agentic-qe 3.6.0 → 3.6.2

package/.claude/agents/v3/qe-devils-advocate.md

@@ -0,0 +1,218 @@
+---
+name: qe-devils-advocate
+version: "3.6.0"
+updated: "2026-02-09"
+description: Meta-agent that challenges other agents' outputs by finding gaps, questioning assumptions, and critiquing completeness
+v2_compat: null
+domain: quality-assessment
+---
+
+<qe_agent_definition>
+<identity>
+You are the V3 QE Devil's Advocate, the adversarial reviewer in Agentic QE v3.
+Mission: Challenge other agents' outputs to surface gaps, blind spots, false positives, and unquestioned assumptions before results reach users.
+Domain: quality-assessment (ADR-064)
+V2 Compatibility: New in v3 -- no v2 equivalent.
+</identity>
+
+<implementation_status>
+Working:
+- Missing edge case detection (boundary values, null/undefined, concurrency)
+- False positive detection in security scans and coverage reports
+- Coverage gap critique (structural vs semantic coverage gaps)
+- Security blind spot identification (missing threat vectors)
+- Assumption questioning (implicit preconditions, happy-path bias)
+- Boundary value gap analysis (off-by-one, overflow, empty collections)
+- Error handling gap detection (missing catch blocks, swallowed errors)
+- Configurable severity thresholds and confidence filters
+- Per-review and cumulative statistics tracking
+
+Partial:
+- Integration with Queen Coordinator task pipeline
+- Cross-domain challenge coordination
+
+Planned:
+- Learning from past challenge outcomes (which challenges were acted on)
+- Auto-escalation for repeated unchallenged gaps
+</implementation_status>
+
+<default_to_action>
+Review outputs immediately when a ChallengeTarget is provided.
+Apply all applicable strategies without confirmation.
+Filter results by configured minConfidence and minSeverity.
+Report challenges in descending severity order.
+Always produce a summary even when no challenges are found.
+</default_to_action>
+
+<parallel_execution>
+Run all applicable challenge strategies concurrently against the target.
+Strategies are independent -- missing-edge-case, false-positive, coverage-gap, etc. run in parallel.
+Aggregate and sort results by severity after all strategies complete.
+Use up to 7 concurrent strategies per review.
+</parallel_execution>
+
+<capabilities>
+- **Missing Edge Case Detection**: Identify untested boundary values, null handling, concurrency, and error paths in test generation outputs
+- **False Positive Detection**: Flag likely false positives in security scans and coverage reports by checking for vague descriptions, low confidence, and known false-positive patterns
+- **Coverage Gap Critique**: Challenge coverage claims by checking for missing negative tests, missing integration paths, and semantic gaps not visible in line coverage
+- **Security Blind Spot Identification**: Find missing threat vectors (injection, auth bypass, SSRF, deserialization) not covered by security scan results
+- **Assumption Questioning**: Surface implicit assumptions in quality assessments, requirements validations, and defect predictions
+- **Boundary Value Gap Analysis**: Detect missing tests for off-by-one errors, integer overflow, empty/max-size collections, and Unicode edge cases
+- **Error Handling Gap Detection**: Find missing error handling for network failures, timeouts, malformed input, and resource exhaustion
+</capabilities>
+
+<memory_namespace>
+Reads:
+- aqe/v3/domains/test-generation/results/* - Test generation outputs to challenge
+- aqe/v3/domains/coverage-analysis/results/* - Coverage reports to critique
+- aqe/v3/domains/security-compliance/scans/* - Security scans to review
+- aqe/v3/domains/quality-assessment/reports/* - Quality reports to question
+
+Writes:
+- aqe/v3/devils-advocate/reviews/* - Challenge review results
+- aqe/v3/devils-advocate/stats/* - Cumulative challenge statistics
+- aqe/v3/devils-advocate/patterns/* - Learned gap patterns
+
+Coordination:
+- aqe/v3/queen/tasks/* - Task status updates
+- aqe/v3/domains/*/results/* - Cross-domain output access
+</memory_namespace>
+
+<learning_protocol>
+**MANDATORY**: When executed via Claude Code Task tool, you MUST call learning MCP tools.
+
+### Query Past Challenge Patterns BEFORE Review
+
+```typescript
+mcp__agentic-qe__memory_retrieve({
+  key: "devils-advocate/patterns",
+  namespace: "learning"
+})
+```
+
+### Required Learning Actions (Call AFTER Review)
+
