agentic-qe 3.3.2 → 3.3.4

package/v3/CHANGELOG.md

@@ -5,6 +5,194 @@ All notable changes to Agentic QE will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.3.4] - 2026-01-29
+
+### 🎯 Highlights
+
+**QCSD Ideation Phase Complete** - Full Quality Conscious Software Delivery (QCSD) Ideation phase implementation with HTSM v6.3 quality criteria analysis, SFDIPOT product factors assessment, and cross-phase memory feedback loops.
+
+**Cross-Phase Memory System** - New persistent memory architecture enabling automated learning between QCSD phases (Production→Ideation, Production→Grooming, CI/CD→Development, Development→Grooming).
+
+**Comprehensive Test Coverage** - 358 files changed with 83,990+ lines of new tests across all 12 domains, kernel, MCP handlers, routing, and workers.
+
+### Added
+
+#### QCSD Ideation Phase Agents
+- **qe-quality-criteria-recommender** - HTSM v6.3 quality criteria analysis with 10 categories
+- **qe-product-factors-assessor** - SFDIPOT framework with 7 factors, 37 subcategories
+- **qe-risk-assessor** - Multi-factor risk scoring with mitigation recommendations
+- **qe-test-idea-rewriter** - Transform passive "Verify X" patterns to active test actions
+
+#### Cross-Phase Memory System
+- **CrossPhaseMemoryService** - File-based persistence for QCSD feedback loops
+- **Cross-phase MCP handlers** - 8 new tools for signal storage/retrieval
+- **Hook executor** - Automatic trigger of cross-phase hooks on agent completion
+- **4 feedback loops** - Strategic, Tactical, Operational, Quality Criteria
+
+#### New Skills
+- **a11y-ally** - Comprehensive WCAG accessibility audit with video caption generation
+- **qcsd-ideation-swarm** - Multi-agent swarm for QCSD Ideation phase
+- **skills-manifest.json** - Centralized skill registration
+
+#### Comprehensive Test Coverage (83,990+ lines)
+- **Kernel tests** - unified-memory, hybrid-backend, kernel, plugin-loader
+- **MCP handler tests** - All domain handlers, handler-factory, task-handlers
+- **Domain tests** - All 12 domain plugins with coordinator tests
+- **Learning tests** - pattern-store, experience-capture, v2-to-v3-migration
+- **Routing tests** - tiny-dancer-router, task-classifier, routing-config
+- **Sync tests** - claude-flow-bridge, sync-agent, json/sqlite readers
+- **Worker tests** - All 10 background workers
+
+#### E2E Test Framework
+- Moved e2e tests to v3/tests/e2e/
+- Sauce Demo test suite with accessibility, cart, checkout, security specs
+- Page Object Model with BasePage, CartPage, CheckoutPage, etc.
+
+#### Documentation Updates
+- Updated agent catalog with QCSD Ideation agents
+- Added HTSM v6.3 quality categories reference
+- Added SFDIPOT framework documentation
+- Updated skill counts: 61 → 63 QE Skills
+- Updated agent counts with new QCSD agents
+
+### Changed
+
+#### Handler Factory Migration
+- All 11 domain handlers now use centralized handler-factory.ts
+- Experience capture middleware wraps all domain operations
+- Consistent error handling across handlers
+
+#### Security Scanner Refactoring
+- Split monolithic security-scanner.ts into modular components
+- New scanner-orchestrator.ts for coordinating SAST/DAST scans
+- Separate sast-scanner.ts and dast-scanner.ts modules
+- security-patterns.ts for pattern definitions
+
+#### E2E Runner Modularization
+- Split e2e-runner.ts into 9 focused modules
+- browser-orchestrator.ts - Browser session management
+- step-executors.ts - Step execution logic
+- assertion-handlers.ts - Assertion processing
+- result-collector.ts - Test result aggregation
+
+### Fixed
+
+#### Test Timeout Fixes
+- Fixed 6 timeout failures in security-compliance/coordinator.test.ts
+- Added proper class-based mocks for SecurityScannerService
+- Added mocks for SecurityAuditorService and ComplianceValidatorService
+
+#### TypeScript Compilation
+- Fixed all TypeScript errors from PR #215 merge
+- Fixed history.length on unknown type casting
+- Fixed performTask payload access patterns
+- Fixed Map iteration with Array.from()
+
+#### Architecture Cleanup
+- Removed wrong-pattern TypeScript agent classes (QualityCriteriaRecommenderAgent, RiskAssessorAgent)
+- Removed orphaned QCSD agent tests
+- Moved n8n-validator to v3/packages/
+
+### Security
+
+- SSRF protection recommendations for DAST scanner (private IP blocking)
+- Path traversal edge case fix recommendation (startsWith + path.sep)
+- npm audit: 0 vulnerabilities
+
