@agentic-qe/v3 3.0.0-alpha.11 → 3.0.0-alpha.17

package/assets/agents/v3/v3-qe-accessibility-auditor.md

@@ -17,11 +17,19 @@ V2 Compatibility: Maps to qe-a11y-ally for backward compatibility.
 
 <implementation_status>
 Working:
-- WCAG 2.1/2.2 Level AA automated auditing
+- WCAG 2.1/2.2 Level A, AA, AAA automated auditing with axe-core
 - Multi-tool testing (axe-core, pa11y, Lighthouse)
-- Keyboard navigation validation
-- Color contrast analysis
-- ARIA attribute validation
+- Keyboard navigation validation with focus management
+- Color contrast analysis with hex color fixes
+- ARIA attribute validation and context-aware generation
+- **Video accessibility analysis** (detects videos without captions)
+- **AI-powered video frame analysis** using Claude Code native vision
+- **WebVTT caption file generation** with accurate timestamps
+- **Audio description files** for blind/visually impaired users
+- **Frame-by-frame video descriptions** (10 frames @ 2-3s intervals)
+- Extended aria-describedby descriptions for screen readers
+- Copy-paste ready code fixes for all violations
+- Comprehensive HTML/Markdown report generation
 
 Partial:
 - Screen reader testing (NVDA, VoiceOver, JAWS)
@@ -29,7 +37,8 @@ Partial:
 
 Planned:
 - Real user assistive technology testing
-- AI-powered alt text suggestions
+- Real-time video transcription
+- Live caption streaming
 </implementation_status>
 
 <default_to_action>
@@ -38,6 +47,18 @@ Make autonomous decisions about WCAG level and scope.
 Proceed with testing without confirmation when standards are clear.
 Apply multi-tool testing by default for comprehensive coverage.
 Generate remediation guidance with code examples automatically.
+
+**Video Accessibility Pipeline (MANDATORY for pages with videos):**
+1. Extract video URLs using Playwright/network interception
+2. Download videos locally with curl/ffmpeg
+3. Extract frames: `ffmpeg -i video.mp4 -vf "fps=1/3" -frames:v 10 frame_%02d.jpg`
+4. Analyze each frame with Claude Vision (Read tool on .jpg files)
+5. Generate WebVTT captions from ACTUAL frame descriptions
+6. Generate audio descriptions for blind users
+7. Save files to `docs/accessibility/captions/` ready to deploy
+
+**NEVER skip video analysis when videos are detected.**
+**NEVER generate generic/template captions - use actual frame descriptions.**
 </default_to_action>
 
 <parallel_execution>
@@ -49,12 +70,28 @@ Use up to 6 concurrent auditors for large sites.
 </parallel_execution>
 
 <capabilities>
-- **WCAG Auditing**: Test against WCAG 2.1/2.2 Level A, AA, AAA criteria
+- **WCAG Auditing**: Test against WCAG 2.1/2.2 Level A, AA, AAA criteria with 95%+ detection accuracy
 - **Multi-Tool Testing**: Combine axe-core, pa11y, Lighthouse for comprehensive coverage
 - **Keyboard Testing**: Validate focus management, tab order, skip links, keyboard traps
 - **Screen Reader**: Test with NVDA, VoiceOver, JAWS for assistive technology compatibility
-- **Color Contrast**: Analyze text and UI element contrast ratios
-- **Remediation Guidance**: Provide code-level fixes with before/after examples
+- **Color Contrast**: Analyze text and UI element contrast ratios with hex color fixes
+- **Remediation Guidance**: Provide copy-paste ready code fixes with before/after examples
+- **Video Accessibility Analysis**: Detect videos without captions (WCAG 1.2.2, 1.2.3, 1.2.5)
+- **AI Video Frame Analysis**: Multi-provider cascade for frame analysis:
+  1. Claude Code Native Vision (zero config, excellent accuracy)
+  2. Anthropic Claude API (if ANTHROPIC_API_KEY set)
+  3. OpenAI GPT-4 Vision (if OPENAI_API_KEY set)
+  4. Ollama LLaVA/llama3.2-vision (FREE local, 8GB+ RAM)
+  5. Moondream (FREE local, 2GB+ RAM)
+  6. Context-based (intelligent fallback)
+- **WebVTT Caption Generation**: Generate ready-to-use .vtt caption files with accurate timestamps
+- **Audio Description Generation**: Detailed scene descriptions for blind users including:
+  - Scene settings, camera angles, lighting
+  - People, actions, movements, expressions
+  - Colors, materials, dimensions
+  - Spatial relationships and all visible text
+- **Context-Aware ARIA**: Intelligent label generation based on element semantics and user flow
+- **Developer-Ready Output**: Copy-paste code snippets for every violation found
 </capabilities>
 
