@su-record/vibe 2.9.18 → 2.9.20

package/commands/vibe.run.md

@@ -1,1925 +1,1940 @@
----
-description: Execute implementation from SPEC
-argument-hint: "feature name" or --phase N
----
-
-# /vibe.run
-
-Execute **Scenario-Driven Implementation** with automatic quality verification.
-
-> **Core Principle**: Scenarios are both the implementation unit and verification criteria. All scenarios passing = Quality guaranteed.
-
-## Usage
-
-```
-/vibe.run "feature-name"              # Full implementation
-/vibe.run "feature-name" --phase 1    # Specific Phase only
-/vibe.run "feature-name" ultrawork    # ULTRAWORK mode (recommended)
-/vibe.run "feature-name" ulw          # Short alias for ultrawork
-```
-
----
-
-## Codex Plugin Integration
-
-> **Codex 플러그인 감지**: 워크플로우 시작 시 아래 명령으로 자동 감지.
->
-> ```bash
-> CODEX_AVAILABLE=$(node "{{VIBE_PATH}}/hooks/scripts/codex-detect.js" 2>/dev/null || echo "unavailable")
-> ```
->
-> `available`이면 `/codex:rescue` (구현 위임), `/codex:review` (코드 리뷰) 자동 호출. `unavailable`이면 기존 워크플로우로 동작.
-
----
-
-> **⏱️ Timer**: Call `getCurrentTime` tool at the START. Record the result as `{start_time}`.
-
-## File Reading Policy (Mandatory)
-
-- **SPEC/Feature 파일**: 반드시 `Read` 도구로 전체 파일을 읽을 것 (Grep 금지)
-- **소스코드 파일**: 구현/수정 대상 파일은 반드시 `Read` 도구로 전체 읽은 후 작업할 것
-- **Grep 사용 제한**: 파일 위치 탐색(어떤 파일에 있는지 찾기)에만 사용. 파일 내용 파악에는 반드시 Read 사용
-- **에이전트 spawn 시**: 프롬프트에 "대상 파일을 Read 도구로 전체 읽은 후 구현하라"를 반드시 포함할 것
-- **부분 읽기 금지**: Grep 결과의 주변 몇 줄만 보고 수정하지 말 것. 전체 맥락을 파악해야 기존 코드와 일관된 구현 가능
-
-## **Scenario-Driven Development (SDD)**
-
-> Automate **Scenario = Implementation = Verification** so even non-developers can trust quality
-
-### Core Flow
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│                    SCENARIO-DRIVEN IMPLEMENTATION                │
-│                                                                  │
-│   Load Feature file                                              │
-│        ↓                                                        │
-│   ┌──────────────────────────────────────────────────────────┐  │
-│   │ Scenario 1: Happy Path                                    │  │
-│   │   Given → When → Then                                     │  │
-│   │        ↓                                                  │  │
-│   │   [Implement] → [Verify immediately] → ✅ Pass            │  │
-│   └──────────────────────────────────────────────────────────┘  │
-│        ↓                                                        │
-│   ┌──────────────────────────────────────────────────────────┐  │
-│   │ Scenario 2: Edge Case                                     │  │
-│   │   Given → When → Then                                     │  │
-│   │        ↓                                                  │  │
-│   │   [Implement] → [Verify] → ❌ Fail → [Fix] → ✅ Pass      │  │
-│   └──────────────────────────────────────────────────────────┘  │
-│        ↓                                                        │
-│   ┌──────────────────────────────────────────────────────────┐  │
-│   │ Scenario N: ...                                           │  │
-│   │   [Implement] → [Verify immediately] → ✅ Pass            │  │
-│   └──────────────────────────────────────────────────────────┘  │
-│        ↓                                                        │
-│   ┌──────────────────────────────────────────────────────────┐  │
-│   │  📊 QUALITY REPORT                                        │  │
-│   │  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━   │  │
-│   │  Scenarios: 5/5 passed ✅                                 │  │
-│   │  Quality score: 94/100                                    │  │
-│   │  Build: ✅ | Tests: ✅                                    │  │
-│   └──────────────────────────────────────────────────────────┘  │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-### Scenario = Implementation Unit
-
-**Traditional approach (Phase-based)**:
-```
-Phase 1 → Phase 2 → Phase 3 → ... → Verify at the end
-                                      ↓
-                              "Where did it go wrong?"
-```
-
-**SDD approach (Scenario-based)**:
-```
-Scenario 1 → Implement → Verify ✅
-Scenario 2 → Implement → Verify ✅
-Scenario 3 → Implement → Verify ❌ → Fix → ✅
-...
-All pass = Quality guaranteed
-```
-
-### Automated Verification (Closed Loop)
-
-**자율적 AI 코딩 = 구현 + 검증 + 반복. 검증을 AI에게 맡기는 순간 루프가 닫힌다.**
-
-After implementing each scenario, **automatic verification**:
-
-| Verification Item | Auto Check | Method |
-|-------------------|------------|--------|
-| Given (precondition) | State/data preparation confirmed | Code analysis |
-| When (action) | Feature execution possible | Code analysis + Build |
-| Then (result) | Expected result matches | Code analysis + Test |
-| Code quality | Complexity, style, security | Static analysis |
-| **UI behavior** | **실제 브라우저에서 동작 확인** | **E2E Closed Loop** |
-
-### E2E Closed Loop (UI Scenarios)
-
-**UI 시나리오가 포함된 Feature일 때 자동 활성화.**
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  CLOSED LOOP = 구현 + 검증 + 반복                                 │
-│                                                                  │
-│  Scenario 구현 완료                                               │
-│       ↓                                                          │
-│  [E2E 검증] AI가 직접 브라우저 조작                                │
-│       │   (접근성 트리 기반 — 토큰 효율적)                         │
-│       ├─ PASS → Next scenario                                    │
-│       └─ FAIL → Root cause analysis                              │
-│              → Fix code (Read full file first)                   │
-│              → Re-run failed scenario only                       │
-│              → PASS? → Next scenario                             │
-│              → FAIL (same as prev)? → STUCK → Ask user           │
-│                                                                  │
-│  핵심: 검증이 가벼울수록 루프는 더 많이 돈다                       │
-│  - 접근성 트리: button "Sign In" = 15 chars                      │
-│  - DOM 트리: div class="nav-wrapper mx-4..." = 200+ chars        │
-│  - 전자를 써야 한 세션에서 수십 개 시나리오 검증 가능               │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-**Browser Tool Priority:**
-
-| Priority | Tool | 용도 |
-|----------|------|------|
-| 1st | Agent Browser (접근성 트리) | AI 직접 조작, 최소 토큰 |
-| 2nd | Playwright Test Runner | 테스트 코드 실행, pass/fail 반환 |
-| 3rd | Playwright MCP (DOM) | 최후 수단, 토큰 비효율 |
-
-**활성화 조건:**
-- Feature 파일에 UI 관련 시나리오 존재 (form, button, page, navigate 등)
-- `.claude/vibe/e2e/config.json`의 `closedLoop.enabled: true` (기본값)
-- dev server가 실행 중 (`baseURL` 접근 가능)
-
-### Auto-Fix on Failure
-
-```
-Scenario verification failed (코드 분석 또는 E2E)
-      ↓
-[Collect evidence]
-  - E2E: screenshot, console errors, accessibility tree snapshot
-  - Code: build errors, test failures, type errors
-      ↓
-[Root cause analysis] - Which Then condition failed?
-      ↓
-[Read target file FULLY] - Grep으로 훑지 말 것
-      ↓
-[Implement fix] - Fix only that part
-      ↓
-[Re-verify] - Re-run ONLY failed scenario (save tokens)
-      ↓
-Repeat until pass (라운드 수 캡 없음 — stuck 감지로 종료)
-```
-
-**Termination conditions:**
-- ✅ PASS → 다음 scenario
-- ⚠️ Stuck (같은 failure가 이전 라운드와 동일) → 사용자 질문:
-    1. 직접 수정 방법 제공 (예: "LoginForm.tsx 42줄 수정") → 재시도
-    2. "proceed" — 남은 실패 TODO 기록 후 다음 scenario
-    3. "abort" — 구현 중단
-- `ultrawork` 모드: stuck 시 프롬프트 없이 TODO 기록 후 다음 scenario
-
----
-
-## **ULTRAWORK Mode** (ulw)
-
-> Include `ultrawork` or `ulw` in your command to activate **maximum performance mode**.
-
-### What ULTRAWORK Enables
-
-When you include `ultrawork` (or `ulw`), ALL of these activate automatically:
-
-| Feature | Description |
-|---------|-------------|
-| **Parallel Exploration** | 3+ Task(haiku) agents run simultaneously |
-| **Boulder Loop** | Auto-continues until ALL phases complete |
-| **Context Compression** | Aggressive auto-save at 70%+ context |
-| **No Pause** | Doesn't wait for confirmation between phases |
-| **External LLMs** | Auto-consults GPT/Gemini if enabled |
-| **Error Recovery** | Loops until 100% or stuck; on stuck auto-records TODO and proceeds (no user prompt) |
-| **Race Review (v2.6.9)** | Multi-LLM review (GPT+Gemini) with cross-validation |
-
-### Boulder Loop (Inspired by Sisyphus)
-
-Like Sisyphus rolling the boulder, ULTRAWORK **keeps going until done**:
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│                    BOULDER LOOP (ultrawork)                      │
-│                                                                  │
-│   ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐  │
-│   │ Phase 1  │───→│ Phase 2  │───→│ Phase 3  │───→│ Phase N  │  │
-│   └──────────┘    └──────────┘    └──────────┘    └──────────┘  │
-│        │               │               │               │         │
-│        ↓               ↓               ↓               ↓         │
-│   [Parallel]      [Parallel]      [Parallel]      [Parallel]    │
-│   [Implement]     [Implement]     [Implement]     [Implement]   │
-│   [Test]          [Test]          [Test]          [Test]        │
-│        │               │               │               │         │
-│        └───────────────┴───────────────┴───────────────┘         │
-│                              │                                   │
-│                              ↓                                   │
-│                     ┌──────────────┐                             │
-│                     │  ALL DONE?   │                             │
-│                     └──────────────┘                             │
-│                       │         │                                │
-│                      NO        YES                               │
-│                       │         │                                │
-│                       ↓         ↓                                │
-│                   [Continue]  [🎉 Complete!]                     │
-│                                                                  │
-│   NO STOPPING until acceptance criteria met or error limit hit   │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-### Ralph Loop (Completion Verification) - CRITICAL
-
-> **Inspired by [ghuntley.com/ralph](https://ghuntley.com/ralph)**: "Deterministically bad in an undeterministic world" - Keep iterating until TRULY complete.
-
-**Problem**: AI often claims "complete" when implementation is partial.
-
-**Solution**: RTM-based automated coverage verification with iteration tracking.
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│                    RALPH LOOP (Mandatory)                        │
-│                                                                  │
-│   After ALL phases complete:                                     │
-│                                                                  │
-│   ┌──────────────────────────────────────────────────────────┐  │
-│   │  RTM COVERAGE VERIFICATION [Iteration {{ITER}}/{{MAX}}]   │  │
-│   │                                                           │  │
-│   │  Generate RTM via core tools:                             │  │
-│   │  → generateTraceabilityMatrix("{feature-name}")           │  │
-│   │                                                           │  │
-│   │  Coverage Metrics (automated):                            │  │
-│   │  □ Requirements coverage: {coveragePercent}%              │  │
-│   │  □ SPEC → Feature mapping: {featureCovered}/{total}       │  │
-│   │  □ Feature → Test mapping: {testCovered}/{total}          │  │
-│   │  □ Build successful?                                      │  │
-│   │  □ Tests passing?                                         │  │
-│   │                                                           │  │
-│   │  UNCOVERED: {uncoveredRequirements[]}                     │  │
-│   └──────────────────────────────────────────────────────────┘  │
-│                              │                                   │
-│                   ┌──────────┴──────────┐                       │
-│                   │  Coverage == 100%?  │                       │
-│                   └──────────┬──────────┘                       │
-│                       │              │                          │
-│                      NO            YES                          │
-│                       │              │                          │
-│                       ↓              ↓                          │
-│              ┌────────────────┐  ┌────────────────┐             │
-│              │ IMPLEMENT      │  │ ✅ TRULY DONE  │             │
-│              │ UNCOVERED      │  │                │             │
-│              │ REQUIREMENTS   │  │ Report final   │             │
-│              │ (auto-extract) │  │ RTM coverage   │             │
-│              └───────┬────────┘  └────────────────┘             │
-│                      │                                          │
-│                      └──────────→ [Re-generate RTM]             │
-│                                                                  │
-│                      │                                          │
-│                      ↓                                          │
-│              Stuck? (coverage unchanged from prev iteration)    │
-│                      │                                          │
-│                      ├─ Interactive: Ask user                   │
-│                      │     1. Provide resolution → retry        │
-│                      │     2. "proceed" → TODO + done           │
-│                      │     3. "abort" → stop                    │
-│                      └─ ultrawork: TODO + done                  │
-│                                                                  │
-│   NO iteration cap — loop until 100% OR stuck                   │
-│   ZERO TOLERANCE for silent scope reduction                     │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-**Ralph Loop with RTM:**
-
-```bash
-# Generate RTM for coverage verification
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.generateTraceabilityMatrix('{feature-name}', {projectPath: process.cwd()}).then(r => console.log(JSON.stringify(r, null, 2))))"
-```
-
-**RTM provides automated metrics:**
-
-| Metric | Description |
-|--------|-------------|
-| `totalRequirements` | Total REQ-* items in SPEC |
-| `specCovered` | Requirements with SPEC mapping |
-| `featureCovered` | Requirements with Feature scenarios |
-| `testCovered` | Requirements with test files |
-| `coveragePercent` | Overall coverage percentage |
-| `uncoveredRequirements` | List of missing REQ-* IDs |
-
-**Ralph Loop Rules:**
-
-| Rule | Description |
-|------|-------------|
-| **No Scope Reduction** | Never say "simplified" or "basic version" - implement FULL request |
-| **Iteration Tracking** | Display `[{{ITER}}]` to show progress (no max — loop until done) |
-| **RTM-Based Gap List** | Use `uncoveredRequirements` array - no manual comparison |
-| **Coverage Threshold** | Must reach 100% coverage to complete |
-| **No Iteration Cap** | Loop until 100% coverage OR stuck (convergence detected) |
-| **Stuck Handling** | If coverage % unchanged between iterations → ask user (proceed/abort/fix), or ultrawork → TODO + done |
-| **Diminishing Returns (Narrowing)** | Iteration 3+ → focus on core requirements (REQ-*-001~003) first; P2/P3 requirements continue to loop but lower priority |
-
-**Ralph Loop Output Format:**
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🔄 RALPH VERIFICATION [Iteration 1]
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-📊 RTM Coverage Report: login
-
-Requirements Traceability:
-  Total Requirements: 9
-  SPEC Covered: 9/9 (100%)
-  Feature Covered: 5/9 (55%)
-  Test Covered: 4/9 (44%)
-
-  ✅ REQ-login-001: Login form UI → Scenario 1 → login.test.ts
-  ✅ REQ-login-002: Email validation → Scenario 2 → validation.test.ts
-  ✅ REQ-login-003: Password validation → Scenario 2 → validation.test.ts
-  ❌ REQ-login-004: Remember me checkbox → NOT IMPLEMENTED
-  ❌ REQ-login-005: Forgot password link → NOT IMPLEMENTED
-  ✅ REQ-login-006: API integration → Scenario 3 → api.test.ts
-  ❌ REQ-login-007: Loading state → NOT IMPLEMENTED
-  ❌ REQ-login-008: Error toast → NOT IMPLEMENTED
-  ✅ REQ-login-009: Session storage → Scenario 4 → (no test)
-
-Overall Coverage: 55% ⚠️ BELOW 100% TARGET
-
-UNCOVERED REQUIREMENTS (auto-extracted from RTM):
-  1. REQ-login-004: Remember me checkbox
-  2. REQ-login-005: Forgot password link
-  3. REQ-login-007: Loading state
-  4. REQ-login-008: Error toast notifications
-
-⚠️ NOT COMPLETE - Implementing uncovered requirements...