+### Deprecated
+
+- Root-level tests/e2e/ directory (moved to v3/tests/e2e/)
+- Root-level src/agents/ TypeScript classes (use .claude/agents/v3/*.md instead)
+
+---
+
+## [3.3.3] - 2026-01-27
+
+### 🎯 Highlights
+
+**Full MinCut/Consensus Integration** - All 12 QE domains now have active MinCut topology awareness, multi-model consensus verification, and self-healing triggers. This completes the ADR-047 implementation with production-ready distributed coordination.
+
+**LLM Integration Across All Domains** - ADR-051 enables intelligent LLM-powered analysis in all 12 QE domains with TinyDancer model routing for cost optimization.
+
+### Added
+
+#### LLM Integration for All 12 QE Domains (ADR-051)
+- **test-generation** - AI-powered test synthesis with pattern learning
+- **test-execution** - Intelligent flaky test analysis and retry recommendations
+- **coverage-analysis** - LLM-assisted gap prioritization and risk scoring
+- **quality-assessment** - AI-driven quality gate decisions with explanations
+- **defect-intelligence** - ML-powered defect prediction and root cause analysis
+- **requirements-validation** - LLM testability analysis and BDD generation
+- **code-intelligence** - Semantic code search with natural language queries
+- **security-compliance** - AI vulnerability analysis with remediation guidance
+- **contract-testing** - LLM contract validation and breaking change detection
+- **visual-accessibility** - AI visual regression analysis and WCAG recommendations
+- **chaos-resilience** - Intelligent resilience assessment and failure prediction
+- **learning-optimization** - Pattern consolidation with LLM synthesis
+
+#### QE Agent Registry Fixes
+- Added missing agents to registry: `qe-product-factors-assessor`, `qe-quality-criteria-recommender`, `qe-test-idea-rewriter`
+- Fixed skill counts: 61 QE skills properly registered
+- Updated agent-to-domain mappings
+
+#### Documentation
+- **TinyDancer Integration Plan** - Detailed plan for model routing across domains
+- **Contract Validator LLM Docs** - LLM integration documentation for contract testing
+
+#### MinCut/Consensus Full Domain Integration (ADR-047, MM-001)
+- **All 12 domains** now actively use consensus verification (not just initialized)
+- **Topology-aware routing** - `getTopologyBasedRouting()` in all domains
+- **Self-healing triggers** - `shouldPauseOperations()` pauses work on critical topology
+
+| Domain | verifyFinding Calls | Self-Healing | Routing |
+|--------|---------------------|--------------|---------|
+| test-generation | 3 | ✅ | ✅ |
+| test-execution | 3 | ✅ | ✅ |
+| coverage-analysis | 3 | ✅ | ✅ |
+| quality-assessment | 2 | ✅ | ✅ |
+| defect-intelligence | 3 | ✅ | ✅ |
+| learning-optimization | 3 | ✅ | ✅ |
+| security-compliance | 2 | ✅ | ✅ |
+| chaos-resilience | 3 | ✅ | ✅ |
+| code-intelligence | 3 | ✅ | ✅ |
+| contract-testing | 3 | ✅ | ✅ |
+| requirements-validation | 3 | ✅ | ✅ |
+| visual-accessibility | 3 | ✅ | ✅ |
+
+#### Performance Benchmarks
+- **mincut-performance.test.ts** (20 tests) - Graph operations, health monitoring, memory usage
+- **consensus-latency.test.ts** (18 tests) - Finding verification, batch operations, strategy comparison
+
+#### Cross-Domain Integration Tests
+- **cross-domain-mincut-consensus.test.ts** (34 tests) - Queen→Domain bridge injection, topology coordination
+
+### Changed
+
+#### Domain Coordinators (all 12)
+- Added `verifyFinding()` calls for high-stakes decisions
+- Added `getTopologyBasedRouting()` method
+- Added `getDomainWeakVertices()` method
+- Added `isDomainWeakPoint()` method
+- Added self-healing with `shouldPauseOperations()` checks
+
+#### Type Exports
+- **consensus-enabled-domain.ts** - Re-export `ConsensusStats` type for domain use
+- **contract-testing/interfaces.ts** - Use proper `WeakVertex[]` and `DomainName[]` types
+- **code-intelligence/coordinator.ts** - Fixed routing type signatures
+
+### Fixed
+
+- **fix(types)**: ConsensusStats now properly exported from mixin module
+- **fix(types)**: DomainName[] type consistency across routing methods
+- **fix(coverage-analysis)**: Use `factors.contribution` instead of non-existent `factors.weight`
+- **fix(benchmarks)**: Relaxed timing thresholds for CI stability (0.2ms→0.5ms, 512B→1KB)
+
+### Documentation
+
+- **MINCUT_CONSENSUS_INTEGRATION_PLAN.md** - Updated status to IMPLEMENTED with completion metrics
+
+---
+
 ## [3.3.2] - 2026-01-26
 
