agentic-qe 3.7.16 → 3.7.18

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

@@ -0,0 +1,91 @@
+# Step 1: Analyze Story Content (Flag Detection)
+
+## Prerequisites
+- Refinement swarm skill invoked
+- User story / acceptance criteria provided
+- OUTPUT_FOLDER determined
+
+## Instructions
+
+### MANDATORY: Complete this analysis before Step 2.
+
+### Step 0: Retrieve Ideation Phase Signals (Cross-Phase Consumption)
+
+Before analyzing story content, retrieve the most recent Ideation phase signals from memory.
+
+**MCP Method (preferred):**
+
+```javascript
+mcp__agentic-qe__memory_query({
+  pattern: "qcsd-ideation-*",
+  namespace: "qcsd-ideation",
+  limit: 1
+})
+```
+
+**CLI Fallback:**
+
+```bash
+npx @claude-flow/cli@latest memory search --query "qcsd-ideation" --namespace qcsd-ideation --limit 1
+```
+
+### Step 1: Scan Story Content and Detect Flags
+
+Scan the user story, acceptance criteria, and any referenced specifications to SET these flags. Do not skip any flag.
+
+### Flag Detection (Check ALL SEVEN)
+
+```
+HAS_API = FALSE
+  Set TRUE if story mentions ANY of: API, endpoint, REST, GraphQL, gRPC,
+  webhook, contract, consumer, provider, OpenAPI, Swagger, request/response,
+  HTTP method, payload, schema, backward compatibility
+
+HAS_REFACTORING = FALSE
+  Set TRUE if story mentions ANY of: refactor, restructure, rewrite,
+  migrate, replace, consolidate, extract, decompose, simplify, optimize,
+  technical debt, legacy, modernize, upgrade
+
+HAS_DEPENDENCIES = FALSE
+  Set TRUE if story mentions ANY of: dependency, library, package, version,
+  upgrade, third-party, external service, integration, SDK, plugin,
+  framework update, breaking change, compatibility
+
+HAS_SECURITY = FALSE
+  Set TRUE if story mentions ANY of: authentication, authorization, encryption,
+  token, OAuth, JWT, RBAC, security, vulnerability, OWASP, input validation,
+  sanitization, XSS, CSRF, injection, secrets, credentials
+
+HAS_MIDDLEWARE = FALSE
+  Set TRUE if story mentions ANY of: middleware, ESB, message broker, MQ,
+  Kafka, RabbitMQ, integration bus, API gateway, message queue, pub/sub,
+  event bus, service bus, ActiveMQ, NATS, Redis Streams
+
+HAS_SAP_INTEGRATION = FALSE
+  Set TRUE if story mentions ANY of: SAP, OData, RFC, BAPI, IDoc,
+  S/4HANA, EWM, ECC, ABAP, CDS view, Fiori, SAP Cloud Integration,
+  SAP PI/PO, SAP Gateway, SAP connector
+
+HAS_AUTHORIZATION = FALSE
+  Set TRUE if story mentions ANY of: SoD, segregation of duties,
+  role conflict, authorization object, T-code, user role,
+  access control matrix, GRC, RBAC policy, permission matrix,
+  privilege escalation, role assignment
+```
+
+### MANDATORY: Output Flag Detection Results
+
+You MUST output flag detection results in a structured format showing each flag value with evidence before proceeding.
+
+## Success Criteria
+- [ ] Ideation phase signals retrieved (or documented as unavailable)
+- [ ] All SEVEN flags evaluated with evidence
+- [ ] Flag detection results output
+- [ ] Expected agent count calculated (3 core + conditional count + 1 transformation)
+
+## Output
+Provide: All flag values with evidence, expected agent count, Ideation baseline data.
+
+## Navigation
+- On success: proceed to Step 2 by reading `steps/02-core-agents.md`
+- On failure: halt and report which flags could not be evaluated