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🔄 RALPH VERIFICATION [Iteration 2]
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-📊 RTM Coverage Report: login
-
-Requirements Traceability:
-  Total Requirements: 9
-  SPEC Covered: 9/9 (100%)
-  Feature Covered: 9/9 (100%)
-  Test Covered: 9/9 (100%)
-
-Overall Coverage: 100% ✅ TARGET REACHED
-
-Build: ✅ Passed
-Tests: ✅ 12/12 Passed
-Type Check: ✅ No errors
-
-✅ RALPH VERIFIED COMPLETE!
-
-📄 RTM saved: .claude/vibe/rtm/login-rtm.md
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-```
-
-**When to Trigger Ralph Loop:**
-
-1. After all phases complete
-2. Before final quality report
-3. Whenever user says "ultrawork" or "ralph"
-
-**Forbidden Responses (VIOLATIONS):**
-
-| ❌ NEVER Say | ✅ Instead |
-|-------------|-----------|
-| "I've implemented a basic version" | Implement the FULL version |
-| "This is a simplified approach" | Implement as specified |
-| "You can add X later" | Add X now |
-| "For demonstration purposes" | Implement production-ready |
-| "The core functionality is done" | ALL functionality must be done |
-
-### ULTRAWORK Example
-
-```
-User: /vibe.run "brick-game" ultrawork
-
-Claude:
-🚀 ULTRAWORK MODE ACTIVATED
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-📄 SPEC: .claude/vibe/specs/brick-game.md
-🎯 4 Phases detected
-⚡ Boulder Loop: ENABLED (will continue until all phases complete)
-🔄 Auto-retry: ON (loop until 100% or stuck → auto-TODO)
-💾 Context compression: AGGRESSIVE
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🏔️ BOULDER ROLLING... Phase 1/4
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-⚡ [PARALLEL] Launching 3 exploration agents...
-✅ Exploration complete (7.2s)
-🔨 Implementing...
-✅ Phase 1 complete
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🏔️ BOULDER ROLLING... Phase 2/4
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-⚡ [PARALLEL] Launching 3 exploration agents...
-✅ Exploration complete (6.8s)
-🔨 Implementing...
-❌ Test failed: collision detection
-🔄 Auto-retry [iteration 1]...
-🔨 Fixing...
-✅ Phase 2 complete
-
-[...continues automatically...]
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎉 BOULDER REACHED THE TOP!
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-✅ All 4 phases complete
-✅ All acceptance criteria passed
-✅ Build succeeded
-✅ Tests passed
-
-⏱️ Total: 8m 24s
-📊 Retries: 2
-💾 Context saved: 3 checkpoints
-```
-
-### Normal vs ULTRAWORK Comparison
-
-| Aspect | Normal | ULTRAWORK |
-|--------|--------|-----------|
-| Phase transition | May pause | Auto-continues |
-| On error | Reports and stops | Auto-retries (3x) |
-| Context 70%+ | Warning only | Auto-compress + save |
-| Exploration | Sequential possible | FORCED parallel |
-| Completion | Phase-by-phase | Until ALL done |
-
----
-
-## Automation Level System
-
-Magic keywords in the user input automatically set the **AutomationLevel**, which controls how much the AI self-advances vs. pausing for confirmation.
-
-### Level Definitions
-
-| Level | Name | Keyword(s) | Auto-advance | Auto-retry | Stuck Behavior | Parallel Agents | Checkpoints |
-|-------|------|------------|--------------|------------|----------------|-----------------|-------------|
-| L0 | Manual | `manual` | No | No | Ask user every step | No | All |
-| L1 | Guided | `guided`, `verify` | No | No | Ask user on stuck | No | All |
-| L2 | Semi-auto | `quick` (default) | Yes | Yes (low cap: 2) | Ask user after 2 retries | No | Key points |
-| L3 | Auto | `ultrawork`, `ulw` | Yes | Yes (no cap) | Auto-TODO + proceed | Yes | Checkpoint-only |
-| L4 | Full-auto | `ralph`, `ralplan` | Yes | Yes (no cap) | Auto-TODO + proceed | Yes | None |
-
-### Detection Rule
-
-The AI detects automation level from word boundaries in the user's input:
-
-```
-/vibe.run "login"              → L2 Semi-auto (default)
-/vibe.run "login" ultrawork    → L3 Auto
-/vibe.run "login" ralph        → L4 Full-auto
-/vibe.run "login" verify       → L1 Guided
-```
-
-### Confirmation Matrix
-
-Certain action types require user confirmation depending on the active level:
-
-| Action | L0 | L1 | L2 | L3 | L4 |
-|--------|----|----|----|----|-----|
-| `destructive` | confirm | confirm | confirm | confirm | auto |
-| `architecture_choice` | confirm | confirm | confirm | auto | auto |
-| `implementation_scope` | confirm | confirm | confirm | auto | auto |
-| `phase_advance` | confirm | confirm | auto | auto | auto |
-| `fix_strategy` | confirm | confirm | auto | auto | auto |
-| `retry` | confirm | auto | auto | auto | auto |
-
-**Rule**: When confirmation is required, pause and display a checkpoint before proceeding.
-
----
-
-## Interactive Checkpoints
-
-Checkpoints are decision gates inserted at critical points in the workflow. At L3/L4, most checkpoints are **auto-resolved** using the default option.
-
-### Checkpoint Types
-
-| Type | When It Fires | Default Option |
-|------|--------------|----------------|
-| `requirements_confirm` | Before starting Phase 1 | Confirm (a) |
-| `architecture_choice` | When architecture approach is ambiguous | Clean/balanced (b) |
-| `implementation_scope` | Before any large scope change (6+ files) | Approve (a) |
-| `verification_result` | After each VerificationLoop iteration below threshold | Continue fixing (a) |
-| `fix_strategy` | When critical issues are found during quality gate | Fix all (a) |
-
-### Checkpoint Format
-
-When a checkpoint fires, display it in this format:
-
-```
-──────────────────────────────────────────────────
-CHECKPOINT: Requirements Confirmation
-──────────────────────────────────────────────────
-Feature: login
-
-Requirements:
-  1. User can log in with email and password
-  2. Invalid credentials show an error message
-  3. Remember me persists session for 30 days
-
-Options:
-  a) Confirm
-     Proceed with these requirements as stated.
-  b) Revise
-     Modify or clarify requirements before proceeding.
-  c) Abort
-     Cancel this workflow.
-
-Default: a
-──────────────────────────────────────────────────
-```
-
-### Auto-Resolution (L3/L4)
-
-At automation levels L3 and L4, checkpoints that do not require confirmation are auto-resolved:
-
-```
-[AUTO] CHECKPOINT: implementation_scope → option a (Approve) [auto-resolved]
-```
-
----
-
-## Rules Reference
-
-**Must follow `~/.claude/vibe/rules/` (global):**
-
-- `core/development-philosophy.md` - Surgical precision, modify only requested scope
-- `core/quick-start.md` - Korean, DRY, SRP, YAGNI
-- `standards/complexity-metrics.md` - Functions ≤20 lines, nesting ≤3 levels
-- `quality/checklist.md` - Code quality checklist
-
-**Language guide:** `~/.claude/vibe/languages/{stack}.md` (global reference)
-
----
-
-## Coding Guidelines (Mandatory)
-
-### Type Safety: Use Types Explicitly
-
-> **Core Principle**: Use types explicitly in every language that has a type system!
-
-Type definitions are not just language syntax — they are a **core engineering philosophy for simplifying and controlling complex software**.
-
-### Applies to ALL Typed Languages
-
-| Category | Languages | Key Principle |
-|----------|-----------|---------------|
-| **Static Typed** | Java, C#, C++, Go, Rust, Swift, Kotlin, Scala | Types = compile-time contracts |
-| **Gradual Typed** | TypeScript, Python (typing), PHP (typed), Ruby (RBS) | Types = optional safety nets |
-| **Functional** | Haskell, OCaml, F#, Elm | Types = logical proofs |
-
-### Universal Anti-Patterns (All Languages)
-
-| ❌ Forbidden Pattern | Why | ✅ Instead |
-|---------------------|-----|-----------|
-| Type escape hatches (`any`, `Any`, `Object`, `void*`, `interface{}`) | Loses type info, runtime errors | Concrete types or `unknown` + guards |
-| Type suppression (`@ts-ignore`, `# type: ignore`, `@SuppressWarnings`) | Hides errors | Fix actual type issues |
-| Raw generic types (`List`, `Map` without params) | Loses type safety | `List<User>`, `Map<String, Order>` |
-| Excessive casting (`as`, `(Type)`, `unsafe`) | Bypasses compiler | Type guards or pattern matching |
-
-### Language-Specific Guidelines
-
-**TypeScript/JavaScript:**
-```typescript
-// ❌ BAD
-function process(data: any): any { return data.foo; }
-
-// ✅ GOOD
-function process(data: unknown): Result {
-  if (isValidData(data)) return data.foo;
-  throw new Error('Invalid');
-}
-```
-
-**Python:**
-```python
-# ❌ BAD
-def process(data: Any) -> Any: return data["key"]
-
-# ✅ GOOD
-def process(data: UserData) -> str: return data["name"]
-```
-
-**Java/Kotlin:**
-```java
-// ❌ BAD
-List items = new ArrayList();  // Raw type
-Object data = getData();       // Lost type info
-
-// ✅ GOOD
-List<User> users = new ArrayList<>();
-User user = getUser();
-```
-
-**Go:**
-```go
-// ❌ BAD
-func process(data interface{}) interface{} { ... }
-
-// ✅ GOOD
-func process(data UserRequest) (UserResponse, error) { ... }
-```
-
-**Rust:**
-```rust
-// ❌ BAD (unnecessary unsafe or Box<dyn Any>)
-let data: Box<dyn Any> = get_data();
-
-// ✅ GOOD
-let data: UserData = get_data()?;
-```
-
-**C#:**
-```csharp
-// ❌ BAD
-object data = GetData();
-dynamic result = Process(data);
-
-// ✅ GOOD
-UserData data = GetData();
-Result result = Process(data);
-```
-
-### Type Safety Rules (Universal)
-
-| Rule | Description |
-|------|-------------|
-| **Boundary Validation** | Validate only at system boundaries (API, JSON, user input) |
-| **Internal Trust** | After validation, pass only precise types internally |
-| **No Type Escape** | Never use escape hatches to "fix" type errors |
-| **Explicit Signatures** | Specify types in function/method signatures |
-| **Generics with Params** | Always use generics with type parameters |
-
-### Quality Gate: Type Violations Block Merge
-
-| Violation | Action |
-|-----------|--------|
-| Type escape hatches (`any`, `Any`, `Object`, `interface{}`, etc.) | ❌ Block |
-| Type suppression comments | ❌ Block |
-| Raw generic types | ❌ Block |
-| Missing function return types | ⚠️ Warning |
-| Excessive type casting | ⚠️ Warning |
-
-## Description
-
-Read PTCF structured SPEC document and execute implementation immediately.
-
-> **PLAN, TASKS documents unnecessary** - SPEC is the executable prompt
-
-## Model Orchestration (Intelligent Routing)
-
-Automatically select optimal model based on **task complexity analysis**.