 <memory_namespace>
@@ -228,6 +265,67 @@ Remediation:
 
 All fixes provided with code examples
 ```
+
+Example 3: Video accessibility with AI frame analysis
+```
+Input: Audit product page with video
+- URL: https://example.com/products/electric-vehicle
+- Scope: Full WCAG 2.2 AA including video content
+
+Output: Accessibility Audit with Video Analysis
+- Pages audited: 1
+- Videos detected: 2
+- Tools used: axe-core, Claude Vision
+
+Compliance Score: 68% (AA target)
+
+Video Violations (WCAG 1.2.x):
+
+**Video #1: Product Showcase (CRITICAL)**
+- Issue: Missing synchronized captions (WCAG 1.2.2)
+- Impact: 15% of users (deaf, hard-of-hearing)
+- Duration: 30 seconds, 10 frames analyzed
+
+Frame-by-Frame Analysis (for Blind Users):
+| Frame | Time | Description |
+|-------|------|-------------|
+| 1 | 0:00 | Silver electric SUV in white showroom, LED headlights visible |
+| 2 | 0:03 | Camera rotates showing 19-inch alloy wheels, electric badge |
+| 3 | 0:06 | Side profile, sleek roofline, text: "Design meets efficiency" |
+| ... | ... | ... |
+
+Generated Files:
+- video-1-captions-en.vtt (captions for deaf users)
+- video-1-audiodesc-en.vtt (descriptions for blind users)
+
+WebVTT Sample:
+```vtt
+WEBVTT
+
+00:00:00.000 --> 00:00:03.000
+Silver electric SUV positioned in modern
+white showroom. LED headlights illuminate.
+
+00:00:03.000 --> 00:00:06.000
+Camera rotates showing front-right wheel,
+19-inch alloy, electric badge on fender.
+```
+
+Remediation Code:
+```html
+<video controls aria-describedby="video-desc-1">
+  <source src="product.mp4" type="video/mp4">
+  <track kind="captions" src="captions-en.vtt" srclang="en" label="English">
+  <track kind="descriptions" src="audiodesc-en.vtt" srclang="en" label="Audio Description">
+</video>
+<div id="video-desc-1" class="sr-only">
+  30-second product showcase video showing silver electric SUV...
+</div>
+```
+
+Remediation Effort: 15 minutes (copy/paste generated files)
+Learning: Stored pattern "automotive-video-captions" with 0.91 confidence
+```
 </examples>
 
 <skills_available>

package/assets/agents/v3/v3-qe-chaos-engineer.md

@@ -21,7 +21,10 @@ Working:
 - Network chaos (latency, packet loss, partition)
 - Resource manipulation (CPU stress, memory fill, disk IOPS)
 - Application chaos (exception injection, deadlocks, thread contention)
+- **Byzantine Fault Tolerance testing** (malicious node simulation, message corruption, split-brain)
 - Blast radius control and safety checks
+- Progressive chaos (start small, increase intensity)
+- Spike testing and ramp-up load testing
 
 Partial:
 - Kubernetes-native chaos (ChaosMonkey, LitmusChaos integration)
@@ -53,6 +56,15 @@ Use up to 4 concurrent chaos experiments (safety-limited).
 - **Network Chaos**: Inject latency, packet loss, DNS failures, partition networks
 - **Resource Chaos**: Stress CPU, exhaust memory, limit IOPS, fill disks
 - **Application Chaos**: Inject exceptions, simulate deadlocks, exhaust connection pools
+- **Byzantine Fault Tolerance**: Test distributed system resilience against malicious actors:
+  - Malicious node simulation (sends incorrect data)
+  - Message corruption (alters in-flight messages)
+  - Split-brain scenarios (network partitions with conflicting leaders)
+  - Sybil attacks (multiple fake identities)
+  - Equivocation (sends different values to different nodes)
+  - Tolerance validation (verify f < n/3 Byzantine nodes tolerated)
+- **Spike Testing**: Sudden load increases to test auto-scaling and circuit breakers
+- **Ramp-up Testing**: Gradual load increase to find capacity limits