package/.claude/skills/qcsd-refinement-swarm/steps/02-core-agents.md

@@ -0,0 +1,40 @@
+# Step 2: Spawn Core Agents (Parallel Batch 1)
+
+## Prerequisites
+- Step 1 (Flag Detection) completed
+- User story content available
+- OUTPUT_FOLDER determined
+
+## Instructions
+
+### CRITICAL: Spawn ALL THREE core agents in ONE message
+
+| Agent | Domain | Role |
+|-------|--------|------|
+| qe-product-factors-assessor | requirements-validation | SFDIPOT (7 factors, 37 subcategories) analysis |
+| qe-bdd-generator | requirements-validation | BDD Gherkin scenario generation |
+| qe-requirements-validator | requirements-validation | INVEST validation and testability scoring |
+
+### Agent 1: Product Factors Assessor
+Analyze the story using SFDIPOT product factors (Structure, Function, Data, Interface, Platform, Operations, Time). Must read the SFDIPOT reference material first. Save to: `${OUTPUT_FOLDER}/02-product-factors.md`
+
+### Agent 2: BDD Generator
+Generate comprehensive BDD Gherkin scenarios from the story acceptance criteria. Must produce Given/When/Then scenarios covering happy paths, edge cases, and negative cases. Save to: `${OUTPUT_FOLDER}/03-bdd-scenarios.md`
+
+### Agent 3: Requirements Validator
+Validate the story against INVEST criteria (Independent, Negotiable, Valuable, Estimable, Small, Testable). Assess testability scoring and acceptance criteria completeness. Save to: `${OUTPUT_FOLDER}/04-requirements-validation.md`
+
+### Post-Spawn Confirmation
+Tell the user all 3 core agents are running and WAIT.
+
+## Success Criteria
+- [ ] All THREE agents spawned in ONE message
+- [ ] Each agent has complete story content
+- [ ] Post-spawn confirmation sent
+
+## Output
+Confirmation that 3 agents are running in background.
+
+## Navigation
+- On success: proceed to Step 3 by reading `steps/03-batch1-results.md`
+- On failure: respawn missing agents

package/.claude/skills/qcsd-refinement-swarm/steps/03-batch1-results.md

@@ -0,0 +1,40 @@
+# Step 3: Wait for Batch 1 & Extract Results
+
+## Prerequisites
+- Step 2: All 3 core agents spawned
+
+## Instructions
+
+### WAIT for all three agents to complete. Do NOT proceed early.
+
+### Extract from each agent:
+
+**From qe-product-factors-assessor:**
+- SFDIPOT factor scores and priorities
+- Key quality risks identified
+- Product factor coverage assessment
+
+**From qe-bdd-generator:**
+- Number of BDD scenarios generated
+- Coverage of acceptance criteria
+- Edge cases and negative scenarios
+
+**From qe-requirements-validator:**
+- INVEST validation score
+- Testability score
+- Acceptance criteria completeness percentage
+- Missing requirements identified
+
+### Output Metrics Summary Box with all extracted values.
+
+## Success Criteria
+- [ ] All three agents returned results
+- [ ] All metrics extracted
+- [ ] Summary box output with actual values
+
+## Output
+Complete Batch 1 metrics for use in subsequent steps.
+
+## Navigation
+- On success: proceed to Step 4 by reading `steps/04-conditional-agents.md`
+- On failure: retry failed agents

package/.claude/skills/qcsd-refinement-swarm/steps/04-conditional-agents.md

@@ -0,0 +1,35 @@
+# Step 4: Spawn Conditional Agents (Parallel Batch 2)
+
+## Prerequisites
+- Step 3 completed with core metrics
+- Flag values from Step 1
+
+## Instructions
+
+### IF A FLAG IS TRUE, YOU MUST SPAWN THAT AGENT
+
+| Flag | Agent | Domain | Output File |
+|------|-------|--------|-------------|
+| HAS_API | qe-contract-validator | contract-testing | `05-contract-validation.md` |
+| HAS_REFACTORING | qe-impact-analyzer | code-intelligence | `06-impact-analysis.md` |
+| HAS_DEPENDENCIES | qe-dependency-mapper | code-intelligence | `07-dependency-analysis.md` |
+| HAS_SECURITY | (handled by core) | - | - |
+| HAS_MIDDLEWARE | qe-middleware-validator | enterprise-integration | `08-middleware-validation.md` |
+| HAS_SAP_INTEGRATION | qe-odata-contract-tester | enterprise-integration | `09-odata-contracts.md` |
+| HAS_AUTHORIZATION | qe-sod-analyzer | enterprise-integration | `10-sod-analysis.md` |
+
+If ALL flags are FALSE, skip to Step 5.
+
+Otherwise, spawn all applicable agents in ONE message and WAIT for completion.
+
+## Success Criteria
+- [ ] All flagged agents spawned (or skipped if no flags)
+- [ ] Agent count validation passed
+- [ ] All conditional agents completed
+
+## Output
+Conditional agent scores for decision logic.
+
+## Navigation
+- On success: proceed to Step 5 by reading `steps/05-decision-synthesis.md`
+- On failure: spawn missing agents

