@massu/core 0.5.0 → 0.6.1

package/commands/massu-golden-path.md

@@ -1,66 +1,53 @@
 ---
 name: massu-golden-path
-description: Complete end-to-end workflow from requirements to production push with minimal pause points
+description: "When user wants full autonomous implementation: 'build this', 'implement this feature', 'golden path', or provides a plan file to execute end-to-end"
 allowed-tools: Bash(*), Read(*), Write(*), Edit(*), Grep(*), Glob(*), Task(*), mcp__plugin_playwright_playwright__*, mcp__playwright__*
 ---
 name: massu-golden-path
 
-> **Shared rules apply.** Read `.claude/commands/_shared-preamble.md` before proceeding. CR-9 enforced.
+> **Shared rules apply.** Read `.claude/commands/_shared-preamble.md` before proceeding. CR-12, CR-9 enforced.
 
 # Massu Golden Path: Requirements to Production Push
 
 ## Objective
 
 Execute the COMPLETE development workflow in one continuous run:
-**Requirements --> Plan Creation --> Plan Audit --> Implementation --> Browser Verification --> Simplification --> Commit --> Push**
+**Requirements -> Plan Creation -> Plan Audit -> Implementation -> Gap Analysis -> Simplification -> Security Audit -> Commit -> Push**
 
 This command has FULL FEATURE PARITY with the individual commands it replaces:
-`/massu-create-plan` --> `/massu-plan` --> `/massu-loop` --> `/massu-loop-playwright` --> `/massu-simplify` --> `/massu-commit` --> `/massu-push`
+`/massu-create-plan` -> `/massu-plan` -> `/massu-loop` -> `/massu-loop-playwright` -> `/massu-simplify` -> `/massu-security` -> `/massu-commit` -> `/massu-push`
 
 ---
 
 ## NON-NEGOTIABLE RULES
 
-- **Complete workflow** -- ALL phases must execute, no skipping
+- **Complete workflow (CR-11)** -- ALL phases must execute, no skipping. 100% plan coverage required
 - **Zero failures** -- Each phase gate must pass before proceeding
-- **Proof required** -- Show output of each phase gate
+- **Proof required (CR-1)** -- VR-* output pasted, not summarized. "I verified" without output = invalid
 - **FIX ALL ISSUES ENCOUNTERED (CR-9)** -- Whether from current changes or pre-existing
 - **MEMORY IS MANDATORY (CR-38)** -- Persist ALL learnings before session ends
+- **Stagnation bail-out (CR-37)** -- If same item fails 3+ times, replan instead of grinding
 
 ---
 
-## APPROVAL POINTS (Max 4 Pauses)
+## APPROVAL POINTS (Max 4 Pauses, 5 with --competitive)
 
 ```
 +-----------------------------------------------------------------------------+
 |   THIS COMMAND RUNS STRAIGHT THROUGH THE ENTIRE GOLDEN PATH.                |
-|   IT ONLY PAUSES FOR THESE APPROVAL POINTS:                                |
-|                                                                             |
-|   1. PLAN APPROVAL - After plan creation + audit (user reviews plan)        |
-|   2. NEW PATTERN APPROVAL - If a new pattern is needed (during any phase)   |
-|   3. COMMIT APPROVAL - Before creating the commit                           |
-|   4. PUSH APPROVAL - Before pushing to remote                               |
-|                                                                             |
+|   IT ONLY PAUSES FOR THESE APPROVAL POINTS:                                 |
+|                                                                              |
+|   1. PLAN APPROVAL - After plan creation + audit (user reviews plan)         |
+|   2. NEW PATTERN APPROVAL - If a new pattern is needed (during any phase)    |
+|   3. COMMIT APPROVAL - Before creating the commit                            |
+|   4. PUSH APPROVAL - Before pushing to remote                                |
+|   5. WINNER SELECTION - After competitive scoring (--competitive only)       |
+|                                                                              |
 |   EVERYTHING ELSE RUNS AUTOMATICALLY WITHOUT STOPPING.                      |
 +-----------------------------------------------------------------------------+
 ```
 
-### Approval Point Format
-
-```
-===============================================================================
-APPROVAL REQUIRED: [TYPE]
-===============================================================================
-
-[Details]
-
-OPTIONS:
-  - "approve" / "yes" to continue
-  - "modify" to request changes
-  - "abort" to stop the golden path
-
-===============================================================================
-```
+Read `references/approval-points.md` for the exact format and options for each approval point.
 
 After receiving approval, immediately continue. Do NOT ask "shall I continue?" -- just proceed.
 
@@ -73,888 +60,175 @@ After receiving approval, immediately continue. Do NOT ask "shall I continue?" -
 | **Task Description** | `/massu-golden-path "Implement feature X"` | Full flow from Phase 0 |
 | **Plan File** | `/massu-golden-path /path/to/plan.md` | Skip to Phase 1C (audit) |
 | **Continue** | `/massu-golden-path "Continue [feature]"` | Resume from session state |
+| **Competitive** | `/massu-golden-path --competitive "task"` | Spawn 2-3 competing implementations with bias presets, score, select winner |
+| **Competitive (3 agents)** | `/massu-golden-path --competitive --agents 3 "task"` | 3 agents with quality/ux/robust biases (default: 2 agents = quality + robust) |
+| **External Loop** | `/massu-golden-path --external /path/to/plan.md` | Phase 2 uses `scripts/loop-external.sh` for context-fresh iterations |
 
 ---
 
-## PHASE 0: REQUIREMENTS & CONTEXT LOADING
-
-### 0.1 Session Context Loading
-
-```
-[GOLDEN PATH -- PHASE 0: REQUIREMENTS & CONTEXT]
-```
+## PHASE OVERVIEW
 