+**1. Store Challenge Review Experience:**
+```typescript
+mcp__agentic-qe__memory_store({
+  key: "devils-advocate/outcome-{timestamp}",
+  namespace: "learning",
+  value: {
+    agentId: "qe-devils-advocate",
+    taskType: "challenge-review",
+    reward: <calculated_reward>,
+    outcome: {
+      targetType: "<test-generation|coverage-analysis|security-scan|...>",
+      targetAgentId: "<agent that produced the output>",
+      challengeCount: <number>,
+      highSeverityCount: <number>,
+      overallScore: <0-1>,
+      verdict: "PASSED|CHALLENGED"
+    },
+    patterns: {
+      gapsFound: ["<types of gaps found>"],
+      strategiesUsed: ["<strategies that produced findings>"]
+    }
+  }
+})
+```
+
+**2. Submit Review Result to Queen:**
+```typescript
+mcp__agentic-qe__task_submit({
+  type: "challenge-review-complete",
+  priority: "p1",
+  payload: {
+    targetAgentId: "...",
+    targetType: "...",
+    challengeCount: <number>,
+    highSeverityCount: <number>,
+    summary: "...",
+    challenges: [...]
+  }
+})
+```
+
+### Reward Calculation Criteria (0-1 scale)
+| Reward | Criteria |
+|--------|----------|
+| 1.0 | Actionable critical findings confirmed by follow-up |
+| 0.9 | High-severity gaps found with clear evidence |
+| 0.7 | Medium gaps found, strategies well-targeted |
+| 0.5 | Review completed, minor findings only |
+| 0.3 | Review completed, no significant findings (clean output) |
+| 0.0 | Review failed or produced only noise/false challenges |
+</learning_protocol>
+
+<output_format>
+- JSON for structured challenge results (challenges array, scores, summary)
+- Markdown for human-readable challenge reports
+- Challenges sorted by severity (critical > high > medium > low > informational)
+- Include challenge count, overall confidence score, and per-strategy breakdown
+</output_format>
+
+<examples>
+Example 1: Challenge test generation output
+```
+Input: Review test-generation output from agent test-gen-001
+  - 5 tests generated for UserService.createUser()
+  - All tests check happy path with valid data
+
+Output: CHALLENGED (Score: 0.38, 4 challenges)
+  [HIGH] Missing edge case: No test for duplicate email
+  [HIGH] Missing edge case: No test for empty/null username
+  [MEDIUM] Boundary value gap: No max-length validation test
+  [LOW] Error handling gap: No test for database connection failure
+  Summary: 5 tests cover only the happy path. No negative tests,
+  no boundary tests, no error handling tests. Test suite has
+  significant gaps in edge case coverage.
+```
+
+Example 2: Challenge security scan output
+```
+Input: Review security-scan output from agent sec-scan-001
+  - 0 vulnerabilities found
+  - Scanned: SQL injection, XSS
+
+Output: CHALLENGED (Score: 0.52, 2 challenges)
+  [HIGH] Security blind spot: No SSRF testing performed
+  [MEDIUM] Security blind spot: No deserialization checks
+  Summary: Scan covers injection and XSS but misses SSRF,
+  deserialization, and authentication bypass vectors.
+```
+</examples>
+
+<v3_integration>
+### Code Implementation
+The Devil's Advocate agent is implemented in `v3/src/agents/devils-advocate/`:
+- `agent.ts` - Core `DevilsAdvocate` class with `review()` method
+- `strategies.ts` - 7 pluggable challenge strategies
+- `types.ts` - Type definitions for targets, challenges, results
+
+### Usage
+```typescript
+import { DevilsAdvocate } from '@agentic-qe/v3';
+
+const da = DevilsAdvocate.createDevilsAdvocate({ minConfidence: 0.5 });
+
+const result = da.review({
+  type: 'test-generation',
+  agentId: 'test-gen-001',
+  domain: 'test-generation',
+  output: { testCount: 3, tests: [] },
+  timestamp: Date.now(),
+});
+```
+
+### Strategies
+| Strategy | Applies To | Detects |
+|----------|-----------|---------|
+| MissingEdgeCaseStrategy | test-generation | Untested edge cases, null handling |
+| FalsePositiveDetectionStrategy | security-scan, coverage-analysis | Likely false positives |
+| CoverageGapCritiqueStrategy | coverage-analysis | Semantic gaps in coverage |
+| SecurityBlindSpotStrategy | security-scan | Missing threat vectors |
+| AssumptionQuestioningStrategy | quality-assessment, defect-prediction, requirements | Implicit assumptions |
+| BoundaryValueGapStrategy | test-generation | Off-by-one, overflow, empty collections |
+| ErrorHandlingGapStrategy | test-generation, contract-validation | Missing error handling |
+</v3_integration>
+</qe_agent_definition>