package/.claude/skills/qcsd-refinement-swarm/steps/05-decision-synthesis.md

@@ -0,0 +1,43 @@
+# Step 5: Synthesize Results & Determine Recommendation
+
+## Prerequisites
+- Core and conditional agent results available
+
+## Instructions
+
+### Apply READY/CONDITIONAL/NOT-READY decision logic EXACTLY
+
+Derive metrics from agent outputs and apply the decision tree:
+
+**NOT-READY conditions (ANY triggers NOT-READY):**
+- Testability score < 40%
+- INVEST validation fails on 3+ criteria
+- Critical SFDIPOT gaps in Structure or Function
+- Missing acceptance criteria for core functionality
+- Contract validation failures on existing APIs (if applicable)
+
+**READY conditions (ALL required for READY):**
+- Testability score >= 70%
+- INVEST validation passes on 5+ criteria
+- All SFDIPOT factors assessed with no critical gaps
+- BDD scenarios cover all acceptance criteria
+- No unresolved blocking issues from conditional agents
+
+**Default: CONDITIONAL**
+
+### Record decision with full metrics and rationale.
+
+If CONDITIONAL, list specific conditions that must be met.
+If NOT-READY, list specific blockers with remediation steps.
+
+## Success Criteria
+- [ ] All metrics derived from actual agent outputs
+- [ ] Decision logic applied exactly
+- [ ] FINAL RECOMMENDATION: READY / CONDITIONAL / NOT-READY
+
+## Output
+Recommendation, metrics, rationale, and improvement actions.
+
+## Navigation
+- On success: proceed to Step 6 by reading `steps/06-report-generation.md`
+- On failure: resolve missing metrics

package/.claude/skills/qcsd-refinement-swarm/steps/06-report-generation.md

@@ -0,0 +1,42 @@
+# Step 6: Generate Refinement Report
+
+## Prerequisites
+- Step 5 completed with FINAL RECOMMENDATION
+- All agent reports available
+
+## Instructions
+
+Generate the complete Refinement Report and save to `${OUTPUT_FOLDER}/01-executive-summary.md`.
+
+The report MUST include:
+1. Executive Summary with recommendation and key metrics
+2. SFDIPOT Product Factors Analysis (from qe-product-factors-assessor)
+3. BDD Scenarios (from qe-bdd-generator)
+4. Requirements Validation (from qe-requirements-validator)
+5. Conditional Analysis sections (for each conditional agent that ran)
+6. Recommended Actions (P0/P1/P2 prioritized)
+7. Test Strategy Summary
+
+### Report Validation Checklist
+
+```
++-- Executive Summary is complete with all metrics
++-- Recommendation matches decision logic
++-- SFDIPOT section covers all 7 factors
++-- BDD section includes all generated scenarios
++-- Requirements section includes INVEST scores
++-- Conditional sections present for all spawned agents
++-- Actions are specific and actionable
++-- Report saved to output folder
+```
+
+## Success Criteria
+- [ ] Complete report with all sections
+- [ ] Report saved to correct path
+
+## Output
+Confirmation that report is saved.
+
+## Navigation
+- On success: proceed to Step 7 by reading `steps/07-learning-persistence.md`
+- On failure: complete missing sections