-
-### Complexity-Based Model Selection
-
-| Complexity Score | Model | When to Use |
-|------------------|-------|-------------|
-| 0-7 (Low) | **Haiku** | Simple fixes, searches, single file changes |
-| 8-19 (Medium) | **Sonnet** | Standard features, 3-5 files, integrations |
-| 20+ (High) | **Opus** | Architecture, security, multi-service, 6+ files |
-
-### Complexity Signals
-
-The following signals increase complexity score:
-
-| Signal | Score |
-|--------|-------|
-| Architecture change | +15 |
-| Security implication | +12 |
-| Multi-service | +8 |
-| Refactoring | +12 |
-| 6+ files | +15 |
-| 3-5 files | +8 |
-| New feature | +5 |
-| Bug fix | -3 |
-| Documentation | -5 |
-
-### Agent Tier System
-
-Each agent has tier variants for cost optimization:
-
-| Agent | Low (Haiku) | Medium (Sonnet) | High (Opus) |
-|-------|-------------|-----------------|-------------|
-| explorer | explorer-low | explorer-medium | explorer |
-| implementer | implementer-low | implementer-medium | implementer |
-| architect | architect-low | architect-medium | architect |
-
-### Task Calls by Role
-
-| Task Type | Model | Task Parameter |
-|-----------|-------|----------------|
-| Simple search | Haiku | `model: "haiku"` |
-| Codebase exploration | Haiku/Sonnet | Auto-selected |
-| Core implementation | Sonnet | `model: "sonnet"` |
-| Test writing | Haiku | `model: "haiku"` |
-| Architecture decisions | Opus | Main session |
-| Final review | Opus | Main session |
-
-### External LLM Usage (When Enabled)
-
-When external LLMs are enabled in `.claude/vibe/config.json`:
-
-| Role | Method | Condition |
-|------|--------|-----------|
-| User direct query | `gpt.question`, `gemini.question` | Hook auto-handles |
-| Internal orchestration | Call global script via Bash | Claude calls directly |
-
-**User questions (Hook auto-handles):**
-- `gpt.question` - GPT architecture consultation
-- `gemini.question` - Gemini Q&A/consultation
-
-**Claude internal calls (directly via Bash):**
-```bash
-# Usage: node llm-orchestrate.js <provider> <mode> [systemPrompt] [prompt]
-#   - If systemPrompt omitted, uses default
-#   - If systemPrompt is "-", uses default and treats next argument as prompt
-
-# [LLM_SCRIPT] = {{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js
-
-# GPT call (short prompt - CLI arg)
-node "[LLM_SCRIPT]" gpt orchestrate-json "[question content]"
-
-# Gemini call
-node "[LLM_SCRIPT]" gemini orchestrate-json "[question content]"
-
-# Custom system prompt usage
-node "[LLM_SCRIPT]" gpt orchestrate-json "You are a code reviewer" "[question content]"
-
-# Long prompt - use --input file (write JSON file first with Write tool)
-# JSON format: {"prompt": "your prompt here"}
-node "[LLM_SCRIPT]" gpt orchestrate-json --input "[SCRATCHPAD]/input.json"
-```
-
-### External LLM Fallback
-
-**IMPORTANT**: When GPT/Gemini hook fails, Claude MUST handle the task directly:
-
-**Fallback behavior**:
-- Do NOT retry the external LLM call
-- Claude handles the task using its own capabilities
-- Continue with the implementation without interruption
-- Log the fallback but don't block progress
-
-## Core Tools (Semantic Analysis & Memory)
-
-Use core tools for accurate codebase understanding and session continuity.
-
-### Tool Invocation
-
-All tools are called via:
-```bash
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.TOOL_NAME({...args}).then(r => console.log(r.content[0].text)))"
-```
-
-### Semantic Analysis Tools
-
-| Tool | Purpose | Usage |
-|------|---------|-------|
-| `findSymbol` | Find symbol definitions | `{symbolName: 'functionName', searchPath: '.'}` |
-| `findReferences` | Find all references | `{symbolName: 'functionName', searchPath: '.'}` |
-| `analyzeComplexity` | Analyze code complexity | `{filePath: 'src/file.ts'}` |
-| `validateCodeQuality` | Validate code quality | `{filePath: 'src/file.ts'}` |
-
-**Example - Find symbol:**
-```bash
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.findSymbol({symbolName: 'login', searchPath: '.'}).then(r => console.log(r.content[0].text)))"
-```
-
-### Memory Tools
-
-| Tool | Purpose | Usage |
-|------|---------|-------|
-| `saveMemory` | Save important decisions | `{key: 'decision-name', value: 'content', category: 'project'}` |
-| `recallMemory` | Recall saved memory | `{key: 'decision-name'}` |
-| `listMemories` | List all memories | `{category: 'project'}` |
-
-**Example - Save important decision:**
-```bash
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.saveMemory({key: 'auth-pattern', value: 'Using JWT with refresh tokens', category: 'project'}).then(r => console.log(r.content[0].text)))"
-```
-
-### Session Management (Auto via Hooks)
-
-- **Session start**: Hook auto-calls `startSession` to restore previous context
-- **Context 80%+**: Hook auto-calls `autoSaveContext` to preserve state
-
-## Process
-
-### 1. Load SPEC + Feature
-
-**Search order (check BOTH file AND folder):**
-
-```
-Step 1: Check if SPLIT structure exists (folder)
-  📁 .claude/vibe/specs/{feature-name}/        → Folder with _index.md + phase files
-  📁 .claude/vibe/features/{feature-name}/      → Folder with _index.feature + phase files
-
-Step 2: If no folder, check single file
-  📄 .claude/vibe/specs/{feature-name}.md       → Single SPEC file
-  📄 .claude/vibe/features/{feature-name}.feature → Single Feature file
-
-Step 3: If neither exists → Error
-```
-
-**Split structure (folder) detected:**
-```
-📁 .claude/vibe/specs/{feature-name}/
-├── _index.md              → Master SPEC (read first for overview)
-├── phase-1-{name}.md      → Phase 1 SPEC
-├── phase-2-{name}.md      → Phase 2 SPEC
-└── ...
-
-📁 .claude/vibe/features/{feature-name}/
-├── _index.feature         → Master Feature (read first for scenario overview)
-├── phase-1-{name}.feature → Phase 1 scenarios
-├── phase-2-{name}.feature → Phase 2 scenarios
-└── ...
-
-→ Load _index.md first, then load phase files in order
-→ Execute phases sequentially (or per --phase flag)
-```
-
-**Single file detected:**
-```
-📄 .claude/vibe/specs/{feature-name}.md      → SPEC (structure, constraints, context)
-📄 .claude/vibe/features/{feature-name}.feature → Feature (scenario = implementation unit)
-```
-
-**Error if NEITHER file NOR folder found:**
-```
-❌ SPEC not found. Searched:
-   - .claude/vibe/specs/{feature-name}/  (folder)
-   - .claude/vibe/specs/{feature-name}.md (file)
-
-   Run /vibe.spec "{feature-name}" first.
-```
-
-### 1-1. Phase Isolation Protocol (Large SPEC Guard)
-
-> **Problem**: Large SPECs (3+ phases, 5+ scenarios) overflow context — agent drifts from SPEC by Phase 3.
-> **Solution**: Load only the current phase's SPEC section. Re-anchor before each scenario.
-
-**Phase Isolation Rules (MANDATORY for 3+ phases):**
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  PHASE ISOLATION PROTOCOL                                       │
-│                                                                  │
-│  ❌ WRONG: Load entire SPEC → implement all phases              │
-│  ✅ RIGHT: Load _index overview → per-phase load → implement    │
-│                                                                  │
-│  Step A: Read _index.md (overview only — phase list, REQ IDs)   │
-│  Step B: For each Phase N:                                       │
-│    1. RE-READ Phase N SPEC section (every time, no memory)       │
-│    2. RE-READ Phase N Feature scenarios                          │
-│    3. Extract Phase N scope: files, scenarios, requirements      │
-│    4. Implement Phase N scenarios                                 │
-│    5. Verify Phase N                                             │
-│    6. Write Phase Checkpoint → .claude/vibe/checkpoints/         │
-│    7. DISCARD Phase N details from working memory                │
-│  Step C: Next Phase — go to Step B                               │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-**Phase Checkpoint Format** (`.claude/vibe/checkpoints/{feature}-phase-{N}.md`):
-
-```markdown
-# Checkpoint: {feature} Phase {N}
-
-## Completed
-- Scenario 1: {name} ✅
-- Scenario 2: {name} ✅
-
-## Files Changed
-- src/auth.service.ts (added login(), validateToken())
-- src/auth.controller.ts (POST /login, POST /refresh)
-
-## State for Next Phase
-- Auth service exports: login(), logout(), validateToken(), refreshToken()
-- JWT secret configured in .env (JWT_SECRET)
-- Test baseline: 12 tests passing
-
-## Remaining Phases
-- Phase {N+1}: {name} — {scenario count} scenarios
-- Phase {N+2}: {name} — {scenario count} scenarios
-```
-
-**SPEC Re-anchoring (Before EVERY scenario):**
-
-```
-Before implementing Scenario X:
-  1. Re-read the EXACT Given/When/Then from Feature file (not from memory!)
-  2. Compare: "Am I about to implement what the SPEC says, or what I think it says?"
-  3. If single-file SPEC: re-read only the current phase section (use line offsets)
-  4. If split SPEC: re-read only phase-N-{name}.md
-```
-
-**Scope Lock (Per Phase):**
-
-```
-At Phase start, declare:
-  MODIFY: [list of files this phase will touch]
-  CREATE: [list of files this phase will create]
-  DO NOT TOUCH: everything else
-
-If implementation requires files outside scope:
-  → STOP. Re-read SPEC. Is this actually needed?
-  → If yes: add to scope with explicit justification
-  → If no: you're drifting. Return to SPEC.
-```
-
-**Context Pressure Handling:**
-
-| Context Level | Action |
-|---------------|--------|
-| < 50% | Normal execution |
-| 50-70% | Save checkpoint, trim exploration results |
-| 70%+ | Save checkpoint → `/new` → resume from checkpoint |
-| Phase boundary | Always save checkpoint regardless of context level |
-
-### 1-2. SPEC-First Gate (Level 3: Spec = Source of Truth)
-
-> **Principle**: SPEC is the source of truth for code. To modify code, update the SPEC first.
-
-**When a change not in the SPEC is needed during implementation:**
-
-```
-Discovery during implementation: "An API endpoint not in SPEC is needed"
-    │
-    ├─ Already in SPEC?
-    │   YES → Implement as-is
-    │
-    ├─ Not in SPEC but within SPEC scope?
-    │   YES → Add to SPEC first (Edit tool → SPEC file)
-    │       → Add corresponding scenario to Feature file
-    │       → Then implement
-    │
-    └─ Outside SPEC scope?
-        YES → Record as TODO and exclude from current scope
-              .claude/vibe/todos/out-of-scope-{item}.md
-```
-
-**Required when changing SPEC:**
-
-1. Update REQ-* IDs in SPEC file (add new requirements)
-2. Add corresponding scenarios to Feature file
-3. New REQ-* must be traceable in RTM after implementation
-
-**Reverse direction also applies — when requirements change from code:**
-
-```
-Test failure → "Is this behavior correct?"
-    │
-    ├─ Check SPEC → If SPEC is correct, fix the code
-    │
-    └─ If SPEC is wrong → Fix SPEC first → Then fix code
-        (Changing code without updating SPEC causes SPEC ↔ code drift)
-```
-
-**Extended Git commit rules:**
-
-- SPEC changes and code changes must be in the same commit
-- Committing feature code without SPEC changes → warning (except intentional refactoring)
-
-### 2. Extract Scenario List
-
-Extract all Scenarios from Feature file:
-
-```markdown
-## Scenarios to Implement
-
-| # | Scenario | Status |
-|---|----------|--------|
-| 1 | Valid login success | ⬜ |
-| 2 | Invalid password error | ⬜ |
-| 3 | Email format validation | ⬜ |
-| 4 | Password reset link | ⬜ |
-
-Total: 4 scenarios
-```
-
-### 3. Scenario-by-Scenario Implementation (Core)
-
-**For each scenario**:
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 1/4: Valid login success
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Given: User is registered
-When: Login with valid email and password
-Then: Login success + JWT token returned
-
-[Step 1] Analyzing implementation...
-  - Required files: auth.service.ts, login.controller.ts
-  - Exploring related code...
-
-[Step 2] Implementing...
-  ✅ auth.service.ts - Added login() method
-  ✅ login.controller.ts - POST /login endpoint
-
-[Step 3] Verifying...
-  ✅ Given: Test user creation possible
-  ✅ When: Login API call succeeded
-  ✅ Then: JWT token return confirmed
-
-✅ Scenario 1 passed!
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-```
-
-**On failure**:
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 2/4: Invalid password error
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-[Step 3] Verifying...
-  ✅ Given: Test user exists
-  ✅ When: Login attempt with wrong password
-  ❌ Then: "Invalid credentials" error message
-     Actual: "Error occurred" returned
-
-[Auto-fix iteration 1]
-  Cause: Error message not properly set
-  Fix: auth.service.ts line 42
-
-[Re-verify]
-  ✅ Then: "Invalid credentials" error message
-
-✅ Scenario 2 passed! (1 fix)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-```
-
----
-
-## **CRITICAL: Parallel Sub-Agent Execution**
-
-> **MUST USE PARALLEL TASK CALLS** - This is REQUIRED, not optional.
-> Sequential execution when parallel is possible = VIOLATION of this workflow.
-
-### Codex 병렬 구현 (Codex 플러그인 활성화 시)
-
-> **활성화 조건**: Codex 플러그인 설치 + 독립적인 시나리오 2개 이상
-> 미설치 시 기존 Claude 서브에이전트 방식으로 동작.