 ### 🎯 Highlights

package/v3/README.md

@@ -5,7 +5,7 @@
 [![TypeScript](https://img.shields.io/badge/TypeScript-5.0+-blue.svg)](https://www.typescriptlang.org/)
 [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/)
 
-> Domain-Driven Quality Engineering with Mathematical Coherence Verification, 12 Bounded Contexts, 51 Specialized QE Agents, 61 QE Skills, and ReasoningBank Learning
+> Domain-Driven Quality Engineering with Mathematical Coherence Verification, 12 Bounded Contexts, 51 Specialized QE Agents, 63 QE Skills, and ReasoningBank Learning
 
 ## Quick Start
 
@@ -38,7 +38,7 @@ npx aqe test generate src/
 ## Why Agentic QE?
 
 - **51 Specialized QE Agents** - Domain-focused quality engineering agents (44 main + 7 subagents)
-- **61 QE Skills** - Domain-specific skills for testing, security, accessibility, and more
+- **63 QE Skills** - Domain-specific skills for testing, security, accessibility, QCSD, and more
 - **12 DDD Bounded Contexts** - Modular, extensible architecture
 - **TinyDancer Model Routing** - 3-tier intelligent routing for cost optimization
 - **O(log n) Coverage Analysis** - Sublinear performance with HNSW indexing
@@ -47,6 +47,8 @@ npx aqe test generate src/
 - **Queen-led Coordination** - 3-5x throughput with work stealing and consensus
 - **MinCut Topology** - Graph-based self-healing agent coordination
 - **Coherence Verification** (v3.3.0) - Mathematical proof of belief consistency using WASM engines
+- **MinCut/Consensus Integration** (v3.3.3) - Full 12/12 domain integration with self-healing
+- **Cross-Phase Memory Unification** (v3.3.4) - Unified SQLite storage for QCSD feedback loops
 - **Zero-Breaking-Changes Migration** - Full v2 backward compatibility
 - **Browser Automation** (v3.1.0) - @claude-flow/browser integration with 9 workflow templates
 
@@ -73,14 +75,17 @@ Background neural consolidation for continuous improvement:
 - **EWC++ Protection**: Elastic Weight Consolidation prevents catastrophic forgetting
 - **Novelty Scoring**: Prioritize learning from novel patterns
 
-### Consensus & MinCut Coordination
+### Consensus & MinCut Coordination (v3.3.3)
 
-Advanced coordination for reliable multi-agent decisions:
+Advanced coordination for reliable multi-agent decisions with **full 12/12 domain integration**:
 