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

@@ -0,0 +1,53 @@
+# Step 7: Store Learnings & Persist State
+
+## Prerequisites
+- Step 6 completed
+- All metrics and flags available
+
+## Instructions
+
+### ENFORCEMENT: ALWAYS RUN. NOT OPTIONAL.
+
+Store refinement findings to memory for cross-phase feedback:
+
+**Step 1: Store to memory**
+
+```javascript
+mcp__agentic-qe__memory_store({
+  key: `qcsd-refinement-${storyId}-${Date.now()}`,
+  namespace: "qcsd-refinement",
+  value: {
+    storyId, recommendation, testabilityScore,
+    investScore, sfdipotPriorities, bddScenarioCount,
+    flags, agentsInvoked, timestamp
+  }
+})
+```
+
+**Step 2: Share learnings**
+
+```javascript
+mcp__agentic-qe__memory_share({
+  sourceAgentId: "qcsd-refinement-swarm",
+  targetAgentIds: ["qe-test-idea-rewriter"],
+  knowledgeDomain: "refinement-test-patterns"
+})
+```
+
+**Step 3: Save persistence record**
+
+Save to `${OUTPUT_FOLDER}/11-learning-persistence.json` with complete metrics, flags, and cross-phase signals.
+
+### Fallback: CLI commands if MCP unavailable.
+
+## Success Criteria
+- [ ] Memory store executed with actual values
+- [ ] Learning persistence JSON saved
+- [ ] Cross-phase signals documented
+
+## Output
+Confirmation of persistence.
+
+## Navigation
+- On success: proceed to Step 8 by reading `steps/08-transformation.md`
+- On failure: retry persistence

package/.claude/skills/qcsd-refinement-swarm/steps/08-transformation.md

@@ -0,0 +1,36 @@
+# Step 8: Apply Test Idea Rewriter (Transformation)
+
+## Prerequisites
+- Step 7 completed
+- BDD scenarios from Step 2 available
+- All test ideas collected from all agents
+
+## Instructions
+
+### ENFORCEMENT: ALWAYS RUN qe-test-idea-rewriter
+
+Spawn qe-test-idea-rewriter to transform ALL test ideas from passive descriptions
+into active, exploratory test charters.
+
+This agent must:
+1. Read all BDD scenarios from qe-bdd-generator output
+2. Read all test suggestions from other agents
+3. Transform each passive test idea ("Verify X works") into an active charter
+4. Apply the Session-Based Test Management (SBTM) charter format
+5. Prioritize test ideas by risk and SFDIPOT factor weights
+
+Save output to: `${OUTPUT_FOLDER}/12-rewritten-test-ideas.md`
+
+WAIT for the agent to complete.
+
+## Success Criteria
+- [ ] qe-test-idea-rewriter spawned and completed
+- [ ] All test ideas transformed
+- [ ] Output saved to correct path
+
+## Output
+Transformed test ideas for final report.
+
+## Navigation
+- On success: proceed to Step 9 by reading `steps/09-final-output.md`
+- On failure: retry the transformation agent

package/.claude/skills/qcsd-refinement-swarm/steps/09-final-output.md

@@ -0,0 +1,46 @@
+# Step 9: Final Output & Completion Summary
+
+## Prerequisites
+- All steps (1-8) completed
+
+## Instructions
+
+Display the completion summary:
+
+```
++---------------------------------------------------------------------+
+|                  QCSD REFINEMENT SWARM COMPLETE                       |
++---------------------------------------------------------------------+
+|                                                                      |
+|  Story Assessed: [Story ID/Title]                                     |
+|  Reports Generated: [count]                                           |
+|  Output Folder: ${OUTPUT_FOLDER}                                     |
+|                                                                      |
+|  REFINEMENT SCORES:                                                   |
+|  +-- Testability Score:       __%                                     |
+|  +-- INVEST Score:            __/6                                    |
+|  +-- SFDIPOT Coverage:        __%                                     |
+|  +-- BDD Scenarios:           __ generated                           |
+|  +-- Test Ideas Rewritten:    __                                      |
+|                                                                      |
+|  RECOMMENDATION: [READY / CONDITIONAL / NOT-READY]                    |
+|  REASON: [1-2 sentence rationale]                                     |
+|                                                                      |
+|  DELIVERABLES:                                                        |
+|  +-- 01-executive-summary.md                                          |
+|  +-- 02-product-factors.md                                            |
+|  +-- 03-bdd-scenarios.md                                              |
+|  +-- 04-requirements-validation.md                                    |
+|  [conditional reports as applicable]                                  |
+|  +-- 11-learning-persistence.json                                     |
+|  +-- 12-rewritten-test-ideas.md                                       |
+|                                                                      |
++---------------------------------------------------------------------+
+```
+
+## Success Criteria
+- [ ] Completion summary displayed
+- [ ] All deliverables listed
+
+## Navigation
+- This is the terminal step.