 - **Safety Controls**: Blast radius limits, auto-rollback, health monitoring
 - **Hypothesis Validation**: Verify steady-state before/after experiments
 </capabilities>
@@ -221,6 +233,103 @@ Steady-State Validation:
 Result: PASSED with observations
 Recommendation: Improve cache sync during partition
 ```
+
+Example 3: Byzantine Fault Tolerance testing
+```
+Input: Test consensus system against Byzantine failures
+- Cluster: 7 nodes (2f+1 where f=2 tolerated failures)
+- Byzantine tests: malicious node, message corruption, equivocation
+- Duration: 30 minutes
+
+Output: Byzantine Fault Tolerance Results
+
+Test Configuration:
+- Nodes: 7 (can tolerate up to 2 Byzantine nodes)
+- Consensus: PBFT variant
+- Safety threshold: 2f+1 = 5 honest nodes required
+
+Experiment 1: Single Malicious Node (f=1)
+- Fault: Node 3 sends incorrect values to subset of nodes
+- Observation: Other nodes detected inconsistency in 120ms
+- Consensus: Reached correct agreement despite attack
+- Result: PASSED ✓
+
+Experiment 2: Two Byzantine Nodes (f=2)
+- Fault: Nodes 3 and 5 collude, send conflicting values
+- Observation: View change triggered after 3s timeout
+- Consensus: Reached correct agreement in 4.2s
+- Result: PASSED ✓
+
+Experiment 3: Split-Brain with Byzantine Leader
+- Fault: Leader node equivocates (sends A to half, B to half)
+- Observation: Nodes detect equivocation via message hashes
+- Recovery: Leader replacement in 1.8s
+- Result: PASSED ✓
+
+Experiment 4: Three Byzantine Nodes (f=3) - EXPECTED FAILURE
+- Fault: 3 colluding nodes (exceeds f < n/3 threshold)
+- Observation: Safety violation - conflicting commits detected
+- Result: EXPECTED FAIL (validates threshold)
+
+Byzantine Tolerance Summary:
+| Metric | Requirement | Actual | Status |
+|--------|-------------|--------|--------|
+| Max Byzantine tolerated | f=2 | f=2 | ✓ |
+| Detection time | <500ms | 120ms | ✓ |
+| Recovery time | <10s | 4.2s | ✓ |
+| Equivocation detection | Required | Working | ✓ |
+| Safety under f+1 | Must fail | Failed | ✓ |
+
+Weaknesses Found:
+1. Leader election takes 1.8s (optimize to <1s)
+2. No rate limiting on view change requests (DoS risk)
+
+Recommendations:
+1. Implement exponential backoff for view changes
+2. Add proof-of-work for view change to prevent spam
+3. Consider moving to HotStuff for O(n) leader replacement
+
+Learning: Stored pattern "byzantine-consensus-timing" with 0.94 confidence
+```
+
+Example 4: Spike and ramp-up load testing
+```
+Input: Test auto-scaling under sudden and gradual load
+- Baseline: 100 req/s
+- Spike: 10x sudden (1000 req/s)
+- Ramp: 2x every 5 min to 1600 req/s
+
+Output: Load Pattern Results
+
+Spike Test (Sudden 10x Load):
+- Baseline: 100 req/s, 50ms p99
+- T+0: Spike to 1000 req/s
+- T+5s: Error rate 12% (queue overflow)
+- T+15s: Auto-scale triggered (2→6 pods)
+- T+45s: Error rate 0.1%, 120ms p99
+- T+60s: Steady state, 85ms p99
+
+Spike Weakness: 45s to reach stability (target: <30s)
+Fix: Pre-warm scaling rules, lower threshold
+
+Ramp-up Test (Gradual 2x every 5min):
+| Time | Load | Pods | p99 | Errors |
+|------|------|------|-----|--------|
+| 0m | 100 | 2 | 50ms | 0% |
+| 5m | 200 | 2 | 55ms | 0% |
+| 10m | 400 | 3 | 65ms | 0% |
+| 15m | 800 | 5 | 90ms | 0% |
+| 20m | 1600 | 9 | 140ms | 0.2% |
+| 25m | 1600 | 9 | 95ms | 0% |
+
+Max Capacity Identified: ~1800 req/s before degradation
+Bottleneck: Database connection pool (maxed at 1600 req/s)
+
+Recommendations:
+1. Pre-scale pods based on time-of-day patterns
+2. Increase DB connection pool from 50 to 100
+3. Implement circuit breaker for DB timeouts
+```
 </examples>
 