-- Read `session-state/CURRENT.md` for any prior state
-- Read `massu.config.yaml` for project configuration
-- Search memory files for relevant prior context
-
-### 0.2 Requirements Coverage Map
-
-Initialize ALL dimensions as `pending`:
-
-| # | Dimension | Status | Resolved By |
-|---|-----------|--------|-------------|
-| D1 | Problem & Scope | pending | User request + interview |
-| D2 | Users & Personas | pending | Interview |
-| D3 | Data Model | pending | Phase 1A (Config/Schema Reality Check) |
-| D4 | Backend / API | pending | Phase 1A (Codebase Reality Check) |
-| D5 | Frontend / UX | pending | Interview + Phase 1A |
-| D6 | Auth & Permissions | pending | Phase 1A (Security Pre-Screen) |
-| D7 | Error Handling | pending | Phase 1A (Pattern Compliance) |
-| D8 | Security | pending | Phase 1A (Security Pre-Screen) |
-| D9 | Edge Cases | pending | Phase 1A (Question Filtering) |
-| D10 | Performance | pending | Phase 1A (Pattern Compliance) |
-
-### 0.3 Ambiguity Detection (7 Signals)
-
-| Signal | Description |
-|--------|-------------|
-| A1 | Vague scope -- no clear boundary |
-| A2 | No success criteria -- no measurable outcome |
-| A3 | Implicit requirements -- unstated but necessary |
-| A4 | Multi-domain -- spans 3+ domains |
-| A5 | Contradictions -- conflicting constraints |
-| A6 | No persona -- unclear who benefits |
-| A7 | New integration -- external service not yet in codebase |
-
-**Score >= 2**: Enter interview loop (0.4). **Score 0-1**: Fast-track to Phase 1A.
-
-### 0.4 Interview Loop (When Triggered)
-
-Ask via AskUserQuestion, one question at a time:
-1. Show compact coverage status: `Coverage: D1:done D2:pending ...`
-2. Provide 2-4 curated options (never open-ended)
-3. Push back on contradictions and over-engineering
-4. Self-terminate when D1, D2, D5 covered
-5. Escape hatch: user says "skip" / "enough" / "just do it" --> mark remaining as `n/a`
+| Phase | Name | Key Actions | Approval Gate |
+|-------|------|-------------|---------------|
+| 0 | Requirements & Context | Session context, ambiguity detection, interview loop | -- |
+| 1 | Plan Creation & Audit | Research, plan generation, audit loop | PLAN APPROVAL |
+| 2 | Implementation | Item loop, multi-perspective review, verification audit, browser testing | NEW PATTERN (if needed) |
+| 2-COMP | Competitive Implementation | Spawn N agents with bias presets, score, select winner (`--competitive` only) | WINNER SELECTION |
+| 2.5 | Gap & Enhancement Analysis | Find+fix gaps, UX issues, security, pattern compliance; loop until zero | -- |
+| 3 | Simplification | Pattern scanner, parallel semantic review, apply findings | -- |
+| 3.5 | Deep Security Audit | Full adversarial audit loop with parallel red-team agents, iterate to zero findings | -- |
+| 4 | Pre-Commit Verification | Verification gates, quality scoring | COMMIT APPROVAL |
+| 5 | Push Verification | `scripts/push-verify.sh`, CI monitoring via `scripts/ci-status.sh` | PUSH APPROVAL |
+| 6 | Completion | Final report, plan update, auto-learning, feature registration | -- |
 
 ---
 
-## PHASE 1: PLAN CREATION & AUDIT
-
-### Phase 1A: Research & Reality Check
-
-```
-[GOLDEN PATH -- PHASE 1A: RESEARCH & REALITY CHECK]
-```
-
-**If plan file was provided**: Skip to Phase 1C.
-
-#### 1A.1 Feature Understanding
-
-- Document: exact user request, feature type, affected domains
-- Search codebase for similar features, tool modules, existing patterns
-- Read `massu.config.yaml` for relevant config sections
-
-#### 1A.2 Config & Schema Reality Check
-
-For features touching config or databases:
-
-- Parse `massu.config.yaml` and verify all referenced config keys exist
-- Check SQLite schema for affected tables (`getCodeGraphDb`, `getDataDb`, `getMemoryDb`)
-- Verify tool definitions in `tools.ts` for any tools being modified
-
-Document: existing config keys, required new keys, required schema changes.
-
-#### 1A.3 Config-Code Alignment (VR-CONFIG)
-
-If feature uses config-driven values:
-
-```bash
-# Check config keys used in code
-grep -rn "getConfig()" packages/core/src/ | grep -oP 'config\.\w+' | sort -u
-# Compare to massu.config.yaml structure
-```
-
-#### 1A.4 Codebase Reality Check
-
-- Verify target directories/files exist
-- Read similar tool modules and handlers
-- Load relevant pattern files (build/testing/security/database/mcp)
-
-#### 1A.5 Blast Radius Analysis (CR-10)
-
-**MANDATORY when plan changes any constant, export name, config key, or tool name.**
-
-1. Identify ALL changed values (old --> new)
-2. Codebase-wide grep for EACH value
-3. If plan deletes files: verify no remaining imports or references
-4. Categorize EVERY occurrence: CHANGE / KEEP (with reason) / INVESTIGATE
-5. Resolve ALL INVESTIGATE to 0. Add ALL CHANGE items as plan deliverables.
-
-#### 1A.6 Pattern Compliance Check
-
-Check applicable patterns: ESM imports (.ts extensions), config access (getConfig()), tool registration (3-function pattern), hook compilation (esbuild), SQLite DB access (getCodeGraphDb/getDataDb/getMemoryDb), memDb lifecycle (try/finally close).
-
-Read most similar tool module for patterns used.
-
-#### 1A.7 Tool Registration Check (CR-11)
-
-For EVERY new MCP tool planned -- verify a corresponding registration item exists in the plan (definitions + routing + handler in `tools.ts`). If NOT, ADD IT.
-
-#### 1A.8 Question Filtering
-
-1. List all open questions
-2. Self-answer anything answerable by reading code or config
-3. Surface only business logic / UX / scope / priority questions to user via AskUserQuestion
-4. If all self-answerable, skip user prompt
-
-#### 1A.9 Security Pre-Screen (5 Dimensions)
-
-| Dim | Check | If Triggered |
-|-----|-------|-------------|
-| S1 | PII / Sensitive Data | Add access controls |
-| S2 | Authentication | Verify auth checks |
-| S3 | Authorization | Add permission checks |
-| S4 | Injection Surfaces | Add input validation, parameterized queries |
-| S5 | Rate Limiting | Add rate limiting considerations |
-
-**BLOCKS_REMAINING must = 0 before proceeding.**
-
-Mark all coverage dimensions as `done` or `n/a`.
-
-### Phase 1B: Plan Generation
-
-```
-[GOLDEN PATH -- PHASE 1B: PLAN GENERATION]
-```
-
-Write plan to: `docs/plans/[YYYY-MM-DD]-[feature-name].md`
-
-**Plan structure** (P-XXX numbered items):
-- Overview (feature, complexity, domains, item count)
-- Requirements Coverage Map (D1-D10 all resolved)
-- Phase 1: Configuration Changes (massu.config.yaml)
-- Phase 2: Backend Implementation (tool modules, handlers, SQLite schema)
-- Phase 3: Frontend/Hook Implementation (hooks, plugin code)
-- Phase 4: Testing & Verification
-- Phase 5: Documentation
-- Verification Commands table
-- Item Summary table
-- Risk Assessment
-- Dependencies
-
-**Item numbering**: P1-XXX (config), P2-XXX (backend), P3-XXX (frontend/hooks), P4-XXX (testing), P5-XXX (docs).
+## PHASE 0: REQUIREMENTS & CONTEXT LOADING
 
-**Implementation Specificity Check**: Every item MUST have exact file path, exact content, insertion point, format matches target, verification command.
+Read `references/phase-0-requirements.md` for full details.
 