package/.claude/agents/v3/qe-quality-criteria-recommender.md

@@ -245,7 +245,7 @@ interface QualityCriteriaAnalysis {
 ```
 
 ## Template Location
-Helper files installed to `.claude/agents/v3/helpers/quality-criteria/`:
+Helper files installed to `.claude/helpers/v3/quality-criteria/`:
 - `quality-criteria-reference-template.html` - HTML output template (MUST read before generating)
 - `htsm-categories.md` - Detailed category definitions
 - `evidence-classification.md` - Evidence type guidelines
@@ -404,7 +404,7 @@ if (!valid) {
 ### Output Validation
 If HTML output requested, always read template first:
 ```
-.claude/agents/v3/helpers/quality-criteria/quality-criteria-reference-template.html
+.claude/helpers/v3/quality-criteria/quality-criteria-reference-template.html
 ```
 </final_validation>
 </qe_agent_definition>

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

@@ -441,5 +441,5 @@ When ALL phases complete -> <promise>DEPLOYMENT_READY</promise>
 ---
 
 **Origin**: Adapted from Ralph Wiggum plugin (anthropics/claude-code)
-**Specialized for**: Agentic QE v3 Fleet with 59 QE agents
+**Specialized for**: Agentic QE v3 Fleet with 60 QE agents
 **Domains**: test-generation, test-execution, coverage-analysis, quality-assessment

package/.claude/skills/release/SKILL.md

@@ -144,27 +144,23 @@ Verify init completes without errors and creates the expected project structure
 # Version output
 node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js --version
 
-# Doctor check
-node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js doctor
+# System status
+node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js status
 ```
 Both must succeed without errors.
 
-#### 8d. Verify MCP Tools
-```bash
-# Verify MCP server can start and list tools
-node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js mcp --list-tools 2>&1 | head -30
-```
-Should list available MCP tools without crashing.
-
-#### 8e. Verify Self-Learning & Fleet Capabilities
+#### 8d. Verify Self-Learning & Fleet Capabilities
 ```bash
 cd /tmp/aqe-release-test
 
-# Verify memory/learning subsystem
-node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js memory list 2>&1 | head -10
+# Verify learning subsystem
+node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js learning stats 2>&1 | head -10
 
-# Verify agent spawning works
+# Verify agent listing works
 node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js agent list 2>&1 | head -10
+
+# Verify health check
+node /workspaces/agentic-qe-new/v3/dist/cli/bundle.js health 2>&1 | head -10
 ```
 These should respond (even if empty results) without errors, confirming the subsystems initialize properly.
 
@@ -177,34 +173,24 @@ rm -rf /tmp/aqe-release-test
 
 ### 9. Local CI Test Suite
 
-Run the same tests that CI runs on PRs (`optimized-ci.yml`) and during publish (`npm-publish.yml`). Skip e2e browser tests unless the user explicitly requests them.
+Run the same tests that CI runs on PRs and during publish. Skip e2e browser tests unless the user explicitly requests them.
 
 ```bash
 cd /workspaces/agentic-qe-new/v3
 
-# Journey tests (highest-value signal, from optimized-ci.yml)
-npm run test:journeys
-
-# Code Intelligence tests (MinCut/Graph algorithms, from optimized-ci.yml)
-npm run test:code-intelligence
-
-# Contract tests (if they exist, from optimized-ci.yml)
-npm run test:contracts 2>/dev/null || echo "No contract tests"
-
-# Infrastructure tests (from optimized-ci.yml)
-npm run test:infrastructure 2>/dev/null || echo "No infrastructure tests"
-
-# Regression tests (from optimized-ci.yml)
-npm run test:regression 2>/dev/null || echo "No regression tests"
-
-# Performance gates (from optimized-ci.yml)
+# Performance gates (fast — validates perf thresholds)
 npm run performance:gate
 
+# Regression tests (runs full unit suite)
+npm run test:regression
+
 # Full test:ci suite (from npm-publish.yml — excludes browser/e2e)
 npm run test:ci
 ```
 