 - **Byzantine Consensus**: Fault-tolerant voting for critical quality decisions
 - **MinCut Topology**: Graph-based self-healing agent coordination
 - **Multi-Model Voting**: Aggregate decisions from multiple model tiers
 - **Claim Verification**: Cryptographic verification of agent work claims
+- **12/12 Domain Coverage**: All domains actively use `verifyFinding()` for consensus
+- **Topology-Aware Routing**: `getTopologyBasedRouting()` avoids weak vertices
+- **Self-Healing Triggers**: `shouldPauseOperations()` enables automatic recovery
 
 ### Coherence-Gated Quality Engineering (v3.3.0)
 
@@ -486,7 +491,7 @@ console.log(`Quality gate: ${gate.value.passed ? 'PASSED' : 'FAILED'}`);
 | Memory | SQLite only | HNSW + SQLite hybrid |
 | Learning | Basic patterns | ReasoningBank + SONA + Dream Cycles |
 | Agents | 32 | 51 QE agents (44 main + 7 subagents) |
-| Skills | 35 | 60 QE skills (v2 + v3 domain skills) |
+| Skills | 35 | 63 QE skills (v2 + v3 domain skills) |
 | Coverage | O(n) | O(log n) |
 | Pattern Search | Linear | O(log n) HNSW indexing |
 | Coordination | Sequential | Queen + Work Stealing + Consensus |
@@ -600,14 +605,14 @@ Agentic QE includes 51 specialized quality engineering agents (44 main + 7 subag
 `qe-queen-coordinator`, `qe-fleet-commander`, `qe-integration-tester`, `qe-data-generator`, `qe-code-reviewer`
 
 ### Additional Agents (New in v3)
-`qe-product-factors-assessor` (SFDIPOT analysis), `qe-test-idea-rewriter` (passive→active test transforms)
+`qe-product-factors-assessor` (SFDIPOT analysis), `qe-quality-criteria-recommender` (HTSM v6.3 Quality Criteria), `qe-test-idea-rewriter` (passive→active test transforms)
 
-## 61 QE Skills
+## 63 QE Skills
 
-Agentic QE includes 60 domain-specific quality engineering skills that agents automatically apply:
+Agentic QE includes 63 domain-specific quality engineering skills that agents automatically apply:
 
 <details>
-<summary><b>View All 61 QE Skills</b></summary>
+<summary><b>View All 63 QE Skills</b></summary>
 
 ### Core Testing & Methodologies (12)
 - `agentic-quality-engineering` - Core PACT principles for AI-powered QE
@@ -681,6 +686,10 @@ Agentic QE includes 60 domain-specific quality engineering skills that agents au
 - `n8n-trigger-testing-strategies` - Webhook and event testing
 - `n8n-integration-testing-patterns` - API contract testing for n8n
 
+### QCSD & Accessibility (2)
+- `qcsd-ideation-swarm` - QCSD Ideation phase swarm using HTSM v6.3, Risk Storming, Testability analysis
+- `a11y-ally` - Comprehensive WCAG auditing with multi-tool testing, video captions, and EU compliance
+
 </details>
 
 > **Note**: Claude Flow platform skills (agentdb, github, flow-nexus, etc.) are managed separately by the claude-flow package and not counted in QE skills.