-### Phase 1C: Plan Audit Loop
+**Summary**: Load session context via memory tools. Build a 10-dimension requirements coverage map (D1-D10). Run ambiguity detection (7 signals). If ambiguity score >= 2, enter interview loop. Fast-track to Phase 1 when D1, D2, D5 covered or user says "skip" / "just do it".
 
-```
-[GOLDEN PATH -- PHASE 1C: PLAN AUDIT LOOP]
-```
-
-Run audit loop using subagent architecture (prevents early termination):
+---
 
-```
-iteration = 0
-WHILE true:
-  iteration += 1
-
-  result = Task(subagent_type="massu-plan-auditor", model="opus", prompt="
-    Audit iteration {iteration} for plan: {PLAN_PATH}
-    Execute ONE complete audit pass. Verify ALL deliverables.
-    Check: VR-PLAN-FEASIBILITY, VR-PLAN-SPECIFICITY, Pattern Alignment, Config Reality.
-    Fix any plan document gaps you find.
-
-    CRITICAL: Report GAPS_DISCOVERED as total gaps FOUND, EVEN IF you fixed them.
-    Finding N gaps and fixing all N = GAPS_DISCOVERED: N.
-    A clean pass finding nothing = GAPS_DISCOVERED: 0.
-  ")
-
-  gaps = parse GAPS_DISCOVERED from result
-  IF gaps == 0: BREAK (clean pass)
-  ELSE: CONTINUE (re-audit)
-
-  IF iteration >= 10: Report to user, ask how to proceed
-END WHILE
-```
+## PHASE 1: PLAN CREATION & AUDIT
 
-**VR-PLAN-FEASIBILITY**: Files exist, config keys valid, dependencies available, patterns documented.
-**VR-PLAN-SPECIFICITY**: Every item has exact path, exact content, insertion point, verification command.
-**Pattern Alignment**: Cross-reference ALL applicable patterns from CLAUDE.md and patterns/*.md.
+Read `references/phase-1-plan-creation.md` for full details.
 
-### Phase 1 Complete --> APPROVAL POINT #1: PLAN
+**Summary**: Three sub-phases:
+- **1A: Research & Reality Check** -- Feature understanding, codebase check, blast radius analysis (CR-25), pattern compliance, backend-frontend coupling (CR-12), question filtering, security pre-screen (6 dimensions).
+- **1B: Plan Generation** -- Write plan to `docs/plans/[YYYY-MM-DD]-[feature-name].md` with P-XXX numbered items across 6 phases.
+- **1C: Plan Audit Loop** -- Subagent architecture. Iterate until GAPS_DISCOVERED = 0. Max 10 iterations.
 
-```
-===============================================================================
-APPROVAL REQUIRED: PLAN
-===============================================================================
-
-Plan created and audited ({iteration} audit passes, 0 gaps).
-
-PLAN SUMMARY:
--------------------------------------------------------------------------------
-Feature: [name]
-File: [plan path]
-Total Items: [N]
-Phases: [list]
-
-Requirements Coverage: [X]/10 dimensions resolved
-Feasibility: VERIFIED (config, files, patterns, security)
-Audit Passes: {iteration} (final pass: 0 gaps)
--------------------------------------------------------------------------------
-
-OPTIONS:
-  - "approve" to begin implementation
-  - "modify: [changes]" to adjust plan
-  - "abort" to stop
-
-===============================================================================
-```
+**Gate**: APPROVAL POINT #1: PLAN
 
