+- Completion theater: agent may claim "comprehensive tests generated" but leave stubs or hardcoded values — always run the generated tests before accepting
+- Fleet must be initialized before using QE agents: run `npx ruflo doctor --fix` if you get "Fleet not initialized"
+
 ## Coordination
 
 **Primary Agents**: qe-test-generator, qe-pattern-matcher, qe-test-architect

package/.claude/skills/qe-test-generation/config.json

@@ -0,0 +1,14 @@
+{
+  "$schema": "./config-schema.json",
+  "_description": "QE Test Generation configuration. Auto-created on first run. Edit to customize.",
+  "framework": null,
+  "style": null,
+  "coverageTarget": 80,
+  "outputDir": "tests/",
+  "options": {
+    "generateIntegrationTests": true,
+    "generateEdgeCases": true,
+    "maxTestsPerFile": 20
+  },
+  "_setupPrompt": "If framework is null, ask the user: 'Which test framework does this project use? (jest/vitest/mocha/playwright)'. If style is null, ask: 'Do you prefer London (mock-based) or Chicago (state-based) TDD style?'"
+}

package/.claude/skills/qe-test-generation/templates/test-scaffold-jest.md

@@ -0,0 +1,72 @@
+# Jest Unit Test Scaffold
+
+## Class Under Test Pattern
+```typescript
+import { {{ClassName}} } from '../src/{{path}}';
+
+describe('{{ClassName}}', () => {
+  let sut: {{ClassName}};
+
+  beforeEach(() => {
+    sut = new {{ClassName}}(/* dependencies */);
+  });
+
+  describe('{{methodName}}', () => {
+    it('should return expected result for valid input', () => {
+      // Arrange
+      const input = /* valid input */;
+
+      // Act
+      const result = sut.{{methodName}}(input);
+
+      // Assert
+      expect(result).toBe(/* expected */);
+    });
+
+    it('should throw for invalid input', () => {
+      expect(() => sut.{{methodName}}(null)).toThrow();
+    });
+
+    it('should handle edge case: empty input', () => {
+      const result = sut.{{methodName}}(/* empty */);
+      expect(result).toBe(/* expected for empty */);
+    });
+
+    it('should handle edge case: boundary value', () => {
+      const result = sut.{{methodName}}(/* boundary */);
+      expect(result).toBe(/* expected at boundary */);
+    });
+  });
+});
+```
+
+## Integration Test Pattern
+```typescript
+describe('{{Feature}} Integration', () => {
+  beforeAll(async () => {
+    await setupTestDatabase();
+  });
+
+  afterAll(async () => {
+    await teardownTestDatabase();
+  });
+
+  it('should complete the full workflow', async () => {
+    // Step 1: Create
+    const created = await service.create(validPayload);
+    expect(created.id).toBeDefined();
+
+    // Step 2: Read
+    const fetched = await service.getById(created.id);
+    expect(fetched).toMatchObject(validPayload);
+
+    // Step 3: Update
+    const updated = await service.update(created.id, changes);
+    expect(updated.field).toBe(changes.field);
+
+    // Step 4: Delete
+    await service.delete(created.id);
+    await expect(service.getById(created.id)).rejects.toThrow('Not found');
+  });
+});
+```

package/.claude/skills/qe-visual-accessibility/SKILL.md

@@ -1,12 +1,11 @@
 ---
 name: "QE Visual Accessibility"
-description: "Visual regression testing, responsive design validation, and WCAG accessibility compliance testing."
+description: "Use when detecting visual regressions, validating responsive design across viewports, testing WCAG accessibility compliance, or ensuring UI consistency after changes."
 trust_tier: 3
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/qe-visual-accessibility.yaml
-
 ---
 
 # QE Visual Accessibility

package/.claude/skills/quality-metrics/SKILL.md

@@ -16,7 +16,6 @@ validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/quality-metrics.yaml
-
 ---
 
 # Quality Metrics
@@ -24,21 +23,9 @@ validation:
 <default_to_action>
 When measuring quality or building dashboards:
 1. MEASURE outcomes (bug escape rate, MTTD) not activities (test count)
-2. FOCUS on DORA metrics: Deployment frequency, Lead time, MTTD, MTTR, Change failure rate
-3. AVOID vanity metrics: 100% coverage means nothing if tests don't catch bugs
-4. SET thresholds that drive behavior (quality gates block bad code)
-5. TREND over time: Direction matters more than absolute numbers
-
-**Quick Metric Selection:**
-- Speed: Deployment frequency, lead time for changes
-- Stability: Change failure rate, MTTR
-- Quality: Bug escape rate, defect density, test effectiveness
-- Process: Code review time, flaky test rate
-
-**Critical Success Factors:**
-- Metrics without action are theater
-- What you measure is what you optimize
-- Trends matter more than snapshots
+2. AVOID vanity metrics: 100% coverage means nothing if tests don't catch bugs
+3. SET thresholds that drive behavior (quality gates block bad code)
+4. TREND over time: Direction matters more than absolute numbers
 </default_to_action>
 
 ## Quick Reference Card
@@ -49,23 +36,6 @@ When measuring quality or building dashboards:
 - Evaluating testing effectiveness
 - Justifying quality investments
 