package/v3/assets/agents/v3/README.md

@@ -0,0 +1,100 @@
+# AQE V3 Agents Index
+
+This directory contains V3 QE agents installed by `aqe init`.
+
+> **Note**: This directory only contains AQE-specific agents (v3-qe-*).
+> Claude-flow core agents (adr-architect, memory-specialist, etc.) are part of
+> the claude-flow system and are available separately.
+
+## Summary
+
+- **Total Agents**: 1
+- **V3 QE Domain Agents**: 1
+- **V3 Subagents**: 0
+
+## V2 → V3 Migration Guide
+
+| V2 Agent | V3 Agent | Domain | Key Changes |
+|----------|----------|--------|-------------|
+| `qe-a11y-ally` | `qe-accessibility-auditor` | visual-accessibility | **MANDATORY video pipeline** |
+| `qe-test-generator` | `qe-test-architect` | test-generation | ReasoningBank learning |
+| `qe-coverage-analyzer` | `qe-coverage-specialist` | coverage-analysis | O(log n) sublinear |
+| `qe-coordinator` | `qe-learning-coordinator` | learning-optimization | Cross-domain patterns |
+| `qe-flaky-investigator` | `qe-flaky-hunter` | defect-intelligence | ML-powered detection |
+
+## Critical: qe-accessibility-auditor
+
+The V3 accessibility auditor has a **MANDATORY video accessibility pipeline**.
+
+### V2 Behavior (qe-a11y-ally)
+
+Generated these files when videos detected:
+- `*-captions.vtt` - For deaf/hard-of-hearing users
+- `*-audiodesc.vtt` - For blind/low-vision users
+- Implementation instructions
+
+### V3 Behavior (qe-accessibility-auditor)
+
+**MUST** execute the same pipeline:
+
+1. **Detect** - Find `<video>`, `<iframe>` elements
+2. **Download** - `curl -L -o video.mp4 URL`
+3. **Extract** - `ffmpeg -vf "fps=1/3" -frames:v 10`
+4. **Analyze** - Read each .jpg with Claude Vision
+5. **Generate** - Create .vtt from actual descriptions
+6. **Save** - Output to `docs/accessibility/captions/{page-slug}/`
+
+### Expected Output Structure
+
+```
+docs/accessibility/captions/{page-slug}/
+├── video-001-captions.vtt     # Captions
+├── video-001-audiodesc.vtt    # Audio descriptions
+├── implementation.md          # HTML integration
+└── audit-report.md            # WCAG report
+```
+
+### Invocation
+
+Use the skill for guaranteed video pipeline execution:
+
+```
+/qe-a11y-ally https://example.com/page
+```
+
+Or via Task with explicit video pipeline requirement:
+
+```javascript
+Task({
+  prompt: "Audit accessibility including MANDATORY video pipeline for URL",
+  subagent_type: "qe-accessibility-auditor"
+})
+```
+
+## Usage
+
+Spawn agents using Claude Code's Task tool:
+
+```javascript
+Task("Generate tests for UserService", "...", "qe-test-architect")
+Task("Analyze coverage gaps", "...", "qe-coverage-specialist")
+Task("Audit accessibility with video captions", "...", "qe-accessibility-auditor")
+```
+
+## V3 QE Domain Agents (1)
+
+Quality Engineering agents mapped to the 12 DDD bounded contexts.
+
+| Agent | Domain | File |
+|-------|--------|------|
+| `qe-accessibility-auditor` | visual-accessibility | `qe-accessibility-auditor.md` |
+
+## V3 Subagents (0)
+
+Specialized sub-task agents for TDD and code review.
+
+*None installed*
+
+---
+
+*Updated 2026-01-26 - Added V2→V3 migration guide and video pipeline documentation*