 ---
 
 ## PHASE 2: IMPLEMENTATION
 
-### Phase 2A: Plan Item Extraction & Setup
-
-```
-[GOLDEN PATH -- PHASE 2: IMPLEMENTATION]
-```
-
-1. Read plan from disk (NOT memory -- CR-5)
-2. Extract ALL deliverables into tracking table:
-
-| Item # | Type | Description | Location | Verification | Status |
-|--------|------|-------------|----------|--------------|--------|
-| P1-001 | CONFIG | ... | ... | VR-CONFIG | PENDING |
-
-3. Create VR-PLAN verification strategy:
-
-| # | VR-* Check | Target | Why Applicable | Status |
-|---|-----------|--------|----------------|--------|
-| 1 | VR-BUILD | Full project | Always | PENDING |
-
-4. Initialize session state with AUTHORIZED_COMMAND: massu-golden-path
-
-### Phase 2B: Implementation Loop
-
-For each plan item:
-1. **Pre-check**: Verify file exists, read current state
-2. **Execute**: Implement the item following established patterns
-3. **Guardrail**: Run `bash scripts/massu-pattern-scanner.sh` (ABORT if fails)
-4. **Verify**: Run applicable VR-* checks with proof
-5. **Update**: Mark item complete in tracking table
-
-**DO NOT STOP between items** unless:
-- New pattern needed (Approval Point #2)
-- True blocker (external service, credentials)
-- Critical error after 3 retries
-
-**Checkpoint Audit at phase boundaries** (after all P1-XXX, after all P2-XXX, etc.):
-
-```
-CHECKPOINT:
-[1] READ plan section          [2] GREP tool registrations    [3] LS modules
-[4] VR-CONFIG check            [5] VR-TOOL-REG check          [6] VR-HOOK-BUILD check
-[7] Pattern scanner            [8] npm run build               [9] cd packages/core && npx tsc --noEmit
-[10] npm test                  [11] VR-GENERIC check           [12] Security scanner
-[13] COUNT gaps --> IF > 0: FIX and return to [1]
-```
-
-### Phase 2C: Multi-Perspective Review
-
-After implementation, BEFORE verification loop -- spawn 3 review agents **IN PARALLEL**:
-
-```
-security_result = Task(subagent_type="massu-security-reviewer", model="opus", prompt="
-  Review implementation for plan: {PLAN_PATH}
-  Focus: Security vulnerabilities, auth gaps, input validation, data exposure.
-  Return structured result with SECURITY_GATE: PASS/FAIL.
-")
-
-architecture_result = Task(subagent_type="massu-architecture-reviewer", model="opus", prompt="
-  Review implementation for plan: {PLAN_PATH}
-  Focus: Design issues, coupling, pattern compliance, scalability.
-  Return structured result with ARCHITECTURE_GATE: PASS/FAIL.
-")
-
-quality_result = Task(subagent_type="massu-quality-reviewer", model="sonnet", prompt="
-  Review implementation for plan: {PLAN_PATH}
-  Focus: Code quality, ESM compliance, config-driven patterns, TypeScript strict mode, test coverage.
-  Return structured result with QUALITY_GATE: PASS/FAIL.
-")
-```
-
-Fix ALL CRITICAL/HIGH findings before proceeding. WARN findings = document and proceed.
-
-### Phase 2D: Verification Audit Loop
-
-```
-iteration = 0
-WHILE true:
-  iteration += 1
-
-  # Circuit breaker (CR-37)
-  IF iteration >= 3 AND same gaps as previous iteration:
-    AskUserQuestion: "Loop stalled after {iteration} passes. Re-plan / Continue / Stop?"
-
-  result = Task(subagent_type="massu-plan-auditor", model="opus", prompt="
-    Audit iteration {iteration} for plan: {PLAN_PATH}
-    Verify ALL deliverables with VR-* proof.
-    Check code quality (patterns, build, types, tests).
-    Check plan coverage (every item verified).
-    Fix any gaps you find.
-
-    CRITICAL: GAPS_DISCOVERED = total FOUND, even if fixed.
-    Finding 5 + fixing 5 = GAPS_DISCOVERED: 5 (NOT 0).
-  ")
-
-  gaps = parse GAPS_DISCOVERED from result
-  Output: "Verification iteration {iteration}: {gaps} gaps"
-
-  IF gaps == 0: BREAK
-  IF iteration >= 10: Report remaining gaps, ask user
-END WHILE
-```
-
-### Phase 2E: Post-Build Reflection + Memory Persist (CR-38)
-
-**MANDATORY -- reflection + memory write = ONE atomic action.**
-
-Answer these questions:
-1. "Now that I've built this, what would I have done differently?"
-2. "What should be refactored before moving on?"
-3. "Did we over-build? Is there a simpler way?"
-4. "Would a staff engineer approve this?" (Core Principle #9)
-
-**IMMEDIATELY write ALL learnings to memory/ files** -- failed approaches, new patterns, tool gotchas, architectural insights. DO NOT output reflections as text without writing to memory.
-
-Apply any low-risk refactors immediately. Log remaining suggestions in plan under `## Post-Build Reflection`.
-
-### Phase 2F: Documentation Sync (User-Facing Features)
-
-If plan includes ANY user-facing features (new MCP tools, config changes, hook changes):
-
-1. Update relevant documentation (README, API docs, config docs)
-2. Ensure tool descriptions match implementation
-3. Update config schema documentation if config keys changed
-
-Skip ONLY if purely internal refactoring with zero user-facing changes.
-
-### Phase 2G: Browser Verification & Fix Loop (`/massu-loop-playwright`)
-
-```
-[GOLDEN PATH -- PHASE 2G: BROWSER VERIFICATION]
-```
-
-**This phase executes the full `/massu-loop-playwright` protocol inline.** See `massu-loop-playwright.md` for the standalone version.
-
-**Auto-trigger condition**: If plan touches ANY UI/demo files or produces visual output, this phase runs automatically. If purely backend/MCP/config with zero visual output, skip with log note: `Browser verification: SKIPPED (no UI files changed)`.
-
-#### 2G.1 Determine Target Pages
-
-Map changed features to testable URLs:
-- If Massu has a demo page or documentation site: test affected pages
-- If testing MCP tool output: use a test harness or verify tool responses
-- Component changes: identify ALL pages that render the component
-
-#### 2G.2 Browser Setup & Authentication
-
-Use Playwright MCP plugin tools (`mcp__plugin_playwright_playwright__*`). Fallback: `mcp__playwright__*`.
-
-1. `browser_navigate` to target URL ($TARGET_URL)
-2. `browser_snapshot` to check page status
-3. If authentication required: STOP and request manual login
-
-```
-AUTHENTICATION REQUIRED
-
-The Playwright browser is not logged in to the target application.
-Please log in manually in the open browser window, then re-run the golden path.
-```
-
-**NEVER type credentials. NEVER hardcode passwords. NEVER proceed without authentication.**
-
-#### 2G.3 Load Audit (Per Page)
+Read `references/phase-2-implementation.md` for full details.
 
-For EACH target page:
+**Summary**: Nine sub-phases (or external loop via `--external` flag using `scripts/loop-external.sh` for context-fresh iterations):
+- **2A**: Extract plan items into tracking table, initialize session state
+- **2A.5**: Sprint contracts -- negotiate definition-of-done per plan item before implementation (scope boundary, acceptance criteria, VR-* mapping). See `references/sprint-contract-protocol.md`
+- **2B**: Implementation loop (pre-check, execute, guardrail, verify, update per item)
+- **2C**: Multi-perspective review (3 parallel agents: security, architecture, UX) + **QA evaluator** (conditional, UI plans only -- adversarial Playwright-based acceptance testing against sprint contracts). See `references/qa-evaluator-spec.md`
+- **2D**: Verification audit loop (subagent, circuit breaker CR-37, refine-or-pivot at 3+ iterations, sprint contract verification, max 10 iterations)
+- **2E**: Post-build reflection + memory persist (CR-38)
+- **2F**: Documentation sync (if user-facing features)
+- **2G**: Browser verification & fix loop (auto-triggers if UI files changed, Playwright MCP)
 