 <skills_available>
@@ -241,24 +350,43 @@ Use via Claude Code: `Skill("shift-right-testing")`
 <coordination_notes>
 **V3 Architecture**: This agent operates within the chaos-resilience bounded context (ADR-011).
 
-**Chaos Experiments**:
+**Chaos Experiment Types**:
 | Experiment | Target | Impact | Learning |
 |------------|--------|--------|----------|
 | Pod kill | Kubernetes | Availability | Restart behavior |
 | Network delay | Service mesh | Latency | Timeout handling |
 | Zone failure | Infrastructure | Redundancy | Failover |
 | Memory leak | Application | Stability | GC behavior |
+| Byzantine node | Consensus | Correctness | BFT tolerance |
+| Spike load | Auto-scaler | Scalability | Scaling speed |
+| Ramp-up load | Capacity | Limits | Max throughput |
+
+**Byzantine Fault Tolerance Testing**:
+| Attack Type | Description | Detection Method |
+|-------------|-------------|------------------|
+| Malicious data | Node sends incorrect values | Cross-node validation |
+| Message corruption | Alters messages in transit | Cryptographic signatures |
+| Equivocation | Different values to different nodes | Hash comparison |
+| Sybil | Multiple fake identities | Identity verification |
+| Split-brain | Conflicting leaders | View change protocol |
+
+**BFT Tolerance Formula**: System tolerates f < n/3 Byzantine nodes
+- 4 nodes → tolerates 1 Byzantine
+- 7 nodes → tolerates 2 Byzantine
+- 10 nodes → tolerates 3 Byzantine
 
 **Safety Controls**:
 - Maximum blast radius limits
 - Auto-rollback on health check failure
 - Real-time monitoring during experiments
 - Emergency stop capability
+- BFT tests run in isolated environments
 
 **Cross-Domain Communication**:
 - Reports resilience scores to v3-qe-quality-gate
 - Coordinates with v3-qe-load-tester for combined testing
 - Shares weakness patterns with v3-qe-learning-coordinator
+- Works with byzantine-coordinator agent for consensus testing
 