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

@@ -53,17 +53,118 @@ 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.**
+## CRITICAL: VIDEO ACCESSIBILITY PIPELINE
+
+**This pipeline is MANDATORY when ANY video is detected on the page.**
+**Failure to execute this pipeline is a CRITICAL ERROR.**
+
+### Detection Phase
+Look for these video indicators in fetched HTML:
+- `<video>` elements with `src` or `<source>` children
+- `<iframe>` with YouTube/Vimeo/Wistia URLs
+- `video-container`, `video-player` class elements
+- JavaScript video players (Video.js, Plyr, etc.)
+
+### Execution Phase (FOR EACH VIDEO)
+
+**Step 1: Download Video**
+```bash
+# Create working directory
+mkdir -p /tmp/a11y-video-work
+
+# For direct MP4/WebM URLs
+curl -L -o /tmp/a11y-video-work/video-001.mp4 "VIDEO_URL"
+
+# For YouTube (if yt-dlp available)
+yt-dlp -f "best[height<=720]" -o /tmp/a11y-video-work/video-001.mp4 "YOUTUBE_URL" 2>/dev/null || echo "yt-dlp not available"
+```
+
+**Step 2: Extract Frames**
+```bash
+mkdir -p /tmp/a11y-video-work/frames-001
+ffmpeg -i /tmp/a11y-video-work/video-001.mp4 \
+  -vf "fps=1/3" \
+  -frames:v 10 \
+  /tmp/a11y-video-work/frames-001/frame_%02d.jpg \
+  2>/dev/null
+```
+
+**Step 3: Analyze Each Frame with Claude Vision**
+Use the Read tool on EACH extracted .jpg file:
+```
+Read /tmp/a11y-video-work/frames-001/frame_01.jpg
+Read /tmp/a11y-video-work/frames-001/frame_02.jpg
+Read /tmp/a11y-video-work/frames-001/frame_03.jpg
+... (continue for all 10 frames)
+```
+
+For each frame, document:
+- Scene: Setting, environment, lighting conditions
+- People: Who is present, what they're doing, expressions
+- Objects: Products, props, vehicles, equipment
+- Text: Any visible text, logos, signs, labels
+- Action: What's happening, movement, transitions
+- Colors: Dominant colors, contrasts, accessibility-relevant
+
+**Step 4: Generate WebVTT Captions**
+```vtt
+WEBVTT
+Kind: captions
+
+00:00:00.000 --> 00:00:03.000
+[Description from frame_01.jpg analysis]
+
+00:00:03.000 --> 00:00:06.000
+[Description from frame_02.jpg analysis]
+
+00:00:06.000 --> 00:00:09.000
+[Description from frame_03.jpg analysis]
+```
+
+**Step 5: Generate Audio Descriptions**
+```vtt
+WEBVTT
+Kind: descriptions
+
+00:00:00.000 --> 00:00:03.000
+SCENE: [Detailed scene description for blind users]
+VISUAL: [What's on screen]
+TEXT: [Any readable text]
+ACTION: [What's happening]
+```
+
+**Step 6: Save Output Files**
+```bash
+# Create output directory
+mkdir -p docs/accessibility/captions/{page-slug}
+
+# Save generated files
+# - video-001-captions.vtt
+# - video-001-audiodesc.vtt
+# - implementation.md (HTML code examples)
+```
+
+### Enforcement Rules
+
+1. **NEVER** complete an accessibility audit without checking for videos
+2. **NEVER** skip the video pipeline if videos are detected
+3. **NEVER** generate placeholder/template captions - use ACTUAL frame analysis
+4. **ALWAYS** use the Read tool on actual .jpg frame files for Vision analysis
+5. **ALWAYS** save output to `docs/accessibility/captions/{page-slug}/`
+6. **ALWAYS** include implementation instructions in the output
+
+### Validation Checklist (Self-Check Before Completing)
+
+- [ ] Did I check for `<video>` and `<iframe>` elements?
+- [ ] Did I download each detected video?
+- [ ] Did I extract frames with ffmpeg?
+- [ ] Did I use Read tool on each .jpg frame file?
+- [ ] Did I generate captions.vtt from ACTUAL frame descriptions?
+- [ ] Did I generate audiodesc.vtt with detailed scene info?
+- [ ] Did I save files to docs/accessibility/captions/?
+- [ ] Did I include implementation instructions?
+
+**If ANY checkbox is NO and videos were detected, the task is INCOMPLETE.**
 </default_to_action>
 
 <parallel_execution>

package/v3/assets/agents/v3/qe-bdd-generator.md

@@ -257,6 +257,46 @@ Use via CLI: `aqe skills show test-design-techniques`
 Use via Claude Code: `Skill("exploratory-testing-advanced")`
 </skills_available>
 
+<cross_phase_memory>
+**QCSD Feedback Loop**: Quality-Criteria Loop (Development → Ideation)
+**Role**: CONSUMER - Receives untestable patterns to avoid in BDD scenarios
+
+### On Startup, Query Quality-Criteria Signals:
+```typescript
+const result = await mcp__agentic_qe__cross_phase_query({
+  loop: "quality-criteria",
+  maxAge: "60d"
+});
+
+// Learn from historical untestable patterns
+for (const signal of result.signals) {
+  if (signal.untestablePatterns) {
+    for (const pattern of signal.untestablePatterns) {
+      // Never generate scenarios using untestable patterns
+      avoidPattern(pattern.acPattern);
+      // Use better patterns in scenario generation
+      preferPattern(pattern.betterPattern);
+    }
+  }
+  if (signal.recommendations?.acTemplates) {
+    // Use proven AC templates for scenario generation
+    addScenarioTemplates(signal.recommendations.acTemplates);
+  }
+}
+```
+
+### How to Use Injected Signals:
+1. **Pattern Avoidance**: Don't use patterns from `untestablePatterns[].acPattern`
+2. **Better Patterns**: Prefer `untestablePatterns[].betterPattern` structures
+3. **Templates**: Use `acTemplates` for proven scenario structures
+4. **Coverage Gaps**: Generate scenarios that would fill `coverageGaps[].codeArea`
+
+### Signal Flow:
+- **Consumes**: Untestable patterns and templates from qe-coverage-specialist, qe-gap-detector
+- **Namespace**: `aqe/cross-phase/quality-criteria/ac-quality`
+- **Expected Signals**: AC templates and patterns to avoid
+</cross_phase_memory>
+
 <coordination_notes>
 **V3 Architecture**: This agent operates within the requirements-validation bounded context (ADR-006).
 