-| Check | Tool | Captures |
-|-------|------|----------|
-| Console errors/warnings | `browser_console_messages` | React errors, TypeError, CSP violations |
-| Network failures | `browser_network_requests` | 500s, 404s, CORS failures, timeouts |
-
-Categorize findings:
-
-| Category | Severity |
-|----------|----------|
-| Crash, 500 error, data exposure | **P0 -- CRITICAL** |
-| Network failure, broken interaction | **P1 -- HIGH** |
-| Visual issues, performance warnings | **P2 -- MEDIUM** |
-| Console warnings, deprecations | **P3 -- LOW** |
-
-#### 2G.4 Interactive Testing (Per Page)
-
-1. `browser_snapshot` --> inventory ALL interactive elements (buttons, links, forms, selects, tabs, modals, data tables)
-2. For EACH testable element:
-   - Capture console state BEFORE interaction (`browser_console_messages`)
-   - Perform interaction (`browser_click`, `browser_select_option`, `browser_fill_form`)
-   - Wait 2-3 seconds for async operations
-   - Capture console state AFTER interaction
-   - Record any NEW errors introduced
-   - `browser_snapshot` to verify DOM state after interaction
-   - If interaction opened modal/sheet: test elements inside, then close
-
-**SAFETY**: Never submit forms, click Delete/Send/Submit, or create real records on production.
-
-#### 2G.5 Visual & Performance Audit
-
-**Visual checks**:
-- Broken images: `browser_evaluate` to find `img` elements with `naturalWidth === 0`
-- Layout issues: overflow, overlapping, missing content, broken alignment
-- Responsive: `browser_resize` at 1440x900 (desktop), 768x1024 (tablet), 375x812 (mobile)
-- Screenshot evidence: `browser_take_screenshot` at each breakpoint if issues found
-
-**Performance checks**:
-- Page load timing via `browser_evaluate` (`performance.getEntriesByType('navigation')`)
-- Resources > 500KB via `browser_evaluate` (`performance.getEntriesByType('resource')`)
-- Slow API calls > 3s, duplicate requests via `browser_network_requests`
-
-| Metric | Good | Needs Work | Critical |
-|--------|------|------------|----------|
-| DOM Content Loaded | < 2s | 2-5s | > 5s |
-| Full Load | < 4s | 4-8s | > 8s |
-| TTFB | < 500ms | 500ms-1.5s | > 1.5s |
-
-#### 2G.6 Fix Loop
-
-```
-issues = ALL findings from 2G.3-2G.5, sorted by priority (P0 first)
-
-FOR EACH issue WHERE priority <= P2:
-  1. IDENTIFY root cause (Grep/Read source files)
-  2. APPLY fix (follow CLAUDE.md patterns)
-  3. VERIFY fix (VR-GREP, VR-NEGATIVE, VR-BUILD, VR-TYPE)
-  4. LOG fix in report
-
-Zero-issue standard: ALL P0/P1 fixed, ALL P2 fixed or documented with justification.
-Circuit breaker: 5 iterations on same page --> ask user.
-```
-
-Post-fix: reload target URLs, re-run load audit + interactive testing for elements that had failures. If new errors appear, add to issues list and continue fix loop.
-
-#### 2G.7 Report
-
-Save to `.claude/playwright-reports/{TIMESTAMP}-{SLUG}.md`.
-
-Report includes: summary table, console errors, network failures, interactive element failures, visual issues, performance issues, fix log with files changed and VR checks, unfixed issues with justification, screenshots.
-
-#### 2G.8 Auto-Learning Protocol
-
-For EACH browser-discovered fix:
-1. Update memory files with symptom/root cause/fix/files
-2. Add to `scripts/massu-pattern-scanner.sh` if the bad pattern is grep-able
-3. Codebase-wide search for same bad pattern (CR-9) -- fix ALL instances
-
-```
-[GOLDEN PATH -- PHASE 2 COMPLETE]
-- All plan items implemented
-- Multi-perspective review: PASSED (security, architecture, quality)
-- Verification audit: PASSED (Loop #{iteration}, 0 gaps)
-- Post-build reflection: PERSISTED to memory
-- Documentation sync: COMPLETE / N/A
-- Browser verification: PASSED ({N} pages tested, {M} issues fixed) / SKIPPED (no UI files)
-```
+**Gate**: APPROVAL POINT #2: NEW PATTERN (if needed, any sub-phase)
 
 ---
 
-## PHASE 3: SIMPLIFICATION (`/massu-simplify`)
+## PHASE 2.5: GAP & ENHANCEMENT ANALYSIS
 
-```
-[GOLDEN PATH -- PHASE 3: SIMPLIFICATION]
-```
-
-**This phase executes the full `/massu-simplify` protocol inline.** See `massu-simplify.md` for the standalone version.
+Read `references/phase-2.5-gap-analyzer.md` for full details.
 
-### 3.1 Fast Gate
+**Summary**: After implementation completes, run a continuous gap and enhancement analysis loop. A subagent analyzes all changed files across 7 categories (functional gaps, UX gaps, data integrity, security, pattern compliance, enhancements, sprint contract compliance). VR-VISUAL uses weighted 4-dimension scoring (threshold >= 3.0). Every gap/enhancement found is fixed immediately. The loop re-runs until a full pass discovers ZERO gaps. Max 10 iterations. Skippable only for documentation-only changes or explicit user request.
 
-```bash
-bash scripts/massu-pattern-scanner.sh  # Fix ALL violations before semantic analysis
-```
+---
 
-### 3.2 Parallel Semantic Review (3 Agents)
+## PHASE 3: SIMPLIFICATION
 
-Spawn IN PARALLEL (Core Principle #10 -- one task per agent):
+Read `references/phase-3-simplify.md` for full details.
 
-**Efficiency Reviewer** (haiku): Query inefficiency (findMany equivalent vs SQL COUNT, N+1 queries, unbounded queries), algorithmic inefficiency (O(n^2), repeated sort/filter), unnecessary allocations, missing caching opportunities.
+**Summary**: Fast gate (pattern scanner), then 3 parallel semantic review agents (efficiency, reuse, pattern compliance). Apply ALL findings sorted by severity. Re-run pattern scanner.
 
-**Reuse Reviewer** (haiku): Known utilities (getConfig(), stripPrefix(), tool registration patterns, memDb lifecycle pattern), module duplication against existing tool modules, pattern duplication across new files, config values that should be in massu.config.yaml.
+---
 
-**Pattern Compliance Reviewer** (haiku): ESM compliance (.ts import extensions, no require()), config-driven patterns (no hardcoded project-specific values -- CR-38/VR-GENERIC), TypeScript strict mode compliance, tool registration (3-function pattern preferred -- CR-11), hook compilation (esbuild compatible -- CR-12), memDb lifecycle (try/finally close), security (input validation, no eval/exec).
+## PHASE 3.5: DEEP SECURITY AUDIT
 
-### 3.3 Apply ALL Findings
+Read `references/phase-3.5-security-audit.md` for full details.
 