package/.claude/skills/sherlock-review/SKILL.md

@@ -101,6 +101,9 @@ npm test -- --coverage
 1. [Action based on findings]
 ```
 
+## Minimum Findings Enforcement
+Every investigation MUST surface at least 3 weighted observations (CRITICAL=3, HIGH=2, MEDIUM=1, LOW=0.5). Elementary observations count at INFORMATIONAL=0.25 weight. A Sherlock investigation that finds nothing is a failed investigation -- Holmes always finds clues.
+
 ---
 
 ## Investigation Scenarios

package/.claude/skills/skill-builder/SKILL.md

@@ -806,6 +806,109 @@ Complete API documentation: [API_REFERENCE.md](docs/API_REFERENCE.md)
 
 ---
 
+## Micro-File Step Architecture (BMAD-006)
+
+For complex skills that exceed 1,000 lines, use the micro-file step architecture
+to combat LLM "lost in the middle" context degradation. When a skill's SKILL.md grows
+beyond a certain threshold, the LLM loses critical details in the middle of the file.
+Splitting into step files ensures each step gets full attention.
+
+### When to Apply
+
+| SKILL.md Size | Recommendation |
+|---------------|----------------|
+| Under 1,000 lines | Single file is fine |
+| 1,000 - 1,500 lines | SHOULD split into step files |
+| Over 1,500 lines | MUST split into step files |
+
+### Directory Convention
+
+```
+my-skill/
+  SKILL.md            # Compact orchestrator (200-300 lines)
+  steps/
+    01-first-step.md   # Self-contained step file
+    02-second-step.md
+    03-third-step.md
+    ...
+```
+
+### Step File Format
+
+Each step file MUST be self-contained and include:
+
+1. **Title** with step number
+2. **Prerequisites** listing what data is needed from prior steps
+3. **Instructions** with complete details (no references back to SKILL.md)
+4. **Success Criteria** as a checklist
+5. **Output** describing what this step produces
+6. **Navigation** with explicit Read tool instructions for the next step
+
+Example step file:
+
+```markdown
+# Step 3: Wait for Core Agent Results
+
+## Prerequisites
+- Step 2 completed (all core agents spawned)
+
+## Instructions
+STOP. Do not proceed until all three background tasks from Step 2 return.
+
+When all agents complete, extract these metrics:
+- Metric A from Agent 1
+- Metric B from Agent 2
+- Metric C from Agent 3
+
+## Success Criteria
+- [ ] All core agents completed
+- [ ] All metrics extracted and recorded
+
+## Output
+Metrics summary for use in Step 4.
+
+## Navigation
+- On success: proceed to Step 4 by reading `steps/04-next-step.md`
+- On failure: retry failed agents from Step 2
+```
+
+### Orchestrator Format
+
+The rewritten SKILL.md orchestrator MUST include:
+
+1. **YAML frontmatter** (preserved exactly from original)
+2. **Title and overview** (brief description of the skill)
+3. **Parameters** section
+4. **Enforcement rules** table
+5. **Step Execution Protocol** with:
+   - Numbered step list with file paths
+   - Explicit Read tool instructions (e.g., `Read({ file_path: "..." })`)
+   - Resume support via `--from-step N`
+6. **Agent/component inventory** table
+7. **Quality gate thresholds** (if applicable)
+8. **Key principle** closing statement
+
+The orchestrator should be 200-300 lines. All procedural detail lives in step files.
+
+### Key Rules
+
+- Step files MUST be under 500 lines each
+- Step files MUST NOT reference content in SKILL.md (self-contained)
+- Step files MUST include explicit `Read()` instructions for the next step (Gap 9 fix)
+- ALL original content MUST be preserved across step files -- nothing is deleted
+- The orchestrator is a table of contents and control flow, not a content dump
+
+### Reference Implementation
+
+See the QCSD swarm skills for working examples:
+- `qcsd-production-swarm/` -- 12 agents, 9 steps, feedback loop
+- `qcsd-refinement-swarm/` -- 10 agents, 9 steps, test idea rewriter
+- `qcsd-cicd-swarm/` -- 10 agents, 9 steps, deployment advisor
+- `qcsd-development-swarm/` -- 10 agents, 9 steps, defect predictor