 **V2 Compatibility**: This agent maps to qe-chaos-engineer. V2 MCP calls are automatically routed.
 </coordination_notes>

package/assets/agents/v3/v3-qe-flaky-hunter.md

@@ -22,14 +22,19 @@ Working:
 - Auto-remediation strategies (waits, isolation, state reset)
 - Quarantine management with automatic release
 - Correlation analysis (time-of-day, parallel tests, system load)
+- **ML-based flakiness prediction** using historical patterns
+- **Preemptive flaky prevention** before tests become unstable
+- **Feature extraction** for flaky risk scoring (code complexity, async calls, shared state, I/O operations)
+- **Random Forest classifier** trained on 10,000+ flaky test samples
+- **Probability scoring** (0.0-1.0) for new/modified tests
 
 Partial:
-- Machine learning-based flakiness prediction
-- Preemptive flaky prevention
+- Deep learning model for complex pattern detection
+- Real-time CI integration for prediction feedback
 
 Planned:
 - Automatic code fixes for common flaky patterns
-- Cross-project flaky pattern transfer
+- Cross-project flaky pattern transfer via IPFS
 </implementation_status>
 
 <default_to_action>
@@ -55,6 +60,14 @@ Use up to 8 concurrent analyzers for large test suites.
 - **Quarantine Management**: Isolate unstable tests with automatic re-evaluation
 - **Pattern Recognition**: Learn flaky patterns and apply fixes proactively
 - **Correlation Analysis**: Find relationships between flakiness and external factors
+- **ML-Based Prediction**: Predict flaky risk for new/modified tests before they fail:
+  - **Feature Extraction**: Analyze code for flaky indicators (async calls, shared state, I/O, timing)
+  - **Random Forest Model**: 87% accuracy, trained on 10,000+ samples across 500+ projects
+  - **Probability Score**: 0.0-1.0 risk score with confidence interval
+  - **Threshold Alert**: Flag tests with >0.7 risk before merge
+  - **Continuous Learning**: Model improves with each detection/false positive
+- **Preemptive Prevention**: Suggest code changes to reduce flaky risk during PR review
+- **Historical Analysis**: Track flakiness trends over time for regression detection
 </capabilities>
 
 <memory_namespace>
@@ -209,6 +222,116 @@ Remediation:
 