-
-독립적인 시나리오(파일 의존성 없음)를 Codex에 위임하여 Claude와 **병렬 구현**:
-
-```
-/codex:rescue "Implement scenario: {scenario-name}. Files: {file-list}. Requirements: {requirements-summary}" --background
-```
-
-- Claude는 다음 시나리오를 **동시에** 구현
-- Codex 완료 시 `/codex:result`로 결과 확인
-- 충돌 발생 시 Claude가 merge 판단
-
-**위임 기준**:
-- 시나리오 간 파일 의존성 없음 (독립적)
-- 시나리오 복잡도 중간 이하
-- 의존성 있는 시나리오는 Claude가 직접 구현
-
-### Mandatory Parallel Exploration (Phase Start)
-
-**BEFORE any implementation, you MUST launch these Task calls IN PARALLEL (single message, multiple tool calls):**
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  STEP 1: PARALLEL EXPLORATION (REQUIRED)                        │
-│                                                                 │
-│  Launch ALL of these in ONE message:                            │
-│                                                                 │
-│  Task(haiku) ─┬─→ "Analyze related files in <context>"          │
-│               │                                                 │
-│  Task(haiku) ─┼─→ "Check dependencies and imports"              │
-│               │                                                 │
-│  Task(haiku) ─┴─→ "Find existing patterns and conventions"      │
-│                                                                 │
-│  [If GPT enabled] Bash: node "[LLM_SCRIPT]" gpt-codex orchestrate-json "[question]"
-│  [If Gemini enabled] Bash: node "[LLM_SCRIPT]" gemini orchestrate-json "[question]"
-└─────────────────────────────────────────────────────────────────┘
-                              │
-                              ↓ (wait for all to complete)
-┌─────────────────────────────────────────────────────────────────┐
-│  STEP 2: SYNTHESIZE (Opus)                                      │
-│  - Review all exploration results                               │
-│  - Decide implementation approach                               │
-│  - Identify files to modify/create                              │
-└─────────────────────────────────────────────────────────────────┘
-                              │
-                              ↓
-┌─────────────────────────────────────────────────────────────────┐
-│  STEP 3: IMPLEMENT + BACKGROUND AGENTS (PARALLEL)               │
-│                                                                 │
-│  Main Agent (sonnet):                                           │
-│  └─→ Execute current phase implementation                       │
-│                                                                 │
-│  Background Agents (haiku, run_in_background=true):             │
-│  ├─→ Task: "Prepare Phase N+1 - analyze required files"         │
-│  ├─→ Task: "Pre-generate test cases for current implementation" │
-│  └─→ Task: "Search for related types/interfaces needed"         │
-│                                                                 │
-│  [ULTRAWORK] All 4 agents run simultaneously!                   │
-└─────────────────────────────────────────────────────────────────┘
-                              │
-                              ↓ (main completes, check backgrounds)
-┌─────────────────────────────────────────────────────────────────┐
-│  STEP 4: TEST + PHASE PIPELINING                                │
-│                                                                 │
-│  Current Phase:                                                 │
-│  └─→ Task(haiku): Write tests using pre-generated cases         │
-│                                                                 │
-│  Next Phase Prep (from background results):                     │
-│  └─→ Already have file analysis, ready to start immediately     │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-### Parallel Task Call Pattern (MUST FOLLOW)
-
-**Correct - Single message with multiple parallel Tasks:**
-```
-<message>
-  Task(haiku, "Analyze src/components/ for existing patterns")
-  Task(haiku, "Check package.json dependencies")
-  Task(haiku, "Find usage of similar features in codebase")
-</message>
-→ All 3 run simultaneously, ~3x faster
-```
-
-**WRONG - Sequential calls (DO NOT DO THIS):**
-```
-<message>Task(haiku, "Analyze...")</message>
-<message>Task(haiku, "Check...")</message>
-<message>Task(haiku, "Find...")</message>
-→ 3x slower, wastes time
-```
-
-### Background Agent Pattern (ULTRAWORK) via Orchestrator
-
-**Launch background agents for next phase via Orchestrator:**
-```bash
-# Start background agent (doesn't block)
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => o.runAgent('Phase 2 prep: Analyze auth API endpoints', 'phase2-prep').then(r => console.log(r.content[0].text)))"
-
-# Multiple backgrounds in parallel
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(async o => {
-  await Promise.all([
-    o.runAgent('Phase 2 prep: Analyze auth API endpoints', 'phase2-prep'),
-    o.runAgent('Pre-generate test cases for login form', 'test-prep'),
-    o.runAgent('Find existing validation patterns', 'pattern-finder')
-  ]);
-  console.log('All background agents started');
-})"
-```
-
-**Check background agent status:**
-```bash
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => console.log(o.status().content[0].text))"
-```
-
-**Get result when ready:**
-```bash
-node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => o.getResult('SESSION_ID').then(r => console.log(r.content[0].text)))"
-```
-
-**Why Background Agents Matter:**
-
-| Without Background | With Background |
-|--------------------|-----------------|
-| Phase 1: 60s | Phase 1: 60s (+ backgrounds running) |
-| Phase 2 prep: 20s | Phase 2 prep: 0s (already done!) |
-| Phase 2: 60s | Phase 2: 60s |
-| **Total: 140s** | **Total: 120s** |
-
-For 5 phases: 4 × 20s saved = **80s faster**
-
-### Why Parallel Matters
-
-| Approach | Time | Cache Benefit |
-|----------|------|---------------|
-| Sequential (3 Tasks) | ~30s | Cache cold on each |
-| **Parallel (3 Tasks)** | **~10s** | **Cache warmed once, shared** |
-
-core ProjectCache (LRU) caches ts-morph parsing results. Parallel calls share the warmed cache.
-
-### UI/UX Design Intelligence (Auto-triggered before Phase 1)
-
-> **조건**: SPEC 또는 Feature에 UI/UX 키워드 포함 시 자동 실행
-> **비활성화**: `.claude/vibe/config.json`에 `"uiUxAnalysis": false` 설정
-
-**Phase 1 시작 전, 2개 에이전트 자동 실행:**
-
-| Agent | Condition | Role |
-|-------|-----------|------|
-| ④ ui-stack-implementer | **항상 실행** | 프레임워크별 컴포넌트 가이드라인 제공 |
-| ⑤ ui-dataviz-advisor | **조건부** (chart/dashboard/visualization 키워드) | 차트/시각화 라이브러리 추천 |
-
-**실행 방법:**
-
-```text
-# ④ 항상 실행 (Haiku)
-Task(subagent_type="ui-stack-implementer",
-  prompt="Provide implementation guidelines for project '{project}' using {detected_stack}. Use core_ui_stack_search for framework-specific patterns.")
-
-# ⑤ 조건부 실행 (Haiku) — SPEC에 차트/대시보드/시각화 키워드 포함 시
-Task(subagent_type="ui-dataviz-advisor",
-  prompt="Recommend data visualization approach for project '{project}'. Use core_ui_search for chart types and react-performance patterns.")
-```
-
-**디자인 시스템 자동 참조:**
-- `.claude/vibe/design-system/{project}/MASTER.md` 존재 시 자동 로드
-- 구현 에이전트가 CSS 변수, 폰트, 색상 팔레트를 직접 참조
-- 페이지별 오버라이드 `pages/{page}.md` 존재 시 우선 적용
-
-### Phase Execution Flow (ULTRAWORK Pipeline)
-
-```
-Phase N Start
-    │
-    ├─→ [PARALLEL] Task(haiku) × 3: Exploration
-    │       - Related code analysis
-    │       - Dependency check
-    │       - Pattern discovery
-    │
-    ↓ (all complete)
-    │
-    ├─→ Opus: Synthesize and decide
-    │
-    ├─→ [PARALLEL PIPELINE] ←── KEY SPEED OPTIMIZATION
-    │       │
-    │       ├─→ Main: Task(sonnet) Implementation
-    │       │
-    │       └─→ Background (run_in_background=true):
-    │               ├─→ Task(haiku): Phase N+1 file analysis
-    │               ├─→ Task(haiku): Test case preparation
-    │               └─→ Task(haiku): Type/interface lookup
-    │
-    ↓ (main completes)
-    │
-    ├─→ Task(haiku): Tests (uses pre-generated cases)
-    │
-    ↓
-Phase N Complete
-    │
-    ↓ (Background results ready - NO WAIT for Phase N+1 exploration!)
-    │
-Phase N+1 Start (IMMEDIATE - exploration already done!)
-```
-
-**Speed Comparison:**
-
-| Mode | Phase Time | 5 Phases Total |
-|------|------------|----------------|
-| Sequential | ~2min/phase | ~10min |
-| Parallel Exploration | ~1.5min/phase | ~7.5min |
-| **ULTRAWORK Pipeline** | **~1min/phase** | **~5min** |
-
-**Why Pipeline is Faster:**
-- Background agents prepare next phase WHILE current phase implements
-- No idle time between phases
-- Test cases pre-generated during implementation
-- Cache stays warm across parallel tasks
-
----
-
-### Agent Teams — Dev Team
-
-> **팀 정의**: `agents/teams/dev-team.md` 참조
-> 설정: `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` + `teammateMode: in-process` (`~/.claude/settings.json` 전역 — postinstall 자동 설정)
-
-**활성화 조건 (Dev Team Full — 4명):**
-- ULTRAWORK 모드 + 3개 이상 시나리오
-- 또는 복잡도 점수 20+ (High)
-
-### Agent Teams — Lite Team (Normal Mode)
-
-> **팀 정의**: `agents/teams/lite-team.md` 참조
-
-**활성화 조건 (Lite Team — 3명):**
-- 일반 모드 + 3개 이상 시나리오
-- 복잡도 점수 8-19 (Medium)
-- 단순 구현(1-2 파일, 시나리오 2개 이하)에서는 기존 병렬 모드 유지
-
-**팀 선택 기준:**
-
-| 조건 | 팀 |
-|------|-----|
-| 시나리오 1-2개, 파일 1-2개 | 기존 병렬 모드 (팀 없음) |
-| 시나리오 3개+, 일반 모드 | **Lite Team (3명)** |
-| ULTRAWORK 또는 복잡도 20+ | Dev Team Full (4명) |
-
-### Agent Teams — Review Team
-
-> **팀 정의**: `agents/teams/review-debate-team.md` 참조
-
-**활성화 조건:**
-
-- `/vibe.review` 실행 후 P1 또는 P2 이슈 2개 이상 발견 시
-- Agent Teams 환경변수 활성화 상태
-
-### Agent Teams — Debug Team
-
-> **팀 정의**: `agents/teams/debug-team.md` 참조
-
-**활성화 조건:**
-
-- 동일 빌드/테스트 실패 3회 이상
-- UltraQA `architecture_question` 상태 진입 시
-
-### Agent Teams — Research Team
-
-> **팀 정의**: `agents/teams/research-team.md` 참조
-
-**활성화 조건:**
-
-- `/vibe.spec` Step 3 리서치 단계
-- Agent Teams 환경변수 활성화 상태
-
-### Agent Teams — Security Team
-
-> **팀 정의**: `agents/teams/security-team.md` 참조
-
-**활성화 조건:**
-
-- auth, payment, user-data, crypto 관련 파일 변경 감지 시
-- 또는 수동으로 `security` 키워드 지정 시
-
-### Agent Teams — Migration Team
-
-> **팀 정의**: `agents/teams/migration-team.md` 참조
-
-**활성화 조건:**
-
-- package.json 주요 의존성 버전 변경 감지 시
-- 또는 수동으로 `migration` 키워드 지정 시
-
-### Agent Teams — Fullstack Team
-
-> **팀 정의**: `agents/teams/fullstack-team.md` 참조
-
-**활성화 조건:**
-
-- SPEC에 frontend + backend 파일이 모두 포함된 경우
-- 또는 수동으로 `fullstack` 키워드 지정 시
-
----
-
-1. **Related code analysis**: Task(haiku) explores `<context>` related code
-2. **File creation/modification**: Task(sonnet) implements per `<output_format>`
-3. **Constraint compliance**: Check `<constraints>`
-4. **Run verification**: Execute verification commands
-
-### 4. Brand Assets Generation (Optional)
-
-When starting a **new project** with brand context in SPEC, auto-generate app icons and favicons:
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎨 BRAND ASSETS GENERATION
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-[Check] Brand assets exist? → Skip if favicon.ico exists
-[Check] Gemini API configured? → Required for image generation
-[Check] SPEC has brand context? → Extract app name, colors, style
-
-[Generate] Creating app icon with Gemini Image API...
-  - Prompt: "App icon for [AppName], [style], [color]..."
-  - Generated: 512x512 master icon
-
-[Resize] Creating platform variants...
-  ✅ favicon.ico (16/32/48)
-  ✅ favicon-16x16.png
-  ✅ favicon-32x32.png
-  ✅ apple-touch-icon.png (180x180)
-  ✅ android-chrome-192x192.png
-  ✅ android-chrome-512x512.png
-  ✅ site.webmanifest
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-✅ Brand assets generated in public/
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-```
-
-**SPEC Brand Context Example:**
-
-```xml
-<context>
-Brand:
-  - App Name: MyApp
-  - Primary Color: #2F6BFF
-  - Style: Modern, minimalist, flat design
-  - Icon Concept: Abstract geometric shape
-</context>
-```
-
-**Trigger Conditions:**
-- First `/vibe.run` execution (no existing icons)
-- SPEC contains brand/design context
-- Gemini API key configured (`vibe gemini key <key>`)
-
-**Manual Generation:**
-```bash
-# [LLM_SCRIPT] = {{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js
-node "[LLM_SCRIPT]" gemini image "App icon for MyApp, primary color #2F6BFF, square format 1:1, simple recognizable design, works well at small sizes, no text or letters, solid or gradient background, modern minimalist" --output "./public/app-icon.png"
-```
-
----
-
-### 5. Race Code Review (GPT + Gemini) + Auto-Fix (v2.6.9)
-
-After all scenarios are implemented, **GPT and Gemini review in parallel with cross-validation**:
-
-> **ULTRAWORK Default**: In ULTRAWORK mode, race review is automatically enabled.
-
-```
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🏁 RACE CODE REVIEW (GPT + Gemini)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-[Step 1] Parallel review execution...
-  ├─ GPT Codex: Reviewing...
-  └─ Gemini: Reviewing...
-
-[Step 2] Cross-validation results:
-  ┌──────────────────────────────────────────────────────────────────┐
-  │ Issue                          │ GPT │ Gemini │ Codex │ Confidence│
-  │────────────────────────────────│─────│────────│───────│───────────│
-  │ Timing attack in password      │ ✅  │ ✅     │ ✅    │ 100% → P1 │
-  │ Rate limiting missing          │ ✅  │ ✅     │ ✅    │ 100% → P1 │
-  │ Magic number usage             │ ✅  │ ❌     │ ❌    │ 50% → P2  │
-  └──────────────────────────────────────────────────────────────────┘
-
-  Summary: 3 issues (P1: 2, P2: 1)
-
-[Step 3] Auto-fixing P1/P2 issues...
-  ✅ auth.service.ts:24 - Applied timingSafeEqual (P1)
-  ✅ auth.controller.ts:15 - Added rate limiter (P1)
-  ✅ auth.service.ts:42 - Extracted constant (P2)
-
-[Step 4] Re-verifying...