+- `qcsd-ideation-swarm/` -- 9 agents, 8 steps, URL browser cascade
+
+---
+
 ## Examples from the Wild
 
 ### Example 1: Simple Documentation Skill

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

@@ -904,7 +904,7 @@
   },
   "metadata": {
     "generatedBy": "Agentic QE Fleet",
-    "fleetVersion": "3.7.16",
+    "fleetVersion": "3.7.18",
     "manifestVersion": "1.3.0",
     "lastUpdated": "2026-02-04T00:00:00.000Z",
     "contributors": [

package/CHANGELOG.md

@@ -5,6 +5,35 @@ All notable changes to the Agentic QE project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.7.18] - 2026-03-11
+
+### Fixed
+
+- **Critical: `aqe init --auto` not installing agents on upgrade** — `preserveOverridesDir()` used `__dirname` (unavailable in ESM bundles), causing a TypeError that crashed the agents installer. Skills were silently installed but agents were skipped, and the phase reported 0/0. Fixed by using `import.meta.url` path resolution with fallback.
+- **Assets phase error isolation** — Agents installer errors no longer crash the entire assets phase. Skills and agents install independently, so a failure in one doesn't prevent the other from completing and reporting counts.
+
+## [3.7.17] - 2026-03-11
+
+### Added
+
+- **Adversarial review with minimum findings** — Review agents now enforce a minimum weighted finding score (CRITICAL=3, HIGH=2, MEDIUM=1, LOW=0.5). When findings fall below threshold, a second-pass review runs with halved confidence. Clean code requires explicit justification documenting files examined, patterns checked, and tools run.
+- **Agent customization overlays** — Users can customize agent behavior via `.claude/agent-overrides/*.yaml` files with replace, append, and config sections. Overlays integrate at init time (agents-installer) and runtime (task router scoring, MCP routing).
+- **Structured validation pipelines** — 13-step requirements validation with blocking gates, weighted category scoring (format, content, quality, traceability, compliance), and markdown report generation.
+- **Mechanical edge case detection** — Regex-based branch enumerator detecting 13 construct types (if-without-else, switch-no-default, try-empty-catch, optional-chaining, nullish-coalescing, promise-no-catch, and more). Available via `aqe coverage gaps <file>` CLI command.
+- **Context compilation for agents** — Parallel source gathering from memory DB, git history, test files, and coverage data with priority-based sorting and token budget management. Integrated into both QETaskRouter and MCP TaskRouterService.
+- **Micro-file step architecture** — 5 QCSD skills (production, refinement, cicd, development, ideation) split into 44 step files under 500 lines each, with compact orchestrator SKILL.md files under 300 lines.
+- **Statusline v3 helper** — New `assets/helpers/statusline-v3.cjs` for improved status display.
+
+### Changed
+
+- **QCSD skills restructured** — Large monolithic QCSD skill files (2000-2700 lines each) replaced with orchestrator + step file architecture for better maintainability.
+- **Agent definitions updated** — qe-devils-advocate, qe-gap-detector, qe-quality-gate, qe-requirements-validator, and 4 reviewer subagents enhanced with minimum finding requirements and mechanical mode support.
+- **Review skills enhanced** — brutal-honesty-review, sherlock-review, and code-review-quality skills now enforce minimum findings.
+- **Hooks phase improvements** — Updated hooks installation for devcontainer compatibility.
+- **Settings merge fix** — Corrected settings merge logic.
+- **Deprecated agents removed** — Cleaned up legacy v2 agents, deprecated subagents, consensus/hive-mind/neural/optimization/reasoning/swarm agent directories, and old slash commands.
+- **DevPod configuration updated** — Refreshed devcontainer settings for new environment.
+
 ## [3.7.16] - 2026-03-10
 
 ### Added

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