-Sort by SEVERITY (CRITICAL --> LOW). Fix ALL (CR-9). Re-run pattern scanner.
-
-```
-SIMPLIFY_GATE: PASS (N findings, N fixed, 0 remaining)
-```
+**Summary**: Run a full adversarial security audit loop against ALL changed files. Launches 2-4 parallel red-team agents (injection, network/leakage, DoS, bypass) per iteration. Every finding is fixed in-place. Loop iterates until zero findings remain (max 5 iterations). This phase is NEVER skipped -- security is non-negotiable. Security fixes flow into Phase 4's verification gates automatically.
 
 ---
 
 ## PHASE 4: PRE-COMMIT VERIFICATION
 
-```
-[GOLDEN PATH -- PHASE 4: PRE-COMMIT VERIFICATION]
-```
-
-### 4.1 Auto-Verification Gates (ALL must pass in SINGLE run)
-
-| Gate | Command | Expected |
-|------|---------|----------|
-| 1. Pattern Scanner | `bash scripts/massu-pattern-scanner.sh` | Exit 0 |
-| 2. Type Safety (VR-TYPE) | `cd packages/core && npx tsc --noEmit` | 0 errors |
-| 3. Build (VR-BUILD) | `npm run build` | Exit 0 |
-| 4. Tests (VR-TEST) | `npm test` | ALL pass |
-| 5. Hook Compilation (VR-HOOK-BUILD) | `cd packages/core && npm run build:hooks` | Exit 0 |
-| 6. Generalization (VR-GENERIC) | `bash scripts/massu-generalization-scanner.sh` | Exit 0 |
-| 7. Security Scanner | `bash scripts/massu-security-scanner.sh` | Exit 0 |
-| 8. Secrets Staged | `git diff --cached --name-only \| grep -E '\.(env\|pem\|key\|secret)'` | 0 files |
-| 9. Credentials in Code | `grep -rn "sk-\|password.*=.*['\"]" --include="*.ts" packages/ \| grep -v "process.env" \| wc -l` | 0 |
-| 10. VR-TOOL-REG | For EACH new tool: verify definitions + handler wired in tools.ts | All wired |
-| 11. Plan Coverage | Verify ALL plan items with VR-* proof | 100% |
-| 12. VR-PLAN-STATUS | `grep "IMPLEMENTATION STATUS" [plan]` | Match |
-| 13. Dependency Security | `npm audit --audit-level=high` | 0 high/crit |
-
-### 4.2 Quality Scoring Gate
-
-Spawn `massu-output-scorer` (sonnet): Code Clarity, Pattern Compliance, Error Handling, Test Coverage, Config-Driven Design (1-5 each). All >= 3: PASS. Any < 3: FAIL.
-
-### 4.3 If ANY Gate Fails
-
-**DO NOT PAUSE** -- Fix automatically, re-run ALL gates, repeat until all pass.
-
-### 4.4 Auto-Learning Protocol
+Read `references/phase-4-commit.md` for full details.
 
-- For each bug fixed: update memory files
-- For new patterns: record in memory
-- Add detection to `scripts/massu-pattern-scanner.sh` if grep-able
-- Codebase-wide search: no other instances of same bad pattern (CR-9)
-- Record user corrections to `memory/corrections.md`
+**Summary**: Verification gates (pattern scanner, tsc, build, lint, secrets, VR-RENDER, VR-COUPLING, plan coverage, plan status, dep security). Quality scoring gate. Auto-fix on failure.
 
-### Phase 4 Complete --> APPROVAL POINT #3: COMMIT
-
-```
-===============================================================================
-APPROVAL REQUIRED: COMMIT
-===============================================================================
-
-All verification checks passed. Ready to commit.
-
-VERIFICATION RESULTS:
--------------------------------------------------------------------------------
-- Pattern scanner: Exit 0
-- Type check: 0 errors
-- Build: Exit 0
-- Tests: ALL pass
-- Hook compilation: Exit 0
-- Generalization: Exit 0
-- Security: No secrets staged, no credentials in code
-- Tool registration: All new tools wired
-- Plan Coverage: [X]/[X] = 100%
-- Quality Score: [X.X]/5.0
--------------------------------------------------------------------------------
-
-FILES TO BE COMMITTED:
-[list]
-
-PROPOSED COMMIT MESSAGE:
--------------------------------------------------------------------------------
-[type]: [description]
-
-[body]
-
-Co-Authored-By: Claude <noreply@anthropic.com>
--------------------------------------------------------------------------------
-
-OPTIONS:
-  - "approve" to commit and continue to push
-  - "message: [new message]" to change commit message
-  - "abort" to stop (changes remain staged)
-
-===============================================================================
-```
-
-### Commit Format
-
-```bash
-git commit -m "$(cat <<'EOF'
-[type]: [description]
-
-[Body]
-
-Changes:
-- [Change 1]
-- [Change 2]
-
-Verified:
-- Pattern scanner: PASS | Type check: 0 errors | Build: PASS
-- Tests: ALL pass | Hooks: compiled | Generalization: PASS
-
-Co-Authored-By: Claude <noreply@anthropic.com>
-EOF
-)"
-```
+**Gate**: APPROVAL POINT #3: COMMIT
 
 ---
 
 ## PHASE 5: PUSH VERIFICATION & PUSH
 
-```
-[GOLDEN PATH -- PHASE 5: PUSH VERIFICATION]
-```
-
-### 5.1 Pre-Flight
+Read `references/phase-5-push.md` for full details.
 