-All mandatory test suites must pass. If any fail, diagnose and fix before continuing.
+Available test scripts: `test:unit`, `test:unit:fast`, `test:unit:heavy`, `test:unit:mcp`, `test:ci`, `test:regression`, `test:safe`, `test:perf`, `test:e2e`, `test:coverage`, `performance:gate`.
+
+All mandatory test suites must pass. Pre-existing MCP handler test failures (tests that need runtime initialization) are acceptable if they also fail on the main branch.
 
 **STOP — show all test results.**
 

package/.claude/skills/skills-manifest.json

@@ -904,7 +904,7 @@
   },
   "metadata": {
     "generatedBy": "Agentic QE Fleet",
-    "fleetVersion": "3.6.0",
+    "fleetVersion": "3.6.2",
     "manifestVersion": "1.3.0",
     "lastUpdated": "2026-02-04T00:00:00.000Z",
     "contributors": [

package/README.md

@@ -9,35 +9,9 @@
 <img alt="NPM Downloads" src="https://img.shields.io/npm/dw/agentic-qe">
 
 
-**V3 (Main)** | [V2 Documentation](v2/docs/V2-README.md) | [Changelog](CHANGELOG.md) | [Contributors](CONTRIBUTORS.md) | [Issues](https://github.com/proffesor-for-testing/agentic-qe/issues) | [Discussions](https://github.com/proffesor-for-testing/agentic-qe/discussions)
+**V3 (Main)** | [V2 Documentation](v2/docs/V2-README.md) | [Release Notes](docs/releases/README.md) | [Changelog](v3/CHANGELOG.md) | [Contributors](CONTRIBUTORS.md) | [Issues](https://github.com/proffesor-for-testing/agentic-qe/issues) | [Discussions](https://github.com/proffesor-for-testing/agentic-qe/discussions)
 
-> **V3** brings Domain-Driven Design architecture, 13 bounded contexts, 59 specialized QE agents, TinyDancer intelligent model routing, ReasoningBank learning with Dream cycles, HNSW vector search, mathematical Coherence verification, full MinCut/Consensus integration across all 13 domains, and deep integration with [Claude Flow](https://github.com/ruvnet/claude-flow) and [Agentic Flow](https://github.com/ruvnet/agentic-flow).
-
-### What's New in v3.6.0
-
-- **Enterprise Integration Domain** — SOAP/WSDL, SAP RFC/BAPI/IDoc, OData, ESB/middleware, message broker, and Segregation of Duties testing (contributed by [@fndlalit](https://github.com/fndlalit))
-- **8 New Agents** — `qe-soap-tester`, `qe-sap-rfc-tester`, `qe-sap-idoc-tester`, `qe-middleware-validator`, `qe-odata-contract-tester`, `qe-message-broker-tester`, `qe-sod-analyzer`, `qe-pentest-validator`
-- **5 New Skills** — `enterprise-integration-testing`, `middleware-testing-patterns`, `wms-testing-patterns`, `observability-testing-patterns`, `pentest-validation` (Tier 3)
-- **Pentest Validation** — Shannon-inspired graduated exploit validation with "No Exploit, No Report" quality gate and 3-tier exploitation
-- **StrongDM Tier 1** — Loop detection + token dashboard for software delivery governance (ADR-062)
-- **Fleet: 59 agents, 75 skills across 13 domains**
-
-### What's New in v3.5.0
-
-- **Governance ON by Default** - @claude-flow/guidance integration with 7 unbreakable QE invariants (ADR-058)
-- **QCSD 2.0 Complete Lifecycle** - All 4 phases: Ideation → Refinement → Development → CI/CD Verification
-- **Infrastructure Self-Healing Enterprise** - 12 enterprise error signatures (SAP, Salesforce, Payment Gateway)
-
-### What's New in v3.4.2
-
-- **Skill Validation System** - 4-layer trust tiers with schemas, validators, and evaluation suites (ADR-056)
-- **CLI Validation Commands** - `aqe skill report`, `aqe eval run`, regression detection
-
-### What's New in v3.4.0
-
-- **AG-UI Protocol** - Anthropic's streaming agent-to-user interface with real-time progress updates
-- **A2A Protocol** - Google's agent-to-agent interoperability standard for cross-tool communication
-- **A2UI Components** - Unified UI combining AG-UI streaming with A2A event handling
+> **V3** brings Domain-Driven Design architecture, 13 bounded contexts, 60 specialized QE agents, TinyDancer intelligent model routing, ReasoningBank learning with Dream cycles, HNSW vector search, mathematical Coherence verification, full MinCut/Consensus integration across all 13 domains, and deep integration with [Claude Flow](https://github.com/ruvnet/claude-flow) and [Agentic Flow](https://github.com/ruvnet/agentic-flow).
 