-  ✅ Build succeeded
-  ✅ Tests passed
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-✅ Race review complete! 3 improvements (2 P1, 1 P2)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-```
-
-**Race Review Invocation (GPT + Gemini in parallel via Bash):**
-
-**🚨 Use --input file to avoid CLI argument length limits and Windows pipe issues.**
-
-1. Save code to review into `[SCRATCHPAD]/review-code.txt` (using Write tool)
-2. Write JSON input file `[SCRATCHPAD]/review-input.json` (using Write tool):
-   - `{"prompt": "Review this code for security, performance, and best practices. Return JSON: {issues: [{id, title, description, severity, suggestion}]}. Code: [CODE_CONTENT]"}`
-   - Where `[CODE_CONTENT]` is the code text (properly JSON-escaped inside the prompt string)
-3. Script path: `[LLM_SCRIPT]` = `{{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js`
-4. Run GPT + Gemini in PARALLEL (two Bash tool calls at once):
-
-```bash
-# GPT review (Bash tool call 1)
-node "[LLM_SCRIPT]" gpt orchestrate-json --input "[SCRATCHPAD]/review-input.json"
-```
-
-```bash
-# Gemini review (Bash tool call 2 - run in parallel)
-node "[LLM_SCRIPT]" gemini orchestrate-json --input "[SCRATCHPAD]/review-input.json"
-```
-
-**Confidence-based Priority:**
-
-| Confidence | Priority | Action |
-|------------|----------|--------|
-| 100% (3/3 or 2/2) | P1 | Auto-fix immediately |
-| 67% (2/3) | P1 | Auto-fix immediately |
-| 50% (1/2) or 33% (1/3) | P2 | Auto-fix with review |
-
-**Fallback handling:**
-- If one LLM fails → Use remaining LLM results (reduced confidence)
-- If all fail → Skip and proceed (log warning)
-
-**Review application rules:**
-
-| Feedback Type | Action |
-|---------------|--------|
-| Security vulnerability (P1) | Auto-fix immediately |
-| Performance improvement (P1/P2) | Auto-fix immediately |
-| Best practices (P2) | Auto-fix |
-| Style/preference (P3) | Apply selectively |
-
-**Conditions:**
-- **ULTRAWORK**: Race review enabled by default
-- **Normal mode**: Use `--race` flag to enable
-- Must re-verify build/tests after fixes
-
-### Codex Code Review (Codex 플러그인 활성화 시)
-
-GPT+Gemini race와 **동시에** Codex review 실행:
-
-```
-/codex:review
-```
-
-결과를 race review 교차 검증에 포함 — 3중 리뷰:
-
-```markdown
-| Issue | GPT | Gemini | Codex | Confidence |
-|-------|-----|--------|-------|------------|
-| {이슈} | ✅/❌ | ✅/❌ | ✅/❌ | {%} |
-```
-
-### 6. Quality Report (Auto-generated)
-
-After all scenarios complete + Gemini review, **quality report is auto-generated**:
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  📊 QUALITY REPORT: login                                        │
-├─────────────────────────────────────────────────────────────────┤
-│                                                                 │
-│  ✅ Scenarios: 4/4 passed                                       │
-│                                                                 │
-│  ┌───────────────────────────────────────────────────────────┐  │
-│  │ # │ Scenario                  │ Status │ Retries │        │  │
-│  │───│───────────────────────────│────────│─────────│        │  │
-│  │ 1 │ Valid login success       │ ✅     │ 0       │        │  │
-│  │ 2 │ Invalid password error    │ ✅     │ 1       │        │  │
-│  │ 3 │ Email format validation   │ ✅     │ 0       │        │  │
-│  │ 4 │ Password reset link       │ ✅     │ 0       │        │  │
-│  └───────────────────────────────────────────────────────────┘  │
-│                                                                 │
-│  📈 Quality score: 94/100                                       │
-│                                                                 │
-│  ┌─────────────────────────────────────────────────────────┐    │
-│  │ Item              │ Result │ Notes                       │    │
-│  │───────────────────│────────│─────────────────────────────│    │
-│  │ Build             │ ✅     │ npm run build succeeded     │    │
-│  │ Tests             │ ✅     │ 12/12 passed                │    │
-│  │ Type check        │ ✅     │ 0 errors                    │    │
-│  │ Complexity        │ ✅     │ All functions ≤30 lines     │    │
-│  │ Security          │ ✅     │ 0 vulnerabilities           │    │
-│  │ Race review       │ ✅     │ 3 improvements applied      │    │
-│  └─────────────────────────────────────────────────────────┘    │
-│                                                                 │
-│  ⏱️ Started: {start_time}                                        │
-│  ⏱️ Completed: {getCurrentTime 결과}                             │
-│                                                                 │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-**What users should check**:
-- Scenario pass rate (4/4 = 100%)
-- Quality score (94/100)
-- Build/test status
-
-**This alone is enough to trust quality.**
-
-### 7. Update Feature File
-
-Auto-update scenario status:
-
-```markdown
-## Coverage
-
-| Scenario | SPEC AC | Status |
-|----------|---------|--------|
-| Valid login success | AC-1 | ✅ |
-| Invalid password error | AC-2 | ✅ |
-| Email format validation | AC-3 | ✅ |
-| Password reset link | AC-4 | ✅ |
-
-**Last verified**: 2024-01-15 14:32
-**Quality score**: 94/100
-```
-
-## TRUST 5 Principles
-
-Follow during implementation:
-
-| Principle | Description |
-|-----------|-------------|
-| **T**est-first | Write tests first |
-| **R**eadable | Clear code |
-| **U**nified | Consistent style |
-| **S**ecured | Consider security |
-| **T**rackable | Logging, monitoring |
-
-## Input
-
-- `.claude/vibe/specs/{feature-name}.md` (PTCF SPEC)
-- `.claude/vibe/features/{feature-name}.feature` (BDD)
-- `CLAUDE.md` (project context)
-
-## Output
-
-- Implemented code files
-- Test files
-- Updated SPEC document (checkmarks)
-
-## Example
-
-```
-User: /vibe.run "login"
-
-Claude:
-📄 Loading SPEC: .claude/vibe/specs/login.md
-📄 Loading Feature: .claude/vibe/features/login.feature
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-📋 Scenarios to Implement
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-| # | Scenario | Status |
-|---|----------|--------|
-| 1 | Valid login success | ⬜ |
-| 2 | Invalid password error | ⬜ |
-| 3 | Email format validation | ⬜ |
-| 4 | Password reset link | ⬜ |
-
-Total: 4 scenarios
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 1/4: Valid login success
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Given: User is registered
-When: Login with valid email and password
-Then: Login success + JWT token returned
-
-⚡ [PARALLEL] Exploring...
-✅ Exploration complete (2.1s)
-
-🔨 Implementing...
-  ✅ auth.service.ts - Added login()
-  ✅ auth.controller.ts - POST /login
-
-🔍 Verifying...
-  ✅ Given: OK
-  ✅ When: OK
-  ✅ Then: OK
-
-✅ Scenario 1 passed!
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 2/4: Invalid password error
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-🔨 Implementing...
-  ✅ auth.service.ts - Password validation logic
-
-🔍 Verifying...
-  ✅ Given: OK
-  ✅ When: OK
-  ❌ Then: "Invalid credentials" error message
-     Actual: "Error" returned
-
-🔄 Auto-fix [iteration 1]...
-  Fix: auth.service.ts line 42
-
-🔍 Re-verifying...
-  ✅ Then: OK
-
-✅ Scenario 2 passed! (1 fix)
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 3/4: Email format validation
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-🔨 Implementing...
-🔍 Verifying...
-✅ Scenario 3 passed!
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🎯 Scenario 4/4: Password reset link
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-🔨 Implementing...
-🔍 Verifying...
-✅ Scenario 4 passed!
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🔍 GEMINI CODE REVIEW
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-📤 Sending code to Gemini...
-📝 Gemini feedback:
-  1. [Security] Need timing attack prevention → Fixing...
-  2. [Performance] Unnecessary DB call → Fixing...
-
-✅ 2 improvements auto-applied
-🔍 Re-verifying... ✅ Passed
-
-┌─────────────────────────────────────────────────────────────────┐
-│  📊 QUALITY REPORT: login                                        │
-├─────────────────────────────────────────────────────────────────┤
-│                                                                 │
-│  ✅ Scenarios: 4/4 passed                                       │
-│                                                                 │
-│  | # | Scenario              | Status | Retries |               │
-│  |---|───────────────────────|───────|─────────|               │
-│  | 1 | Valid login success   | ✅    | 0       |               │
-│  | 2 | Invalid password error| ✅    | 1       |               │
-│  | 3 | Email format validation| ✅   | 0       |               │
-│  | 4 | Password reset link   | ✅    | 0       |               │
-│                                                                 │
-│  📈 Quality score: 94/100                                       │
-│  Build: ✅ | Tests: ✅ | Types: ✅ | Gemini: ✅ (2 applied)     │
-│                                                                 │
-│  ⏱️ Started: {start_time}                                        │
-│  ⏱️ Completed: {getCurrentTime 결과}                             │
-└─────────────────────────────────────────────────────────────────┘
-
-🎉 Implementation complete! All scenarios passed + Gemini review applied.
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-🔍 AUTO REVIEW (13+ Agents)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-⚡ [PARALLEL] 13 expert agents reviewing...
-  - security-reviewer ✅
-  - performance-reviewer ✅
-  - architecture-reviewer ✅
-  - ...
-
-📋 Review results:
-  - P1 Critical: 0
-  - P2 Important: 2
-  - P3 Nice-to-have: 1
-
-🔧 Auto-fixing P2 issues...
-  1. [PERF] N+1 query → Fixed
-  2. [ARCH] Circular dependency → Fixed
-
-✅ Auto Review complete! 2 issues auto-resolved.
-```
-
-### Phase-specific Execution
-
-```
-User: /vibe.run "brick-game" --phase 2
-
-Claude:
-📄 Reading SPEC: .claude/vibe/specs/brick-game.md
-🎯 Executing Phase 2 only.
-
-Phase 2: Game Logic
-1. [ ] Paddle movement implementation
-2. [ ] Ball physics engine
-3. [ ] Brick collision handling
-4. [ ] Score system
-5. [ ] Game over conditions
-
-⚡ Launching parallel exploration...
-[Task(haiku) × 3 launched in parallel]
-
-🚀 Starting implementation...
-```
-
-## Error Handling
-
-On failure:
-1. Check error message
-2. Review `<constraints>`
-3. Fix code and retry
-4. If continues to fail, report to user
-
----
-
-## Quality Gate (Mandatory)
-
-### Implementation Quality Checklist
-
-Before marking any scenario as complete, ALL items must pass:
-
-| Category | Check Item | Weight |
-|----------|------------|--------|
-| **Functionality** | All Given/When/Then conditions verified | 20% |
-| **Functionality** | Edge cases handled per scenario | 10% |
-| **Code Quality** | No `any` types in TypeScript | 10% |
-| **Code Quality** | Functions ≤30 lines, nesting ≤3 levels | 10% |
-| **Code Quality** | No hardcoded values (use constants) | 5% |
-| **Security** | Input validation implemented | 10% |
-| **Security** | Authentication/authorization checked | 5% |
-| **Error Handling** | Try-catch or error states present | 10% |
-| **Error Handling** | User-friendly error messages | 5% |
-| **Testing** | Unit tests exist for core logic | 10% |
-| **Performance** | No N+1 queries or unnecessary loops | 5% |
-
-### Quality Score Calculation
-
-```
-Score = Σ(checked items × weight) / 100
-
-Grades:
-- 95-100: ✅ EXCELLENT - Ready to merge
-- 90-94:  ⚠️ GOOD - Minor improvements required before merge
-- 80-89:  ⚠️ FAIR - Significant improvements required
-- 0-79:   ❌ POOR - Major fixes needed
-```
-
-### Quality Gate Thresholds
-
-| Gate | Minimum Score | Condition |
-|------|---------------|-----------|
-| **Scenario Complete** | 95 | Each scenario must score ≥95 |
-| **Phase Complete** | 95 | Average of all scenarios ≥95 |
-| **Feature Complete** | 95 | All phases complete + Gemini review |
-
-### Auto-Fix Triggers
-
-| Issue Type | Auto-Fix Action |
-|------------|-----------------|
-| Missing error handling | Add try-catch wrapper |
-| Hardcoded values | Extract to constants file |
-| Missing input validation | Add validation schema |
-| Function too long | Suggest split points |
-| N+1 query detected | Add eager loading |
-
-### Auto-Fix 실패 시 Codex Rescue (Codex 플러그인 활성화 시)
-
-P1 auto-fix가 **3회 실패** 시, Codex에 위임:
-
-```
-/codex:rescue "Fix P1 issue: {issue-description}. File: {file-path}. Error: {error-message}"
-```
-
-Codex 수정 완료 후 재검증. Codex도 실패 시 TODO 파일에 기록.
-
-### Forbidden Patterns (Block Merge)
-
-| Pattern | Why Forbidden | Detection |
-|---------|---------------|-----------|
-| `console.log` | Debug code in production | Regex scan |
-| `// TODO` without issue | Untracked work | Comment scan |
-| `any` type | Type safety bypass | TypeScript check |
-| `@ts-ignore` | Type error suppression | TypeScript check |
-| Empty catch blocks | Silent error swallowing | AST analysis |
-| Commented-out code | Dead code | Comment scan |
-
----
-
-## Auto-Retrospective (Post-Implementation)
-
-After ALL phases complete successfully, **automatically** perform a brief retrospective:
-
-### Retrospective Template
-
-```
-## Retrospective: {feature-name}
-
-### What Worked
-- [List effective patterns, tools, approaches used]
-
-### What Didn't
-- [List issues, failures, unexpected blockers]
-
-### Key Decisions
-- [Important architectural or implementation decisions made during this run]
-
-### Lessons Learned
-- [Principle format: "When X, do Y because Z"]
-```
-
-### Execution Steps
-
-1. Generate retrospective based on the implementation session
-2. Save to `.claude/vibe/retros/{feature-name}.md`
-3. Save key lessons via `core_save_memory` (for cross-session recall)
-4. Update `claude-progress.txt` with final status
-
-**Important:**
-
-- Keep it concise (under 20 lines)
-- Focus on **project-specific** insights, not generic knowledge
-- Only save to memory if the lesson is actionable and non-obvious
-
----
-
-## Next Step
-
-```
-/vibe.verify "brick-game"
-```
-
----
-
-ARGUMENTS: $ARGUMENTS
+---
+description: Execute implementation from SPEC
+argument-hint: "feature name" or --phase N
+---
+
+# /vibe.run
+
+Execute **Scenario-Driven Implementation** with automatic quality verification.
+
+> **Core Principle**: Scenarios are both the implementation unit and verification criteria. All scenarios passing = Quality guaranteed.
+
+## Usage
+
+```
+/vibe.run "feature-name"              # Full implementation
+/vibe.run "feature-name" --phase 1    # Specific Phase only
+/vibe.run "feature-name" ultrawork    # ULTRAWORK mode (recommended)
+/vibe.run "feature-name" ulw          # Short alias for ultrawork
+```
+
+---
+
+## Codex Plugin Integration
+
+> **Codex 플러그인 감지**: 워크플로우 시작 시 아래 명령으로 자동 감지.