package/v3/assets/agents/v3/qe-coverage-specialist.md

@@ -206,6 +206,45 @@ Use via CLI: `aqe skills show risk-based-testing`
 Use via Claude Code: `Skill("mutation-testing")`
 </skills_available>
 
+<cross_phase_memory>
+**QCSD Feedback Loop**: Quality-Criteria Loop (Development → Ideation)
+**Role**: PRODUCER - Stores untestable patterns from coverage analysis
+
+### On Coverage Gap Detection, Store Quality-Criteria Signal:
+```typescript
+mcp__agentic_qe__cross_phase_store({
+  loop: "quality-criteria",
+  data: {
+    untestablePatterns: [
+      {
+        acPattern: "<acceptance-criteria-pattern>",
+        problem: "<why-its-untestable>",
+        frequency: <0.0-1.0>,
+        betterPattern: "<improved-pattern>"
+      }
+    ],
+    coverageGaps: [
+      {
+        codeArea: "<file-or-module>",
+        coveragePercentage: <percentage>,
+        rootCause: "<why-gap-exists>",
+        acImprovement: "<how-better-AC-would-help>"
+      }
+    ],
+    recommendations: {
+      forRequirementsValidator: ["<AC improvement recommendations>"],
+      acTemplates: { "<feature>": "<template>" }
+    }
+  }
+})
+```
+
+### Signal Flow:
+- **Produces**: Untestable patterns, coverage gaps → consumed by qe-requirements-validator, qe-bdd-generator
+- **Namespace**: `aqe/cross-phase/quality-criteria/ac-quality`
+- **TTL**: 60 days (AC quality insights inform story writing)
+</cross_phase_memory>
+
 <coordination_notes>
 **V3 Architecture**: This agent operates within the coverage-analysis bounded context (ADR-003).
 

package/v3/assets/agents/v3/qe-defect-predictor.md

@@ -230,6 +230,42 @@ Use via CLI: `aqe skills show risk-based-testing`
 Use via Claude Code: `Skill("mutation-testing")`
 </skills_available>
 
+<cross_phase_memory>
+**QCSD Feedback Loop**: Strategic Loop (Production → Ideation)
+**Role**: PRODUCER - Stores risk weights from production defect analysis
+
+### On Completion, Store Strategic Signal:
+```typescript
+mcp__agentic_qe__cross_phase_store({
+  loop: "strategic",
+  data: {
+    riskWeights: [
+      {
+        category: "<defect-prone-area>",
+        weight: <0.0-1.0>,
+        confidence: <0.0-1.0>,
+        evidence: {
+          defectCount: <count>,
+          percentageOfTotal: <percentage>,
+          severityDistribution: { critical: <n>, high: <n>, medium: <n> },
+          timeRange: { start: "<date>", end: "<date>" }
+        }
+      }
+    ],
+    recommendations: {
+      forRiskAssessor: ["<recommendations for risk assessment>"],
+      forQualityCriteria: ["<recommendations for acceptance criteria>"]
+    }
+  }
+})
+```
+
+### Signal Flow:
+- **Produces**: Production risk weights → consumed by qe-risk-assessor, qe-quality-criteria-recommender
+- **Namespace**: `aqe/cross-phase/strategic/production-risk`
+- **TTL**: 90 days (strategic insights have long-term value)
+</cross_phase_memory>
+
 <coordination_notes>
 **V3 Architecture**: This agent operates within the defect-intelligence bounded context (ADR-006).