 🏗️ **DDD Architecture** | 🧠 **ReasoningBank + Dream Cycles** | 🎯 **TinyDancer Model Routing** | 🔍 **HNSW Vector Search** | 👑 **Queen Coordinator** | 📊 **O(log n) Coverage** | 🔗 **Claude Flow Integration** | 🎯 **13 Bounded Contexts** | 📚 **75 QE Skills** | 🧬 **Coherence Verification** | ✅ **Trust Tiers** | 🛡️ **Governance**
 
@@ -88,7 +62,7 @@ claude "Use qe-flaky-hunter to analyze the last 100 test runs and stabilize flak
 
 **What V3 provides:**
 - ✅ **13 DDD Bounded Contexts**: Organized by business domain (test-generation, coverage-analysis, security-compliance, enterprise-integration, etc.)
-- ✅ **59 QE Agents**: Including Queen Coordinator for hierarchical orchestration (52 main + 7 TDD subagents)
+- ✅ **60 QE Agents**: Including Queen Coordinator for hierarchical orchestration (53 main + 7 TDD subagents)
 - ✅ **TinyDancer Model Routing**: 3-tier intelligent routing (Haiku/Sonnet/Opus) for cost optimization
 - ✅ **ReasoningBank Learning**: HNSW-indexed pattern storage with experience replay
 - ✅ **O(log n) Coverage Analysis**: Sublinear algorithms for efficient gap detection
@@ -134,7 +108,7 @@ claude "Assess code quality and provide deployment recommendation"
 | **AI testing tools are expensive** | TinyDancer 3-tier model routing reduces costs by matching task complexity to appropriate model |
 | **No memory between test runs—every analysis starts from scratch** | ReasoningBank remembers patterns, strategies, and what works for your codebase |
 | **Agents waste tokens reading irrelevant code** | Code Intelligence provides token reduction with semantic search and knowledge graphs |
-| **Quality engineering requires complex coordination** | Queen Coordinator orchestrates 59 agents across 13 domains with consensus and MinCut topology |
+| **Quality engineering requires complex coordination** | Queen Coordinator orchestrates 60 agents across 13 domains with consensus and MinCut topology |
 | **Tools don't understand your testing frameworks** | Works with Jest, Cypress, Playwright, Vitest, Mocha, Jasmine, AVA |
 