+>
+> ```bash
+> CODEX_AVAILABLE=$(node "{{VIBE_PATH}}/hooks/scripts/codex-detect.js" 2>/dev/null || echo "unavailable")
+> ```
+>
+> `available`이면 `/codex:rescue` (구현 위임), `/codex:review` (코드 리뷰) 자동 호출. `unavailable`이면 기존 워크플로우로 동작.
+
+---
+
+> **⏱️ Timer**: Call `getCurrentTime` tool at the START. Record the result as `{start_time}`.
+
+## File Reading Policy (Mandatory)
+
+- **SPEC/Feature 파일**: 반드시 `Read` 도구로 전체 파일을 읽을 것 (Grep 금지)
+- **소스코드 파일**: 구현/수정 대상 파일은 반드시 `Read` 도구로 전체 읽은 후 작업할 것
+- **Grep 사용 제한**: 파일 위치 탐색(어떤 파일에 있는지 찾기)에만 사용. 파일 내용 파악에는 반드시 Read 사용
+- **에이전트 spawn 시**: 프롬프트에 "대상 파일을 Read 도구로 전체 읽은 후 구현하라"를 반드시 포함할 것
+- **부분 읽기 금지**: Grep 결과의 주변 몇 줄만 보고 수정하지 말 것. 전체 맥락을 파악해야 기존 코드와 일관된 구현 가능
+
+## **Scenario-Driven Development (SDD)**
+
+> Automate **Scenario = Implementation = Verification** so even non-developers can trust quality
+
+### Pre-Run Regression Check (MANDATORY, before implementation starts)
+
+시작 직후 필수 실행:
+
+```
+Load skill `vibe-regress` with: list --feature "{feature-name}"
+```
+
+- 미해결 회귀 항목 있으면:
+  - interactive 모드: 사용자에게 "먼저 회귀 테스트 생성 후 진행?" 묻기
+  - ultrawork 모드: 자동 `/vibe.regress generate <slug>`로 예방 테스트 생성 후 진행
+- 미해결 없으면 조용히 통과
+
+또한 `.claude/vibe/contracts/{feature-name}.md`이 있으면 로드 — 구현 시 계약 준수 기준으로 사용.
+
+### Core Flow
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│                    SCENARIO-DRIVEN IMPLEMENTATION                │
+│                                                                  │
+│   Load Feature file                                              │
+│        ↓                                                        │
+│   ┌──────────────────────────────────────────────────────────┐  │
+│   │ Scenario 1: Happy Path                                    │  │
+│   │   Given → When → Then                                     │  │
+│   │        ↓                                                  │  │
+│   │   [Implement] → [Verify immediately] → ✅ Pass            │  │
+│   └──────────────────────────────────────────────────────────┘  │
+│        ↓                                                        │
+│   ┌──────────────────────────────────────────────────────────┐  │
+│   │ Scenario 2: Edge Case                                     │  │
+│   │   Given → When → Then                                     │  │
+│   │        ↓                                                  │  │
+│   │   [Implement] → [Verify] → ❌ Fail → [Fix] → ✅ Pass      │  │
+│   └──────────────────────────────────────────────────────────┘  │
+│        ↓                                                        │
+│   ┌──────────────────────────────────────────────────────────┐  │
+│   │ Scenario N: ...                                           │  │
+│   │   [Implement] → [Verify immediately] → ✅ Pass            │  │
+│   └──────────────────────────────────────────────────────────┘  │
+│        ↓                                                        │
+│   ┌──────────────────────────────────────────────────────────┐  │
+│   │  📊 QUALITY REPORT                                        │  │
+│   │  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━   │  │
+│   │  Scenarios: 5/5 passed ✅                                 │  │
+│   │  Quality score: 94/100                                    │  │
+│   │  Build: ✅ | Tests: ✅                                    │  │
+│   └──────────────────────────────────────────────────────────┘  │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+### Scenario = Implementation Unit
+
+**Traditional approach (Phase-based)**:
+```
+Phase 1 → Phase 2 → Phase 3 → ... → Verify at the end
+                                      ↓
+                              "Where did it go wrong?"
+```
+
+**SDD approach (Scenario-based)**:
+```
+Scenario 1 → Implement → Verify ✅
+Scenario 2 → Implement → Verify ✅
+Scenario 3 → Implement → Verify ❌ → Fix → ✅
+...
+All pass = Quality guaranteed
+```
+
+### Automated Verification (Closed Loop)
+
+**자율적 AI 코딩 = 구현 + 검증 + 반복. 검증을 AI에게 맡기는 순간 루프가 닫힌다.**
+
+After implementing each scenario, **automatic verification**:
+
+| Verification Item | Auto Check | Method |
+|-------------------|------------|--------|
+| Given (precondition) | State/data preparation confirmed | Code analysis |
+| When (action) | Feature execution possible | Code analysis + Build |
+| Then (result) | Expected result matches | Code analysis + Test |
+| Code quality | Complexity, style, security | Static analysis |
+| **UI behavior** | **실제 브라우저에서 동작 확인** | **E2E Closed Loop** |
+
+### E2E Closed Loop (UI Scenarios)
+
+**UI 시나리오가 포함된 Feature일 때 자동 활성화.**
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  CLOSED LOOP = 구현 + 검증 + 반복                                 │
+│                                                                  │
+│  Scenario 구현 완료                                               │
+│       ↓                                                          │
+│  [E2E 검증] AI가 직접 브라우저 조작                                │
+│       │   (접근성 트리 기반 — 토큰 효율적)                         │
+│       ├─ PASS → Next scenario                                    │
+│       └─ FAIL → Root cause analysis                              │
+│              → Fix code (Read full file first)                   │
+│              → Re-run failed scenario only                       │
+│              → PASS? → Next scenario                             │
+│              → FAIL (same as prev)? → STUCK → Ask user           │
+│                                                                  │
+│  핵심: 검증이 가벼울수록 루프는 더 많이 돈다                       │
+│  - 접근성 트리: button "Sign In" = 15 chars                      │
+│  - DOM 트리: div class="nav-wrapper mx-4..." = 200+ chars        │
+│  - 전자를 써야 한 세션에서 수십 개 시나리오 검증 가능               │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+**Browser Tool Priority:**
+
+| Priority | Tool | 용도 |
+|----------|------|------|
+| 1st | Agent Browser (접근성 트리) | AI 직접 조작, 최소 토큰 |
+| 2nd | Playwright Test Runner | 테스트 코드 실행, pass/fail 반환 |
+| 3rd | Playwright MCP (DOM) | 최후 수단, 토큰 비효율 |
+
+**활성화 조건:**
+- Feature 파일에 UI 관련 시나리오 존재 (form, button, page, navigate 등)
+- `.claude/vibe/e2e/config.json`의 `closedLoop.enabled: true` (기본값)
+- dev server가 실행 중 (`baseURL` 접근 가능)
+
+### Auto-Fix on Failure
+
+```
+Scenario verification failed (코드 분석 또는 E2E)
+      ↓
+[Collect evidence]
+  - E2E: screenshot, console errors, accessibility tree snapshot
+  - Code: build errors, test failures, type errors
+      ↓
+[Root cause analysis] - Which Then condition failed?
+      ↓
+[Read target file FULLY] - Grep으로 훑지 말 것
+      ↓
+[Implement fix] - Fix only that part
+      ↓
+[Re-verify] - Re-run ONLY failed scenario (save tokens)
+      ↓
+Repeat until pass (라운드 수 캡 없음 — stuck 감지로 종료)
+```
+
+**Termination conditions:**
+- ✅ PASS → 다음 scenario
+- ⚠️ Stuck (같은 failure가 이전 라운드와 동일) → 사용자 질문:
+    1. 직접 수정 방법 제공 (예: "LoginForm.tsx 42줄 수정") → 재시도
+    2. "proceed" — 남은 실패 TODO 기록 후 다음 scenario
+    3. "abort" — 구현 중단
+- `ultrawork` 모드: stuck 시 프롬프트 없이 TODO 기록 후 다음 scenario
+
+---
+
+## **ULTRAWORK Mode** (ulw)
+
+> Include `ultrawork` or `ulw` in your command to activate **maximum performance mode**.
+
+### What ULTRAWORK Enables
+
+When you include `ultrawork` (or `ulw`), ALL of these activate automatically:
+
+| Feature | Description |
+|---------|-------------|
+| **Parallel Exploration** | 3+ Task(haiku) agents run simultaneously |
+| **Boulder Loop** | Auto-continues until ALL phases complete |
+| **Context Compression** | Aggressive auto-save at 70%+ context |
+| **No Pause** | Doesn't wait for confirmation between phases |
+| **External LLMs** | Auto-consults GPT/Gemini if enabled |
+| **Error Recovery** | Loops until 100% or stuck; on stuck auto-records TODO and proceeds (no user prompt) |
+| **Race Review (v2.6.9)** | Multi-LLM review (GPT+Gemini) with cross-validation |
+
+### Boulder Loop (Inspired by Sisyphus)
+
+Like Sisyphus rolling the boulder, ULTRAWORK **keeps going until done**:
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│                    BOULDER LOOP (ultrawork)                      │
+│                                                                  │
+│   ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐  │
+│   │ Phase 1  │───→│ Phase 2  │───→│ Phase 3  │───→│ Phase N  │  │
+│   └──────────┘    └──────────┘    └──────────┘    └──────────┘  │
+│        │               │               │               │         │
+│        ↓               ↓               ↓               ↓         │
+│   [Parallel]      [Parallel]      [Parallel]      [Parallel]    │
+│   [Implement]     [Implement]     [Implement]     [Implement]   │
+│   [Test]          [Test]          [Test]          [Test]        │
+│        │               │               │               │         │
+│        └───────────────┴───────────────┴───────────────┘         │
+│                              │                                   │
+│                              ↓                                   │
+│                     ┌──────────────┐                             │
+│                     │  ALL DONE?   │                             │
+│                     └──────────────┘                             │
+│                       │         │                                │
+│                      NO        YES                               │
+│                       │         │                                │
+│                       ↓         ↓                                │
+│                   [Continue]  [🎉 Complete!]                     │
+│                                                                  │
+│   NO STOPPING until acceptance criteria met or error limit hit   │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+### Ralph Loop (Completion Verification) - CRITICAL
+
+> **Inspired by [ghuntley.com/ralph](https://ghuntley.com/ralph)**: "Deterministically bad in an undeterministic world" - Keep iterating until TRULY complete.
+
+**Problem**: AI often claims "complete" when implementation is partial.
+
+**Solution**: RTM-based automated coverage verification with iteration tracking.
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│                    RALPH LOOP (Mandatory)                        │
+│                                                                  │
+│   After ALL phases complete:                                     │
+│                                                                  │
+│   ┌──────────────────────────────────────────────────────────┐  │
+│   │  RTM COVERAGE VERIFICATION [Iteration {{ITER}}/{{MAX}}]   │  │
+│   │                                                           │  │
+│   │  Generate RTM via core tools:                             │  │
+│   │  → generateTraceabilityMatrix("{feature-name}")           │  │
+│   │                                                           │  │
+│   │  Coverage Metrics (automated):                            │  │
+│   │  □ Requirements coverage: {coveragePercent}%              │  │
+│   │  □ SPEC → Feature mapping: {featureCovered}/{total}       │  │
+│   │  □ Feature → Test mapping: {testCovered}/{total}          │  │
+│   │  □ Build successful?                                      │  │
+│   │  □ Tests passing?                                         │  │
+│   │                                                           │  │
+│   │  UNCOVERED: {uncoveredRequirements[]}                     │  │
+│   └──────────────────────────────────────────────────────────┘  │
+│                              │                                   │
+│                   ┌──────────┴──────────┐                       │
+│                   │  Coverage == 100%?  │                       │
+│                   └──────────┬──────────┘                       │
+│                       │              │                          │
+│                      NO            YES                          │
+│                       │              │                          │
+│                       ↓              ↓                          │
+│              ┌────────────────┐  ┌────────────────┐             │
+│              │ IMPLEMENT      │  │ ✅ TRULY DONE  │             │
+│              │ UNCOVERED      │  │                │             │
+│              │ REQUIREMENTS   │  │ Report final   │             │
+│              │ (auto-extract) │  │ RTM coverage   │             │
+│              └───────┬────────┘  └────────────────┘             │
+│                      │                                          │
+│                      └──────────→ [Re-generate RTM]             │
+│                                                                  │
+│                      │                                          │
+│                      ↓                                          │
+│              Stuck? (coverage unchanged from prev iteration)    │
+│                      │                                          │
+│                      ├─ Interactive: Ask user                   │
+│                      │     1. Provide resolution → retry        │
+│                      │     2. "proceed" → TODO + done           │
+│                      │     3. "abort" → stop                    │
+│                      └─ ultrawork: TODO + done                  │
+│                                                                  │
+│   NO iteration cap — loop until 100% OR stuck                   │
+│   ZERO TOLERANCE for silent scope reduction                     │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+**Ralph Loop with RTM:**
+
+```bash
+# Generate RTM for coverage verification
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.generateTraceabilityMatrix('{feature-name}', {projectPath: process.cwd()}).then(r => console.log(JSON.stringify(r, null, 2))))"
+```
+
+**RTM provides automated metrics:**
+
+| Metric | Description |
+|--------|-------------|
+| `totalRequirements` | Total REQ-* items in SPEC |
+| `specCovered` | Requirements with SPEC mapping |
+| `featureCovered` | Requirements with Feature scenarios |
+| `testCovered` | Requirements with test files |
+| `coveragePercent` | Overall coverage percentage |
+| `uncoveredRequirements` | List of missing REQ-* IDs |
+
+**Ralph Loop Rules:**
+
+| Rule | Description |
+|------|-------------|
+| **No Scope Reduction** | Never say "simplified" or "basic version" - implement FULL request |
+| **Iteration Tracking** | Display `[{{ITER}}]` to show progress (no max — loop until done) |
+| **RTM-Based Gap List** | Use `uncoveredRequirements` array - no manual comparison |
+| **Coverage Threshold** | Must reach 100% coverage to complete |
+| **No Iteration Cap** | Loop until 100% coverage OR stuck (convergence detected) |
+| **Stuck Handling** | If coverage % unchanged between iterations → ask user (proceed/abort/fix), or ultrawork → TODO + done |
+| **Diminishing Returns (Narrowing)** | Iteration 3+ → focus on core requirements (REQ-*-001~003) first; P2/P3 requirements continue to loop but lower priority |
+
+**Ralph Loop Output Format:**
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🔄 RALPH VERIFICATION [Iteration 1]
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📊 RTM Coverage Report: login
+
+Requirements Traceability:
+  Total Requirements: 9
+  SPEC Covered: 9/9 (100%)
+  Feature Covered: 5/9 (55%)
+  Test Covered: 4/9 (44%)
+
+  ✅ REQ-login-001: Login form UI → Scenario 1 → login.test.ts
+  ✅ REQ-login-002: Email validation → Scenario 2 → validation.test.ts
+  ✅ REQ-login-003: Password validation → Scenario 2 → validation.test.ts
+  ❌ REQ-login-004: Remember me checkbox → NOT IMPLEMENTED
+  ❌ REQ-login-005: Forgot password link → NOT IMPLEMENTED
+  ✅ REQ-login-006: API integration → Scenario 3 → api.test.ts
+  ❌ REQ-login-007: Loading state → NOT IMPLEMENTED
+  ❌ REQ-login-008: Error toast → NOT IMPLEMENTED
+  ✅ REQ-login-009: Session storage → Scenario 4 → (no test)
+
+Overall Coverage: 55% ⚠️ BELOW 100% TARGET
+
+UNCOVERED REQUIREMENTS (auto-extracted from RTM):
+  1. REQ-login-004: Remember me checkbox
+  2. REQ-login-005: Forgot password link
+  3. REQ-login-007: Loading state
+  4. REQ-login-008: Error toast notifications
+
+⚠️ NOT COMPLETE - Implementing uncovered requirements...