-### Meaningful vs Vanity Metrics
-| ✅ Meaningful | ❌ Vanity |
-|--------------|-----------|
-| Bug escape rate | Test case count |
-| MTTD (detection) | Lines of test code |
-| MTTR (recovery) | Test executions |
-| Change failure rate | Coverage % (alone) |
-| Lead time for changes | Requirements traced |
-
-### DORA Metrics
-| Metric | Elite | High | Medium | Low |
-|--------|-------|------|--------|-----|
-| Deploy Frequency | On-demand | Weekly | Monthly | Yearly |
-| Lead Time | < 1 hour | < 1 week | < 1 month | > 6 months |
-| Change Failure Rate | < 5% | < 15% | < 30% | > 45% |
-| MTTR | < 1 hour | < 1 day | < 1 week | > 1 month |
-
 ### Quality Gate Thresholds
 | Metric | Blocking Threshold | Warning |
 |--------|-------------------|---------|
@@ -77,38 +47,6 @@ When measuring quality or building dashboards:
 
 ---
 
-## Core Metrics
-
-### Bug Escape Rate
-```
-Bug Escape Rate = (Production Bugs / Total Bugs Found) × 100
-
-Target: < 10% (90% caught before production)
-```
-
-### Test Effectiveness
-```
-Test Effectiveness = (Bugs Found by Tests / Total Bugs) × 100
-
-Target: > 70%
-```
-
-### Defect Density
-```
-Defect Density = Defects / KLOC
-
-Good: < 1 defect per KLOC
-```
-
-### Mean Time to Detect (MTTD)
-```
-MTTD = Time(Bug Reported) - Time(Bug Introduced)
-
-Target: < 1 day for critical, < 1 week for others
-```
-
----
-
 ## Dashboard Design
 
 ```typescript
@@ -205,17 +143,6 @@ const metricsFleet = await FleetManager.coordinate({
 
 ---
 
-## Common Traps
-
-| Trap | Problem | Solution |
-|------|---------|----------|
-| Coverage worship | 100% coverage, bugs still escape | Measure bug escape rate instead |
-| Test count focus | Many tests, slow feedback | Measure execution time |
-| Activity metrics | Busy work, no outcomes | Measure outcomes (MTTD, MTTR) |
-| Point-in-time | Snapshot without context | Track trends over time |
-
----
-
 ## Related Skills
 - [agentic-quality-engineering](../agentic-quality-engineering/) - Agent coordination
 - [cicd-pipeline-qe-orchestrator](../cicd-pipeline-qe-orchestrator/) - Quality gates
@@ -226,6 +153,4 @@ const metricsFleet = await FleetManager.coordinate({
 
 ## Remember
 
-**Measure outcomes, not activities.** Bug escape rate > test count. MTTD/MTTR > coverage %. Trends > snapshots. Set gates that block bad code. What you measure is what you optimize.
-
 **With Agents:** Agents track metrics automatically, analyze trends, trigger alerts, and make gate decisions. Use agents to maintain continuous quality visibility.

package/.claude/skills/refactoring-patterns/SKILL.md

@@ -15,7 +15,6 @@ trust_tier: 2
 validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
-
 ---
 
 # Refactoring Patterns

package/.claude/skills/regression-testing/SKILL.md

@@ -16,7 +16,6 @@ validation:
   schema_path: schemas/output.json
   validator_path: scripts/validate-config.json
   eval_path: evals/regression-testing.yaml
-
 ---
 
 # Regression Testing
@@ -48,14 +47,6 @@ When verifying changes don't break existing functionality:
 - After dependency updates
 - After environment changes
 
-### Regression Types
-| Type | When | Scope |
-|------|------|-------|
-| **Corrective** | No code change | Full suite |
-| **Progressive** | New features | Existing + new |
-| **Selective** | Specific changes | Changed + dependent |
-| **Complete** | Major refactor | Everything |
-
 ### Test Selection Strategies
 | Strategy | How | Reduction |
 |----------|-----|-----------|
@@ -91,22 +82,6 @@ function selectTests(changedFiles: string[]): string[] {
 
 ---
 
-## Regression Suite Pyramid
-
-```
-         /\
-        /  \  Full Regression (weekly)
-       /    \  - All tests (2-4 hours)
-      /------\
-     /        \  Extended Regression (nightly)
-    /          \  - Unit + integration + critical E2E (30-60 min)
-   /------------\
-  /              \  Quick Regression (per commit)
- /________________\  - Changed code + smoke tests (5-10 min)
-```
-
----
-
 ## CI/CD Integration
 
 ```yaml
@@ -141,27 +116,6 @@ jobs:
 
 ---
 
-## Optimization Techniques
-
-```javascript
-// 1. Parallel execution
-module.exports = {
-  maxWorkers: '50%', // Use half CPU cores
-  testTimeout: 30000
-};
-
-// 2. Sharding across CI workers
-// npm test -- --shard=1/4
-
-// 3. Incremental testing (only changed since last run)
-// Track last run state, skip passing unchanged tests
-
-// 4. Fast-fail on smoke
-// Run critical tests first, abort if they fail
-```
-
----
-
 ## Agent-Driven Regression
 
 ```typescript
@@ -226,8 +180,18 @@ const regressionFleet = await FleetManager.coordinate({
 
 ## Remember
 
-**Regression testing is insurance against change.** Every code change is a risk. Smart regression testing mitigates that risk by testing what matters based on what changed.
+**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.
 
-**Good regression testing is strategic, not exhaustive.** You cannot test everything, every time. Select based on changes, risk, and time budget.
+## Skill Composition
 
-**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.
+- **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