 ---
@@ -240,7 +214,7 @@ The **qe-queen-coordinator** manages the entire fleet with intelligent task dist
 ```
 
 **Capabilities:**
-- Orchestrate 59 QE agents concurrently across 13 domains
+- Orchestrate 60 QE agents concurrently across 13 domains
 - TinyDancer 3-tier model routing (Haiku/Sonnet/Opus) with confidence-based decisions
 - Byzantine fault-tolerant consensus for critical quality decisions
 - MinCut graph-based topology optimization for self-healing coordination
@@ -253,6 +227,34 @@ claude "Use qe-queen-coordinator to orchestrate release validation for v2.1.0 wi
 
 ---
 
+### 🤝 Agent Teams & Fleet Coordination
+
+The Queen Coordinator is extended with **Agent Teams** (ADR-064) for hybrid fleet communication:
+
+| Feature | Description |
+|---------|-------------|
+| **Mailbox Messaging** | Direct agent-to-agent and domain-scoped broadcast messaging |
+| **Distributed Tracing** | TraceContext propagation across messages for end-to-end task visibility |
+| **Dynamic Scaling** | Workload-based auto-scaling with configurable policies and cooldowns |
+| **Competing Hypotheses** | Multi-agent root cause investigation with evidence scoring, auto-triggered on critical failures |
+| **Federation** | Cross-service routing with health monitoring and service discovery |
+| **Circuit Breakers** | Per-domain fault isolation with automatic recovery |
+| **Task DAG** | Topological ordering with cycle detection for multi-step workflows |
+
+**Fleet Tiers** — Activate the level of coordination your project needs:
+
+| Tier | Agents | Best For |
+|------|--------|----------|
+| **Lite** | 1-4 | Small projects, focused tasks |
+| **Standard** | 5-10 | Team projects, multi-domain coordination |
+| **Full** | 11-15 | Enterprise, cross-fleet federation |
+
+```bash
+claude "Use qe-queen-coordinator with agent teams to investigate flaky test failures across test-execution and defect-intelligence domains"
+```
+
+---
+
 ### 🧠 ReasoningBank Learning System
 
 V3 agents learn and improve through the **ReasoningBank** pattern storage:
@@ -466,17 +468,17 @@ npx @claude-flow/cli@latest agent spawn -t qe-test-architect --name test-gen
 
 ---
 
-### 📊 59 Specialized QE Agents
+### 📊 60 Specialized QE Agents
 
 | Category | Count | Highlights |
 |----------|-------|------------|
-| **Main QE Agents** | 52 | Test generation, coverage, security, performance, accessibility, enterprise integration, pentest validation |
+| **Main QE Agents** | 53 | Test generation, coverage, security, performance, accessibility, enterprise integration, pentest validation |
 | **TDD Subagents** | 7 | RED/GREEN/REFACTOR with code review |
 
 **V2 Backward Compatibility**: All V2 agents map to V3 equivalents automatically.
 
 <details>
-<summary><b>📋 View All Main QE Agents (52)</b></summary>
+<summary><b>📋 View All Main QE Agents (53)</b></summary>
 
 | Agent | Domain | Purpose |
 |-------|--------|---------|
@@ -524,6 +526,7 @@ npx @claude-flow/cli@latest agent spawn -t qe-test-architect --name test-gen
 | qe-product-factors-assessor | quality-assessment | SFDIPOT product factors analysis |
 | qe-test-idea-rewriter | test-generation | Transform passive tests to active actions |
 | qe-quality-criteria-recommender | quality-assessment | HTSM v6.3 Quality Criteria analysis |
+| qe-devils-advocate | quality-assessment | Adversarial review of agent outputs |
 
 </details>
 
@@ -794,7 +797,7 @@ agentic-qe/
 │   │   ├── mcp/             # MCP server
 │   │   └── cli/             # V3 CLI
 │   ├── tests/               # 5,600+ tests
-│   └── assets/agents/       # 59 QE agent definitions (52 main + 7 subagents)
+│   └── assets/agents/       # 60 QE agent definitions (53 main + 7 subagents)
 ├── v2/                      # V2 Implementation (Legacy)
 │   ├── src/                 # V2 source code
 │   ├── tests/               # V2 tests

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentic-qe",
-  "version": "3.6.0",
+  "version": "3.6.2",
   "description": "Agentic Quality Engineering V3 - Domain-Driven Design Architecture with 13 Bounded Contexts, O(log n) coverage analysis, ReasoningBank learning, 59 specialized QE agents, mathematical Coherence verification, deep Claude Flow integration",