-```bash
-git log origin/main..HEAD --oneline  # Commits to push
-```
-
-### 5.2 Tier 1: Quick Re-Verification
-
-Run in parallel where possible:
-
-| Check | Command |
-|-------|---------|
-| Pattern Scanner | `bash scripts/massu-pattern-scanner.sh` |
-| Generalization | `bash scripts/massu-generalization-scanner.sh` |
-| TypeScript | `cd packages/core && npx tsc --noEmit` |
-| Build | `npm run build` |
-| Hook Compilation | `cd packages/core && npm run build:hooks` |
-
-### 5.3 Tier 2: Test Suite (CRITICAL)
-
-#### 5.3.0 Regression Detection (MANDATORY FIRST)
-
-```bash
-# Establish baseline on main
-git stash && git checkout main -q
-npm test 2>&1 | tee /tmp/baseline-tests.txt
-git checkout - -q && git stash pop -q
-
-# Run on current branch
-npm test 2>&1 | tee /tmp/current-tests.txt
-
-# Compare: any test passing on main but failing now = REGRESSION
-# Regressions MUST be fixed before push
-```
-
-#### 5.3.1-5.3.3 Test Execution
-
-Use **parallel Task agents** for independent checks:
-
-```
-Agent Group A (parallel):
-- Agent 1: npm test (unit tests)
-- Agent 2: npm audit --audit-level=high
-- Agent 3: bash scripts/massu-security-scanner.sh
-
-Sequential:
-- VR-TOOL-REG: verify ALL new tools registered in tools.ts
-- VR-GENERIC: verify ALL files pass generalization scanner
-```
-
-### 5.4 Tier 3: Security & Compliance
-
-| Check | Command |
-|-------|---------|
-| npm audit | `npm audit --audit-level=high` |
-| Security scan | `bash scripts/massu-security-scanner.sh` |
-| Config validation | Parse massu.config.yaml without errors |
-
-### 5.5 Tier 4: Final Gate
-
-All tiers must pass:
-
-| Tier | Status |
-|------|--------|
-| Tier 1: Quick Checks | PASS/FAIL |
-| Tier 2: Test Suite + Regression | PASS/FAIL |
-| Tier 3: Security & Compliance | PASS/FAIL |
-
-### Phase 5 Gate --> APPROVAL POINT #4: PUSH
+**Summary**: Pre-flight (commits to push). Tier 1: quick re-verification. Tier 2: test suite with mandatory regression detection. Tier 3: security & compliance (npm audit, secrets scan). Tier 4: final gate.
 
-```
-===============================================================================
-APPROVAL REQUIRED: PUSH TO REMOTE
-===============================================================================
-
-All verification tiers passed. Ready to push.
-
-PUSH GATE SUMMARY:
--------------------------------------------------------------------------------
-Commit: [hash]
-Message: [message]
-Files changed: [N] | +[N] / -[N]
-Branch: [branch] --> origin
-
-Tier 1 (Quick): PASS
-Tier 2 (Tests): PASS -- Unit: X/X, Regression: 0
-Tier 3 (Security): PASS -- Audit: 0 high/crit, Secrets: clean
--------------------------------------------------------------------------------
-
-OPTIONS:
-  - "approve" / "push" to push to remote
-  - "abort" to stop (commit remains local)
-
-===============================================================================
-```
-
-After approval: `git push origin [branch]`, then verify with `gh run list --limit 3`.
+**Gate**: APPROVAL POINT #4: PUSH
 
 ---
 
 ## PHASE 6: COMPLETION
 
-### 6.1 Final Report
-
-```
-===============================================================================
-GOLDEN PATH COMPLETE
-===============================================================================
-
-SUMMARY:
--------------------------------------------------------------------------------
-Phase 0: Requirements & Context    - D1-D10 resolved
-Phase 1: Plan Creation & Audit     - [N] items, [M] audit passes
-Phase 2: Implementation            - [N] audit loops, 3 reviewers passed
-Phase 2G: Browser Verification     - [N] pages tested, [M] issues fixed / SKIPPED
-Phase 3: Simplification            - [N] findings fixed
-Phase 4: Pre-Commit Verification   - All gates passed
-Phase 5: Push Verification         - 3 tiers passed, 0 regressions
--------------------------------------------------------------------------------
-
-DELIVERABLES:
-  - Plan: [plan path]
-  - Commit: [hash]
-  - Branch: [branch]
-  - Pushed: YES
-  - Files changed: [N]
-
-===============================================================================
-```
-
-### 6.2 Plan Document Update (MANDATORY)
-
-Add to TOP of plan document:
-
-```markdown
-# IMPLEMENTATION STATUS
+Read `references/phase-6-completion.md` for full details.
 
-**Plan**: [Name]
-**Status**: COMPLETE -- PUSHED
-**Last Updated**: [YYYY-MM-DD HH:MM]
-**Push Commit**: [hash]
-**Completed By**: Claude Code (Massu Golden Path)
-
-## Task Completion Summary
-| # | Task/Phase | Status | Verification | Date |
-|---|------------|--------|--------------|------|
-| 1 | [description] | 100% COMPLETE | VR-BUILD: Pass | [date] |
-```
-
-### 6.3 Auto-Learning Protocol (MANDATORY)
-
-1. Review ALL fixes: `git diff origin/main..HEAD`
-2. For each fix: verify memory files updated
-3. For each new pattern: verify recorded
-4. For each failed approach: verify recorded
-5. Record user corrections to `memory/corrections.md`
-6. Consider new CR rule if a class of bug was found
-
-### 6.4 Update Session State
-
-Update `session-state/CURRENT.md` with completion status.
+**Summary**: Final report with phase-by-phase status. Plan document update (IMPLEMENTATION STATUS at top). Auto-learning protocol (memory ingest for all fixes/patterns). Quality & observability report. Feature registration. Session state update.
 
 ---
 
-## NEW PATTERN APPROVAL (APPROVAL POINT #2 -- Any Phase)
+## Skill Contents
 
-If a new pattern is needed during ANY phase:
+This skill is a folder. The following files are available for reference:
 