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🔄 RALPH VERIFICATION [Iteration 2]
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📊 RTM Coverage Report: login
+
+Requirements Traceability:
+  Total Requirements: 9
+  SPEC Covered: 9/9 (100%)
+  Feature Covered: 9/9 (100%)
+  Test Covered: 9/9 (100%)
+
+Overall Coverage: 100% ✅ TARGET REACHED
+
+Build: ✅ Passed
+Tests: ✅ 12/12 Passed
+Type Check: ✅ No errors
+
+✅ RALPH VERIFIED COMPLETE!
+
+📄 RTM saved: .claude/vibe/rtm/login-rtm.md
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**When to Trigger Ralph Loop:**
+
+1. After all phases complete
+2. Before final quality report
+3. Whenever user says "ultrawork" or "ralph"
+
+**Forbidden Responses (VIOLATIONS):**
+
+| ❌ NEVER Say | ✅ Instead |
+|-------------|-----------|
+| "I've implemented a basic version" | Implement the FULL version |
+| "This is a simplified approach" | Implement as specified |
+| "You can add X later" | Add X now |
+| "For demonstration purposes" | Implement production-ready |
+| "The core functionality is done" | ALL functionality must be done |
+
+### ULTRAWORK Example
+
+```
+User: /vibe.run "brick-game" ultrawork
+
+Claude:
+🚀 ULTRAWORK MODE ACTIVATED
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📄 SPEC: .claude/vibe/specs/brick-game.md
+🎯 4 Phases detected
+⚡ Boulder Loop: ENABLED (will continue until all phases complete)
+🔄 Auto-retry: ON (loop until 100% or stuck → auto-TODO)
+💾 Context compression: AGGRESSIVE
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🏔️ BOULDER ROLLING... Phase 1/4
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+⚡ [PARALLEL] Launching 3 exploration agents...
+✅ Exploration complete (7.2s)
+🔨 Implementing...
+✅ Phase 1 complete
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🏔️ BOULDER ROLLING... Phase 2/4
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+⚡ [PARALLEL] Launching 3 exploration agents...
+✅ Exploration complete (6.8s)
+🔨 Implementing...
+❌ Test failed: collision detection
+🔄 Auto-retry [iteration 1]...
+🔨 Fixing...
+✅ Phase 2 complete
+
+[...continues automatically...]
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎉 BOULDER REACHED THE TOP!
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+✅ All 4 phases complete
+✅ All acceptance criteria passed
+✅ Build succeeded
+✅ Tests passed
+
+⏱️ Total: 8m 24s
+📊 Retries: 2
+💾 Context saved: 3 checkpoints
+```
+
+### Normal vs ULTRAWORK Comparison
+
+| Aspect | Normal | ULTRAWORK |
+|--------|--------|-----------|
+| Phase transition | May pause | Auto-continues |
+| On error | Reports and stops | Auto-retries (3x) |
+| Context 70%+ | Warning only | Auto-compress + save |
+| Exploration | Sequential possible | FORCED parallel |
+| Completion | Phase-by-phase | Until ALL done |
+
+---
+
+## Automation Level System
+
+Magic keywords in the user input automatically set the **AutomationLevel**, which controls how much the AI self-advances vs. pausing for confirmation.
+
+### Level Definitions
+
+| Level | Name | Keyword(s) | Auto-advance | Auto-retry | Stuck Behavior | Parallel Agents | Checkpoints |
+|-------|------|------------|--------------|------------|----------------|-----------------|-------------|
+| L0 | Manual | `manual` | No | No | Ask user every step | No | All |
+| L1 | Guided | `guided`, `verify` | No | No | Ask user on stuck | No | All |
+| L2 | Semi-auto | `quick` (default) | Yes | Yes (low cap: 2) | Ask user after 2 retries | No | Key points |
+| L3 | Auto | `ultrawork`, `ulw` | Yes | Yes (no cap) | Auto-TODO + proceed | Yes | Checkpoint-only |
+| L4 | Full-auto | `ralph`, `ralplan` | Yes | Yes (no cap) | Auto-TODO + proceed | Yes | None |
+
+### Detection Rule
+
+The AI detects automation level from word boundaries in the user's input:
+
+```
+/vibe.run "login"              → L2 Semi-auto (default)
+/vibe.run "login" ultrawork    → L3 Auto
+/vibe.run "login" ralph        → L4 Full-auto
+/vibe.run "login" verify       → L1 Guided
+```
+
+### Confirmation Matrix
+
+Certain action types require user confirmation depending on the active level:
+
+| Action | L0 | L1 | L2 | L3 | L4 |
+|--------|----|----|----|----|-----|
+| `destructive` | confirm | confirm | confirm | confirm | auto |
+| `architecture_choice` | confirm | confirm | confirm | auto | auto |
+| `implementation_scope` | confirm | confirm | confirm | auto | auto |
+| `phase_advance` | confirm | confirm | auto | auto | auto |
+| `fix_strategy` | confirm | confirm | auto | auto | auto |
+| `retry` | confirm | auto | auto | auto | auto |
+
+**Rule**: When confirmation is required, pause and display a checkpoint before proceeding.
+
+---
+
+## Interactive Checkpoints
+
+Checkpoints are decision gates inserted at critical points in the workflow. At L3/L4, most checkpoints are **auto-resolved** using the default option.
+
+### Checkpoint Types
+
+| Type | When It Fires | Default Option |
+|------|--------------|----------------|
+| `requirements_confirm` | Before starting Phase 1 | Confirm (a) |
+| `architecture_choice` | When architecture approach is ambiguous | Clean/balanced (b) |
+| `implementation_scope` | Before any large scope change (6+ files) | Approve (a) |
+| `verification_result` | After each VerificationLoop iteration below threshold | Continue fixing (a) |
+| `fix_strategy` | When critical issues are found during quality gate | Fix all (a) |
+
+### Checkpoint Format
+
+When a checkpoint fires, display it in this format:
+
+```
+──────────────────────────────────────────────────
+CHECKPOINT: Requirements Confirmation
+──────────────────────────────────────────────────
+Feature: login
+
+Requirements:
+  1. User can log in with email and password
+  2. Invalid credentials show an error message
+  3. Remember me persists session for 30 days
+
+Options:
+  a) Confirm
+     Proceed with these requirements as stated.
+  b) Revise
+     Modify or clarify requirements before proceeding.
+  c) Abort
+     Cancel this workflow.
+
+Default: a
+──────────────────────────────────────────────────
+```
+
+### Auto-Resolution (L3/L4)
+
+At automation levels L3 and L4, checkpoints that do not require confirmation are auto-resolved:
+
+```
+[AUTO] CHECKPOINT: implementation_scope → option a (Approve) [auto-resolved]
+```
+
+---
+
+## Rules Reference
+
+**Must follow `~/.claude/vibe/rules/` (global):**
+
+- `core/development-philosophy.md` - Surgical precision, modify only requested scope
+- `core/quick-start.md` - Korean, DRY, SRP, YAGNI
+- `standards/complexity-metrics.md` - Functions ≤20 lines, nesting ≤3 levels
+- `quality/checklist.md` - Code quality checklist
+
+**Language guide:** `~/.claude/vibe/languages/{stack}.md` (global reference)
+
+---
+
+## Coding Guidelines (Mandatory)
+
+### Type Safety: Use Types Explicitly
+
+> **Core Principle**: Use types explicitly in every language that has a type system!
+
+Type definitions are not just language syntax — they are a **core engineering philosophy for simplifying and controlling complex software**.
+
+### Applies to ALL Typed Languages
+
+| Category | Languages | Key Principle |
+|----------|-----------|---------------|
+| **Static Typed** | Java, C#, C++, Go, Rust, Swift, Kotlin, Scala | Types = compile-time contracts |
+| **Gradual Typed** | TypeScript, Python (typing), PHP (typed), Ruby (RBS) | Types = optional safety nets |
+| **Functional** | Haskell, OCaml, F#, Elm | Types = logical proofs |
+
+### Universal Anti-Patterns (All Languages)
+
+| ❌ Forbidden Pattern | Why | ✅ Instead |
+|---------------------|-----|-----------|
+| Type escape hatches (`any`, `Any`, `Object`, `void*`, `interface{}`) | Loses type info, runtime errors | Concrete types or `unknown` + guards |
+| Type suppression (`@ts-ignore`, `# type: ignore`, `@SuppressWarnings`) | Hides errors | Fix actual type issues |
+| Raw generic types (`List`, `Map` without params) | Loses type safety | `List<User>`, `Map<String, Order>` |
+| Excessive casting (`as`, `(Type)`, `unsafe`) | Bypasses compiler | Type guards or pattern matching |
+
+### Language-Specific Guidelines
+
+**TypeScript/JavaScript:**
+```typescript
+// ❌ BAD
+function process(data: any): any { return data.foo; }
+
+// ✅ GOOD
+function process(data: unknown): Result {
+  if (isValidData(data)) return data.foo;
+  throw new Error('Invalid');
+}
+```
+
+**Python:**
+```python
+# ❌ BAD
+def process(data: Any) -> Any: return data["key"]
+
+# ✅ GOOD
+def process(data: UserData) -> str: return data["name"]
+```
+
+**Java/Kotlin:**
+```java
+// ❌ BAD
+List items = new ArrayList();  // Raw type
+Object data = getData();       // Lost type info
+
+// ✅ GOOD
+List<User> users = new ArrayList<>();
+User user = getUser();
+```
+
+**Go:**
+```go
+// ❌ BAD
+func process(data interface{}) interface{} { ... }
+
+// ✅ GOOD
+func process(data UserRequest) (UserResponse, error) { ... }
+```
+
+**Rust:**
+```rust
+// ❌ BAD (unnecessary unsafe or Box<dyn Any>)
+let data: Box<dyn Any> = get_data();
+
+// ✅ GOOD
+let data: UserData = get_data()?;
+```
+
+**C#:**
+```csharp
+// ❌ BAD
+object data = GetData();
+dynamic result = Process(data);
+
+// ✅ GOOD
+UserData data = GetData();
+Result result = Process(data);
+```
+
+### Type Safety Rules (Universal)
+
+| Rule | Description |
+|------|-------------|
+| **Boundary Validation** | Validate only at system boundaries (API, JSON, user input) |
+| **Internal Trust** | After validation, pass only precise types internally |
+| **No Type Escape** | Never use escape hatches to "fix" type errors |
+| **Explicit Signatures** | Specify types in function/method signatures |
+| **Generics with Params** | Always use generics with type parameters |
+
+### Quality Gate: Type Violations Block Merge
+
+| Violation | Action |
+|-----------|--------|
+| Type escape hatches (`any`, `Any`, `Object`, `interface{}`, etc.) | ❌ Block |
+| Type suppression comments | ❌ Block |
+| Raw generic types | ❌ Block |
+| Missing function return types | ⚠️ Warning |
+| Excessive type casting | ⚠️ Warning |
+
+## Description
+
+Read PTCF structured SPEC document and execute implementation immediately.
+
+> **PLAN, TASKS documents unnecessary** - SPEC is the executable prompt
+
+## Model Orchestration (Intelligent Routing)
+
+Automatically select optimal model based on **task complexity analysis**.