   "main": "./v3/dist/index.js",
   "types": "./v3/dist/index.d.ts",

package/scripts/cloud-db-config.json

@@ -19,7 +19,7 @@
     }
   },
   "sync": {
-    "enabled": false,
+    "enabled": true,
     "mode": "incremental",
     "interval": "5m",
     "tables": [

package/v3/CHANGELOG.md

@@ -5,6 +5,50 @@ 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).
 
+## [3.6.2] - 2026-02-10
+
+### Fixed
+
+- **YAML parser empty array crash (Issue #244)** — `aqe init --auto` no longer fails on re-runs when `config.yaml` has empty array fields like `disabled:` with no items. The custom YAML parser now normalizes known array fields after parsing, and `mergeConfigs()` uses defensive `Array.isArray()` checks.
+- **Agent parse errors on helper files (Issue #243)** — Helper reference files (`htsm-categories.md`, `evidence-classification.md`) and the generated `README.md` are no longer placed inside `.claude/agents/v3/` where `claude doctor` would incorrectly parse them as agent definitions. Helpers now install to `.claude/helpers/v3/` and the agents index writes to `.claude/docs/v3-agents-index.md`.
+
+### Changed
+
+- **Helper files location** — Agent helper/reference files (quality-criteria templates, SFDIPOT templates) now install to `.claude/helpers/v3/` instead of `.claude/agents/v3/helpers/`. Updated all path references in `quality-criteria-service.ts` and agent definitions.
+
+## [3.6.1] - 2026-02-09
+
+### Added
+
+- **Agent Teams Integration (ADR-064)** — Hybrid fleet architecture layering Claude Code Agent Teams communication patterns on the existing Queen Coordinator. 4-phase implementation: Foundation, Hybrid Architecture, Learning & Observability, Advanced Patterns.
+- **Agent Teams Adapter** — Direct mailbox messaging between agents with domain-scoped teams (2-4 agents per domain), team lead/teammate model, and subscription-based event delivery.
+- **Fleet Tier Selector** — Tiered fleet activation (smoke/standard/deep/crisis) that controls agent count and token costs based on trigger context (commit, PR, release, incident).
+- **Task Dependency DAG** — Topological ordering with cycle detection for multi-step task workflows. DAGScheduler for automated execution of ready tasks.
+- **TeammateIdle Hook** — Auto-assigns pending tasks to idle agents, reducing Queen bottleneck for task distribution.
+- **TaskCompleted Hook** — Extracts patterns from completed tasks and trains them into ReasoningBank automatically. Quality gate validation with exit code 2 rejection.
+- **Domain Circuit Breakers** — Per-domain fault isolation with configurable failure thresholds, half-open recovery probing, and criticality-based configs.
+- **Domain Team Manager** — Creates and manages domain-scoped agent teams with health monitoring, scaling, and rebalancing.
+- **HNSW Graph Construction** — Real O(log n) HNSW insert and search in unified memory, replacing the O(n) linear scan stub.
+- **Distributed Tracing** — TraceCollector with W3C-style TraceContext propagation encoded into AgentMessage correlationId fields. Queen traces full task lifecycles.
+- **Competing Hypotheses** — HypothesisManager for multi-agent root cause investigation with evidence scoring, confidence tracking, and convergence (evidence-scoring, unanimous, majority, timeout). Auto-triggered on p0/p1 task failures.
+- **Cross-Fleet Federation** — FederationMailbox with service registry, domain-based routing, health monitoring via heartbeats, and graceful degradation for unreachable services.
+- **Dynamic Agent Scaling** — DynamicScaler with workload metrics collection, configurable scaling policies (queue depth, idle ratio, error rate thresholds), cooldown enforcement, and executor callbacks. Wired into Queen's metrics loop.
+- **ReasoningBank Pattern Store Adapter** — Bridges TaskCompletedHook pattern extraction to QEReasoningBank storage with domain detection, type mapping, and confidence propagation.
+- **promotePattern() Implementation** — Completes the ReasoningBank promotion stub: delegates to PatternStore.promote() and publishes pattern:promoted events.
+- **Devil's Advocate Agent** — `qe-devils-advocate` agent that challenges other agents' outputs by finding gaps and questioning assumptions.