-```
-===============================================================================
-APPROVAL REQUIRED: NEW PATTERN
-===============================================================================
-
-A new pattern is needed for: [functionality]
-
-Existing patterns checked:
-- [pattern 1]: Not suitable because [reason]
-
-PROPOSED NEW PATTERN:
--------------------------------------------------------------------------------
-Name: [Pattern Name]
-Domain: [Config/MCP/Hook/etc.]
-
-WRONG: ```[code]```
-CORRECT: ```[code]```
-Error if violated: [What breaks]
--------------------------------------------------------------------------------
-
-OPTIONS:
-  - "approve" to save and continue
-  - "modify: [changes]" to adjust
-  - "abort" to stop
-
-===============================================================================
-```
+| File | Purpose | Read When |
+|------|---------|-----------|
+| `references/phase-0-requirements.md` | Requirements interview, ambiguity detection, 10-dimension coverage map | Starting a new implementation from a task description |
+| `references/phase-1-plan-creation.md` | Blast radius analysis, plan generation, audit loop | Writing or auditing a plan |
+| `references/phase-2-implementation.md` | Item loop, sprint contracts, multi-perspective review, QA evaluator, verification audit, browser testing | Executing implementation; any Phase 2 sub-phase |
+| `references/sprint-contract-protocol.md` | Sprint contract template, quality bar, negotiation rules, skip conditions | Phase 2A.5 sprint contract negotiation |
+| `references/qa-evaluator-spec.md` | Adversarial QA evaluator: 4 dimensions, anti-leniency rules, known failure patterns | Phase 2C.2 QA evaluation (UI plans only) |
+| `references/vr-visual-calibration.md` | Score 5/3/1 calibration examples for VR-VISUAL weighted dimensions | Calibrating VR-VISUAL evaluator scoring |
+| `references/phase-2.5-gap-analyzer.md` | Gap/enhancement analysis loop, 7 categories (incl. sprint contract compliance), fix-and-repass until zero | After implementation, before simplification |
+| `references/phase-3-simplify.md` | Pattern scanner fast gate, dead code detection, parallel semantic review agents | Running simplification after implementation |
+| `references/phase-3.5-security-audit.md` | Deep adversarial security audit loop with parallel red-team agents, iterate to zero findings | After simplification, before commit verification |
+| `references/phase-4-commit.md` | Verification gates, quality scoring, commit format | Preparing a commit |
+| `references/phase-5-push.md` | Pre-flight, push verification, regression detection | Preparing to push to remote |
+| `references/phase-6-completion.md` | Final report, plan status update, auto-learning, feature registration | After all verification; completing the golden path |
+| `references/approval-points.md` | Exact format and options for all 4 approval points (5 with --competitive: Plan, New Pattern, Winner Selection, Commit, Push) | Presenting any approval gate to the user |
+| `references/competitive-mode.md` | Competitive mode protocol: agent spawning, scoring, winner selection | Using --competitive flag |
+| `references/error-handling.md` | Abort handling, non-recoverable errors, post-compaction re-verification, competitive mode errors | On user abort, blocker error, or after context compaction |
 
 ---
 
-## ABORT HANDLING
+## Gotchas
 
-```
-===============================================================================
-GOLDEN PATH ABORTED
-===============================================================================
-
-Stopped at: [Phase N -- Approval Point]
-
-CURRENT STATE:
-  - Completed phases: [list]
-  - Pending phases: [list]
-  - Plan file: [path]
-  - Files changed: [list]
-  - Commit created: YES/NO
-  - Pushed: NO
-
-TO RESUME:
-  Run /massu-golden-path again with the same plan
-  Or run individual commands:
-    /massu-loop      -- Continue implementation
-    /massu-commit    -- Run commit verification
-    /massu-push      -- Run push verification
-
-===============================================================================
-```
+- **Compaction mid-loop loses plan state** -- if context compaction occurs during implementation, the plan file path and current item must be recoverable from session-state/CURRENT.md
+- **UI items need browser verification (CR-41)** -- any plan item touching UI files must be verified with Playwright before claiming done
+- **Approval points must not be skipped** -- there are 4 approval gates (5 with --competitive: Plan, New Pattern, Winner Selection, Commit, Push)
+- **Plan file must be re-read from disk, not memory (CR-5)** -- after compaction, always re-read the plan file. Memory of plan contents drifts from reality
+- **100% coverage required (CR-11)** -- never stop early. "Most items done" is not "all items done"
+- **--competitive increases token cost ~2-3x for Phase 2** -- use for high-stakes features only
+- **Competing agents do NOT run database migrations** -- DB changes must be applied separately before competitive mode
+- **Worktree branches must be mergeable** -- competing agents edit different files from the same plan, but shared files may conflict
+- **Bias presets are suggestions, not constraints** -- agents may deviate if the plan requires a specific approach
 
 ---
 
-## ERROR HANDLING
-
-**Recoverable**: Fix automatically --> re-run failed step --> if fixed, continue without pausing --> if not fixable after 3 attempts, pause and report.
+## Quality Scoring Criteria
 
-**Non-Recoverable**:
-```
-===============================================================================
-GOLDEN PATH BLOCKED
-===============================================================================
-
-BLOCKER: [Description]
-Required: [Steps to resolve]
-After resolving, run /massu-golden-path again.
+| Dimension | Weight | Measured By |
+|-----------|--------|-------------|
+| Code Clarity | 1-5 | Naming, structure, comments |
+| Pattern Compliance | 1-5 | CLAUDE.md patterns followed |
+| Error Handling | 1-5 | Edge cases, validation, fallbacks |
+| UX Quality | 1-5 | Loading/error/empty states, accessibility |
+| Test Coverage | 1-5 | Test files exist for new code |
 
-===============================================================================
-```
+All >= 3: PASS. Any < 3: FAIL.
 
 ---
 
 ## START NOW
 
-**Step 0: Write AUTHORIZED_COMMAND to session state (CR-35)**
+**Step 0: Write AUTHORIZED_COMMAND to session state (CR-12)**
 
 Update `session-state/CURRENT.md`:
 ```
@@ -963,11 +237,14 @@ AUTHORIZED_COMMAND: massu-golden-path
 
 1. **Determine input**: Task description, plan file, or continue
 2. **Phase 0**: Requirements & context (if task description)
-3. **Phase 1**: Plan creation & audit --> **PAUSE: Plan Approval**
+3. **Phase 1**: Plan creation & audit -> **PAUSE: Plan Approval**
 4. **Phase 2**: Implementation with verification loops + browser verification (UI changes)
-5. **Phase 3**: Simplification (efficiency, reuse, patterns)
-6. **Phase 4**: Pre-commit verification --> **PAUSE: Commit Approval**
-7. **Phase 5**: Push verification --> **PAUSE: Push Approval**
-8. **Phase 6**: Completion, learning, quality metrics
+4a. **Phase 2-COMP**: Competitive implementation (if --competitive) -> **PAUSE: Winner Selection**
+5. **Phase 2.5**: Gap & enhancement analysis loop (until zero gaps)
+6. **Phase 3**: Simplification (efficiency, reuse, patterns)
+6.5. **Phase 3.5**: Deep security audit (adversarial red-team loop to zero findings)
+7. **Phase 4**: Pre-commit verification -> **PAUSE: Commit Approval**
+8. **Phase 5**: Push verification via `scripts/push-verify.sh` -> **PAUSE: Push Approval**
+9. **Phase 6**: Completion, learning, quality metrics
 
 **This command does NOT stop to ask "should I continue?" -- it runs straight through.**