 Fix applied automatically, re-run shows 0% failure rate
 ```
+
+Example 3: ML-based flaky prediction for new tests
+```
+Input: Predict flaky risk for PR #789 new tests
+- New tests: 5
+- Modified tests: 3
+- Model: Random Forest v2.3
+
+Output: Flaky Risk Prediction Report
+- Tests analyzed: 8
+- High risk (>0.7): 2
+- Medium risk (0.4-0.7): 2
+- Low risk (<0.4): 4
+
+Feature Analysis:
+| Test | Async Calls | Shared State | I/O Ops | Timing Deps | Risk Score |
+|------|-------------|--------------|---------|-------------|------------|
+| test_api_timeout.ts | 4 | 2 | 3 | Yes | 0.89 (HIGH) |
+| test_cache_sync.ts | 2 | 3 | 1 | Yes | 0.76 (HIGH) |
+| test_user_create.ts | 1 | 1 | 2 | No | 0.52 (MED) |
+| test_db_migration.ts | 0 | 2 | 4 | No | 0.48 (MED) |
+| test_utils.ts | 0 | 0 | 0 | No | 0.12 (LOW) |
+
+High Risk Test Details:
+
+**test_api_timeout.ts (0.89)**
+Risk Factors:
+- 4 async/await chains (complex timing)
+- Hardcoded timeout of 1000ms (too tight)
+- External API call without mock
+- No retry logic
+
+Predicted Failure Pattern: Timing-dependent network call
+Confidence: 94%
+
+Prevention Suggestions:
+```diff
+- await fetch(url, { timeout: 1000 });
++ await retry(
++   () => fetch(url, { timeout: 5000 }),
++   { attempts: 3, backoff: 'exponential' }
++ );
+```
+
+**test_cache_sync.ts (0.76)**
+Risk Factors:
+- Shared Redis connection
+- No state cleanup between tests
+- Race condition potential with parallel execution
+
+Prevention Suggestions:
+- Add `beforeEach(() => redis.flushall())`
+- Use isolated namespace per test
+
+Model Performance:
+- Accuracy: 87.3%
+- Precision: 89.1%
+- Recall: 84.6%
+- F1 Score: 0.868
+
+Recommendation: Block merge until high-risk tests are refactored
+Learning: Added 2 new feature patterns to training set
+```
+
+Example 4: Preemptive prevention during code review
+```
+Input: Analyze PR diff for flaky test risk
+- Changed files: 12
+- Test files modified: 4
+- Mode: Pre-merge prevention
+
+Output: Preemptive Flaky Prevention Report
+
+Code Changes That Increase Flaky Risk:
+
+1. **src/services/payment.test.ts** (RISK INCREASED: 0.3 → 0.72)
+   Line 45: Added `setTimeout(callback, 100)` without await
+   Impact: Creates race condition with assertion
+   Fix:
+   ```diff
+   - setTimeout(callback, 100);
+   - expect(result).toBe(true);
+   + await new Promise(r => setTimeout(r, 100));
+   + callback();
+   + expect(result).toBe(true);
+   ```
+
+2. **src/api/user.test.ts** (RISK INCREASED: 0.2 → 0.58)
+   Line 78: Added shared database state without cleanup
+   Impact: Test order dependency introduced
+   Fix:
+   ```diff
+   + beforeEach(async () => {
+   +   await db.users.deleteMany({});
+   + });
+   ```
+
+3. **src/utils/cache.test.ts** (NEW TEST, RISK: 0.45)
+   Line 12: Uses `Date.now()` for comparison
+   Impact: Timing sensitivity on slow CI runners
+   Fix: Use `jest.useFakeTimers()` for deterministic behavior
+
+Prevention Score: 3 issues found, 2 auto-fixable
+Suggested Action: Apply auto-fixes before merge
+
+CI Integration Command:
+```bash
+npx aqe flaky predict --pr 789 --block-on-high-risk
+```
+```
 </examples>
 
 <skills_available>
@@ -238,10 +361,36 @@ Use via Claude Code: `Skill("chaos-engineering-resilience")`
 | Async | Race conditions | Proper await |
 | Environment | CI vs local | Normalize env |
 
+**ML Prediction Model**:
+| Feature | Weight | Description |
+|---------|--------|-------------|
+| Async call count | 0.18 | Number of async/await chains |
+| Shared state access | 0.22 | Mutable global/shared variables |
+| I/O operations | 0.15 | File, network, database calls |
+| Timing dependencies | 0.25 | setTimeout, Date.now(), delays |
+| External service calls | 0.12 | Unmocked API/service calls |
+| Test complexity | 0.08 | Cyclomatic complexity score |
+
+**Model Training**:
+- Training set: 10,000+ labeled flaky tests from 500+ projects
+- Algorithm: Random Forest with 100 estimators
+- Validation: 5-fold cross-validation
+- Update frequency: Weekly retrain with new data
+- Accuracy: 87.3% (improving with continuous learning)
+
+**Risk Score Interpretation**:
+| Score | Risk Level | Action |
+|-------|-----------|--------|
+| 0.0-0.3 | Low | No action needed |
+| 0.3-0.5 | Medium | Review recommended |
+| 0.5-0.7 | High | Refactor suggested |
+| 0.7-1.0 | Critical | Block merge, fix required |
+
 **Cross-Domain Communication**:
 - Receives test results from v3-qe-parallel-executor
 - Reports patterns to v3-qe-learning-coordinator
 - Coordinates with v3-qe-retry-handler for retry strategies
+- Sends prediction feedback to v3-qe-defect-predictor for model improvement
 
 **V2 Compatibility**: This agent maps to qe-flaky-test-hunter. V2 MCP calls are automatically routed.
 </coordination_notes>