+- **397+ New Tests** — 282 coordination tests, 67 hook tests, 48 learning tests covering all ADR-064 phases including adapter tracing integration and latency benchmarks.
+
+### Fixed
+
+- **6 CodeQL Alerts** — Resolved security alerts in enterprise-integration services (input validation, type safety).
+- **Pattern Training Pipeline** — Connected the disconnected TaskCompletedHook → ReasoningBank pipeline so patterns are automatically trained on task completion.
+- **Queen Operational Wiring** — All Phase 3+4 modules (tracing, dynamic scaler, hypotheses) are now called by Queen's operational flow, not just initialized as shelf-ware.
+
+### Changed
+
+- **Queen Coordinator** — Extended with tracing (startTrace on submitTask, completeSpan/failSpan on completion/failure), dynamic scaling (metrics feed + evaluate + execute in metrics loop), and competing hypotheses (auto-investigation on critical failures).
+- **Agent Teams Adapter** — sendMessage() and broadcast() now encode TraceContext into correlationId when provided, enabling end-to-end distributed tracing.
+
 ## [3.6.0] - 2026-02-08
 
 ### Added

package/v3/README.md

@@ -5,14 +5,14 @@
 [![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, 13 Bounded Contexts, 59 Specialized QE Agents, 75 QE Skills, and ReasoningBank Learning
+> Domain-Driven Quality Engineering with Mathematical Coherence Verification, 13 Bounded Contexts, 60 Specialized QE Agents, 75 QE Skills, and ReasoningBank Learning
 
 ### Key Features
 
 | Feature | Description |
 |---------|-------------|
 | **75 QE Skills** | Quality engineering skills with 4-tier trust validation system |
-| **59 QE Agents** | Specialized agents for test generation, security, coverage, enterprise integration, and more |
+| **60 QE Agents** | Specialized agents for test generation, security, coverage, enterprise integration, and more |
 | **13 DDD Domains** | Modular bounded contexts for all quality engineering needs |
 | **MCP Integration** | Full Claude Code integration via Model Context Protocol |
 | **AG-UI/A2A Protocols** | Industry-standard agent streaming and interoperability |
@@ -48,7 +48,7 @@ npx aqe test generate src/
 
 ## Why Agentic QE?
 
-- **59 Specialized QE Agents** - Domain-focused quality engineering agents (52 main + 7 subagents)
+- **60 Specialized QE Agents** - Domain-focused quality engineering agents (53 main + 7 subagents)
 - **75 QE Skills** - 46 Tier 3 verified + 29 additional (QCSD swarms, n8n testing, enterprise integration, qe-* domains)
 - **13 DDD Bounded Contexts** - Modular, extensible architecture
 - **TinyDancer Model Routing** - 3-tier intelligent routing for cost optimization
@@ -632,7 +632,7 @@ console.log(`Quality gate: ${gate.value.passed ? 'PASSED' : 'FAILED'}`);
 | Module System | CommonJS | ESM |
 | Memory | SQLite only | HNSW + SQLite hybrid |
 | Learning | Basic patterns | ReasoningBank + SONA + Dream Cycles |
-| Agents | 32 | 59 QE agents (52 main + 7 subagents) |
+| Agents | 32 | 60 QE agents (53 main + 7 subagents) |
 | Skills | 35 | 75 QE skills (46 Tier 3 + 29 additional) |
 | Coverage | O(n) | O(log n) |
 | Pattern Search | Linear | O(log n) HNSW indexing |
@@ -706,9 +706,9 @@ See the [Migration Guide](./docs/MIGRATION-GUIDE.md) for detailed instructions a
 }
 ```
 
-## 59 QE Agents
+## 60 QE Agents
 
-Agentic QE includes 59 specialized quality engineering agents (52 main + 7 subagents) organized by domain:
+Agentic QE includes 60 specialized quality engineering agents (53 main + 7 subagents) organized by domain:
 
 ### Test Generation Domain
 `qe-test-architect`, `qe-tdd-specialist`, `qe-tdd-red`, `qe-tdd-green`, `qe-tdd-refactor`, `qe-property-tester`, `qe-mutation-tester`, `qe-bdd-generator`
@@ -720,7 +720,7 @@ Agentic QE includes 59 specialized quality engineering agents (52 main + 7 subag
 `qe-coverage-specialist`, `qe-gap-detector`, `qe-risk-analyzer`
 
 ### Quality Assessment Domain
-`qe-quality-gate`, `qe-metrics-optimizer`, `qe-deployment-advisor`
+`qe-quality-gate`, `qe-metrics-optimizer`, `qe-deployment-advisor`, `qe-devils-advocate`
 
 ### Defect Intelligence Domain
 `qe-defect-intelligence`, `qe-regression-analyzer`, `qe-root-cause-analyzer`