@@ -61,6 +61,26 @@ Use up to 7 concurrent strategies per review.
 - **Error Handling Gap Detection**: Find missing error handling for network failures, timeouts, malformed input, and resource exhaustion
 </capabilities>
 
+<adversarial_review_standards>
+### Minimum Finding Requirements
+Every review MUST surface findings meeting a minimum weighted score of 3.0 (severity weights: CRITICAL=3, HIGH=2, MEDIUM=1, LOW=0.5, INFORMATIONAL=0.25).
+
+### Reviewer Mindset
+You are a skeptical reviewer. Your job is to find problems. The LLM agreeability bias works against quality -- actively resist it. Assume the code has defects until proven otherwise.
+
+### Anti-Patterns (NEVER Do These)
+- NEVER produce a review that says "looks good" without at least 3 observations
+- NEVER accept "no findings" without a Clean Justification containing specific evidence
+- NEVER skip the second-pass deep review when initial findings are below minimum
+
+### Clean Justification Protocol
+If the code is genuinely clean (rare), you MUST provide:
+1. List of specific files examined
+2. List of patterns/anti-patterns checked
+3. List of tools/strategies run
+4. Detailed reasoning why no issues were found
+</adversarial_review_standards>
+
 <memory_namespace>
 Reads:
 - aqe/v3/domains/test-generation/results/* - Test generation outputs to challenge

package/assets/agents/v3/qe-gap-detector.md

@@ -54,8 +54,33 @@ Use up to 6 concurrent analyzers for large codebases.
 - **Test Recommendations**: Prioritized recommendations with effort estimates
 - **Trend Analysis**: Track gap closure over time
 - **Visual Reports**: Gap heatmaps and coverage treemaps
+- **Mechanical Edge Case Mode**: Exhaustive branch enumeration without subjective filtering (BMAD-004)
 </capabilities>
 
+<mechanical_mode>
+## Mechanical/Exhaustive Mode (BMAD-004)
+
+When invoked with `--mechanical` or `--exhaustive` flag, switch to exhaustive branch enumeration mode:
+
+### Exhaustive Mode Behavior
+- Report EVERY unhandled branch path as structured JSON without filtering by risk score
+- No subjective prioritization — purely mechanical enumeration
+- Enumerate: if-without-else, switch-no-default, empty-catch, optional-chaining null paths, promise-no-catch, array-empty-case, logical-or-falsy-trap
+- Output format: UnhandledBranch[] with file, line, column, construct type, trigger condition, current handling, suggested guard
+- Severity is deterministic (based on construct type), not subjective
+
+### Default Mode (unchanged)
+Without the mechanical flag, operate in the standard risk-scored mode with semantic analysis and prioritization.
+
+### Output Formats
+- `--json` — Structured JSON array of UnhandledBranch objects
+- `--table` — Tabular summary grouped by severity
+- `--markdown` — Detailed markdown report with code context
+
+### Implementation
+Uses `src/analysis/branch-enumerator.ts` — a regex-based pattern matcher (no AST parser dependency) that implements the `BranchEnumerator` strategy interface. Detects 13 construct types across TypeScript and JavaScript files.
+</mechanical_mode>
+
 <memory_namespace>
 Reads:
 - aqe/coverage/reports/* - Coverage data (lcov, istanbul, c8)

package/assets/agents/v3/qe-quality-gate.md

@@ -57,6 +57,14 @@ Use up to 6 concurrent evaluators for complex gates.
 - **CI/CD Integration**: Provide gate status to GitHub Actions, Jenkins, GitLab CI
 </capabilities>
 
+<pipeline_integration>
+## Pipeline Integration (BMAD-003)
+
+Quality gates can delegate structured validation to the validation pipeline framework. When evaluating requirements or documentation quality, invoke the requirements validation pipeline for systematic step-by-step assessment with gate enforcement.
+
+Validation pipeline provides: step-by-step structured verdicts, blocking gate enforcement, weighted scoring, and evidence-based reporting.
+</pipeline_integration>
+
 <memory_namespace>
 Reads:
 - aqe/quality-thresholds/* - Configured gate thresholds