+
+### Complexity-Based Model Selection
+
+| Complexity Score | Model | When to Use |
+|------------------|-------|-------------|
+| 0-7 (Low) | **Haiku** | Simple fixes, searches, single file changes |
+| 8-19 (Medium) | **Sonnet** | Standard features, 3-5 files, integrations |
+| 20+ (High) | **Opus** | Architecture, security, multi-service, 6+ files |
+
+### Complexity Signals
+
+The following signals increase complexity score:
+
+| Signal | Score |
+|--------|-------|
+| Architecture change | +15 |
+| Security implication | +12 |
+| Multi-service | +8 |
+| Refactoring | +12 |
+| 6+ files | +15 |
+| 3-5 files | +8 |
+| New feature | +5 |
+| Bug fix | -3 |
+| Documentation | -5 |
+
+### Agent Tier System
+
+Each agent has tier variants for cost optimization:
+
+| Agent | Low (Haiku) | Medium (Sonnet) | High (Opus) |
+|-------|-------------|-----------------|-------------|
+| explorer | explorer-low | explorer-medium | explorer |
+| implementer | implementer-low | implementer-medium | implementer |
+| architect | architect-low | architect-medium | architect |
+
+### Task Calls by Role
+
+| Task Type | Model | Task Parameter |
+|-----------|-------|----------------|
+| Simple search | Haiku | `model: "haiku"` |
+| Codebase exploration | Haiku/Sonnet | Auto-selected |
+| Core implementation | Sonnet | `model: "sonnet"` |
+| Test writing | Haiku | `model: "haiku"` |
+| Architecture decisions | Opus | Main session |
+| Final review | Opus | Main session |
+
+### External LLM Usage (When Enabled)
+
+When external LLMs are enabled in `.claude/vibe/config.json`:
+
+| Role | Method | Condition |
+|------|--------|-----------|
+| User direct query | `gpt.question`, `gemini.question` | Hook auto-handles |
+| Internal orchestration | Call global script via Bash | Claude calls directly |
+
+**User questions (Hook auto-handles):**
+- `gpt.question` - GPT architecture consultation
+- `gemini.question` - Gemini Q&A/consultation
+
+**Claude internal calls (directly via Bash):**
+```bash
+# Usage: node llm-orchestrate.js <provider> <mode> [systemPrompt] [prompt]
+#   - If systemPrompt omitted, uses default
+#   - If systemPrompt is "-", uses default and treats next argument as prompt
+
+# [LLM_SCRIPT] = {{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js
+
+# GPT call (short prompt - CLI arg)
+node "[LLM_SCRIPT]" gpt orchestrate-json "[question content]"
+
+# Gemini call
+node "[LLM_SCRIPT]" gemini orchestrate-json "[question content]"
+
+# Custom system prompt usage
+node "[LLM_SCRIPT]" gpt orchestrate-json "You are a code reviewer" "[question content]"
+
+# Long prompt - use --input file (write JSON file first with Write tool)
+# JSON format: {"prompt": "your prompt here"}
+node "[LLM_SCRIPT]" gpt orchestrate-json --input "[SCRATCHPAD]/input.json"
+```
+
+### External LLM Fallback
+
+**IMPORTANT**: When GPT/Gemini hook fails, Claude MUST handle the task directly:
+
+**Fallback behavior**:
+- Do NOT retry the external LLM call
+- Claude handles the task using its own capabilities
+- Continue with the implementation without interruption
+- Log the fallback but don't block progress
+
+## Core Tools (Semantic Analysis & Memory)
+
+Use core tools for accurate codebase understanding and session continuity.
+
+### Tool Invocation
+
+All tools are called via:
+```bash
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.TOOL_NAME({...args}).then(r => console.log(r.content[0].text)))"
+```
+
+### Semantic Analysis Tools
+
+| Tool | Purpose | Usage |
+|------|---------|-------|
+| `findSymbol` | Find symbol definitions | `{symbolName: 'functionName', searchPath: '.'}` |
+| `findReferences` | Find all references | `{symbolName: 'functionName', searchPath: '.'}` |
+| `analyzeComplexity` | Analyze code complexity | `{filePath: 'src/file.ts'}` |
+| `validateCodeQuality` | Validate code quality | `{filePath: 'src/file.ts'}` |
+
+**Example - Find symbol:**
+```bash
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.findSymbol({symbolName: 'login', searchPath: '.'}).then(r => console.log(r.content[0].text)))"
+```
+
+### Memory Tools
+
+| Tool | Purpose | Usage |
+|------|---------|-------|
+| `saveMemory` | Save important decisions | `{key: 'decision-name', value: 'content', category: 'project'}` |
+| `recallMemory` | Recall saved memory | `{key: 'decision-name'}` |
+| `listMemories` | List all memories | `{category: 'project'}` |
+
+**Example - Save important decision:**
+```bash
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/tools/index.js').then(t => t.saveMemory({key: 'auth-pattern', value: 'Using JWT with refresh tokens', category: 'project'}).then(r => console.log(r.content[0].text)))"
+```
+
+### Session Management (Auto via Hooks)
+
+- **Session start**: Hook auto-calls `startSession` to restore previous context
+- **Context 80%+**: Hook auto-calls `autoSaveContext` to preserve state
+
+## Process
+
+### 1. Load SPEC + Feature
+
+**Search order (check BOTH file AND folder):**
+
+```
+Step 1: Check if SPLIT structure exists (folder)
+  📁 .claude/vibe/specs/{feature-name}/        → Folder with _index.md + phase files
+  📁 .claude/vibe/features/{feature-name}/      → Folder with _index.feature + phase files
+
+Step 2: If no folder, check single file
+  📄 .claude/vibe/specs/{feature-name}.md       → Single SPEC file
+  📄 .claude/vibe/features/{feature-name}.feature → Single Feature file
+
+Step 3: If neither exists → Error
+```
+
+**Split structure (folder) detected:**
+```
+📁 .claude/vibe/specs/{feature-name}/
+├── _index.md              → Master SPEC (read first for overview)
+├── phase-1-{name}.md      → Phase 1 SPEC
+├── phase-2-{name}.md      → Phase 2 SPEC
+└── ...
+
+📁 .claude/vibe/features/{feature-name}/
+├── _index.feature         → Master Feature (read first for scenario overview)
+├── phase-1-{name}.feature → Phase 1 scenarios
+├── phase-2-{name}.feature → Phase 2 scenarios
+└── ...
+
+→ Load _index.md first, then load phase files in order
+→ Execute phases sequentially (or per --phase flag)
+```
+
+**Single file detected:**
+```
+📄 .claude/vibe/specs/{feature-name}.md      → SPEC (structure, constraints, context)
+📄 .claude/vibe/features/{feature-name}.feature → Feature (scenario = implementation unit)
+```
+
+**Error if NEITHER file NOR folder found:**
+```
+❌ SPEC not found. Searched:
+   - .claude/vibe/specs/{feature-name}/  (folder)
+   - .claude/vibe/specs/{feature-name}.md (file)
+
+   Run /vibe.spec "{feature-name}" first.
+```
+
+### 1-1. Phase Isolation Protocol (Large SPEC Guard)
+
+> **Problem**: Large SPECs (3+ phases, 5+ scenarios) overflow context — agent drifts from SPEC by Phase 3.
+> **Solution**: Load only the current phase's SPEC section. Re-anchor before each scenario.
+
+**Phase Isolation Rules (MANDATORY for 3+ phases):**
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  PHASE ISOLATION PROTOCOL                                       │
+│                                                                  │
+│  ❌ WRONG: Load entire SPEC → implement all phases              │
+│  ✅ RIGHT: Load _index overview → per-phase load → implement    │
+│                                                                  │
+│  Step A: Read _index.md (overview only — phase list, REQ IDs)   │
+│  Step B: For each Phase N:                                       │
+│    1. RE-READ Phase N SPEC section (every time, no memory)       │
+│    2. RE-READ Phase N Feature scenarios                          │
+│    3. Extract Phase N scope: files, scenarios, requirements      │
+│    4. Implement Phase N scenarios                                 │
+│    5. Verify Phase N                                             │
+│    6. Write Phase Checkpoint → .claude/vibe/checkpoints/         │
+│    7. DISCARD Phase N details from working memory                │
+│  Step C: Next Phase — go to Step B                               │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+**Phase Checkpoint Format** (`.claude/vibe/checkpoints/{feature}-phase-{N}.md`):
+
+```markdown
+# Checkpoint: {feature} Phase {N}
+
+## Completed
+- Scenario 1: {name} ✅
+- Scenario 2: {name} ✅
+
+## Files Changed
+- src/auth.service.ts (added login(), validateToken())
+- src/auth.controller.ts (POST /login, POST /refresh)
+
+## State for Next Phase
+- Auth service exports: login(), logout(), validateToken(), refreshToken()
+- JWT secret configured in .env (JWT_SECRET)
+- Test baseline: 12 tests passing
+
+## Remaining Phases
+- Phase {N+1}: {name} — {scenario count} scenarios
+- Phase {N+2}: {name} — {scenario count} scenarios
+```
+
+**SPEC Re-anchoring (Before EVERY scenario):**
+
+```
+Before implementing Scenario X:
+  1. Re-read the EXACT Given/When/Then from Feature file (not from memory!)
+  2. Compare: "Am I about to implement what the SPEC says, or what I think it says?"
+  3. If single-file SPEC: re-read only the current phase section (use line offsets)
+  4. If split SPEC: re-read only phase-N-{name}.md
+```
+
+**Scope Lock (Per Phase):**
+
+```
+At Phase start, declare:
+  MODIFY: [list of files this phase will touch]
+  CREATE: [list of files this phase will create]
+  DO NOT TOUCH: everything else
+
+If implementation requires files outside scope:
+  → STOP. Re-read SPEC. Is this actually needed?
+  → If yes: add to scope with explicit justification
+  → If no: you're drifting. Return to SPEC.
+```
+
+**Context Pressure Handling:**
+
+| Context Level | Action |
+|---------------|--------|
+| < 50% | Normal execution |
+| 50-70% | Save checkpoint, trim exploration results |
+| 70%+ | Save checkpoint → `/new` → resume from checkpoint |
+| Phase boundary | Always save checkpoint regardless of context level |
+
+### 1-2. SPEC-First Gate (Level 3: Spec = Source of Truth)
+
+> **Principle**: SPEC is the source of truth for code. To modify code, update the SPEC first.
+
+**When a change not in the SPEC is needed during implementation:**
+
+```
+Discovery during implementation: "An API endpoint not in SPEC is needed"
+    │
+    ├─ Already in SPEC?
+    │   YES → Implement as-is
+    │
+    ├─ Not in SPEC but within SPEC scope?
+    │   YES → Add to SPEC first (Edit tool → SPEC file)
+    │       → Add corresponding scenario to Feature file
+    │       → Then implement
+    │
+    └─ Outside SPEC scope?
+        YES → Record as TODO and exclude from current scope
+              .claude/vibe/todos/out-of-scope-{item}.md
+```
+
+**Required when changing SPEC:**
+
+1. Update REQ-* IDs in SPEC file (add new requirements)
+2. Add corresponding scenarios to Feature file
+3. New REQ-* must be traceable in RTM after implementation
+
+**Reverse direction also applies — when requirements change from code:**
+
+```
+Test failure → "Is this behavior correct?"
+    │
+    ├─ Check SPEC → If SPEC is correct, fix the code
+    │
+    └─ If SPEC is wrong → Fix SPEC first → Then fix code
+        (Changing code without updating SPEC causes SPEC ↔ code drift)
+```
+
+**Extended Git commit rules:**
+
+- SPEC changes and code changes must be in the same commit
+- Committing feature code without SPEC changes → warning (except intentional refactoring)
+
+### 2. Extract Scenario List
+
+Extract all Scenarios from Feature file:
+
+```markdown
+## Scenarios to Implement
+
+| # | Scenario | Status |
+|---|----------|--------|
+| 1 | Valid login success | ⬜ |
+| 2 | Invalid password error | ⬜ |
+| 3 | Email format validation | ⬜ |
+| 4 | Password reset link | ⬜ |
+
+Total: 4 scenarios
+```
+
+### 3. Scenario-by-Scenario Implementation (Core)
+
+**For each scenario**:
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 1/4: Valid login success
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Given: User is registered
+When: Login with valid email and password
+Then: Login success + JWT token returned
+
+[Step 1] Analyzing implementation...
+  - Required files: auth.service.ts, login.controller.ts
+  - Exploring related code...
+
+[Step 2] Implementing...
+  ✅ auth.service.ts - Added login() method
+  ✅ login.controller.ts - POST /login endpoint
+
+[Step 3] Verifying...
+  ✅ Given: Test user creation possible
+  ✅ When: Login API call succeeded
+  ✅ Then: JWT token return confirmed
+
+✅ Scenario 1 passed!
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**On failure**:
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 2/4: Invalid password error
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+[Step 3] Verifying...
+  ✅ Given: Test user exists
+  ✅ When: Login attempt with wrong password
+  ❌ Then: "Invalid credentials" error message
+     Actual: "Error occurred" returned
+
+[Auto-fix iteration 1]
+  Cause: Error message not properly set
+  Fix: auth.service.ts line 42
+
+[Re-verify]
+  ✅ Then: "Invalid credentials" error message
+
+✅ Scenario 2 passed! (1 fix)
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+---
+
+## **CRITICAL: Parallel Sub-Agent Execution**
+
+> **MUST USE PARALLEL TASK CALLS** - This is REQUIRED, not optional.
+> Sequential execution when parallel is possible = VIOLATION of this workflow.
+
+### Codex 병렬 구현 (Codex 플러그인 활성화 시)
+
+> **활성화 조건**: Codex 플러그인 설치 + 독립적인 시나리오 2개 이상
+> 미설치 시 기존 Claude 서브에이전트 방식으로 동작.
+
+독립적인 시나리오(파일 의존성 없음)를 Codex에 위임하여 Claude와 **병렬 구현**:
+
+```
+/codex:rescue "Implement scenario: {scenario-name}. Files: {file-list}. Requirements: {requirements-summary}" --background
+```
+
+- Claude는 다음 시나리오를 **동시에** 구현
+- Codex 완료 시 `/codex:result`로 결과 확인
+- 충돌 발생 시 Claude가 merge 판단
+
+**위임 기준**:
+- 시나리오 간 파일 의존성 없음 (독립적)
+- 시나리오 복잡도 중간 이하
+- 의존성 있는 시나리오는 Claude가 직접 구현
+
+### Mandatory Parallel Exploration (Phase Start)
+
+**BEFORE any implementation, you MUST launch these Task calls IN PARALLEL (single message, multiple tool calls):**
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  STEP 1: PARALLEL EXPLORATION (REQUIRED)                        │
+│                                                                 │
+│  Launch ALL of these in ONE message:                            │
+│                                                                 │
+│  Task(haiku) ─┬─→ "Analyze related files in <context>"          │
+│               │                                                 │
+│  Task(haiku) ─┼─→ "Check dependencies and imports"              │
+│               │                                                 │
+│  Task(haiku) ─┴─→ "Find existing patterns and conventions"      │
+│                                                                 │
+│  [If GPT enabled] Bash: node "[LLM_SCRIPT]" gpt-codex orchestrate-json "[question]"
+│  [If Gemini enabled] Bash: node "[LLM_SCRIPT]" gemini orchestrate-json "[question]"
+└─────────────────────────────────────────────────────────────────┘
+                              │
+                              ↓ (wait for all to complete)
+┌─────────────────────────────────────────────────────────────────┐
+│  STEP 2: SYNTHESIZE (Opus)                                      │
+│  - Review all exploration results                               │
+│  - Decide implementation approach                               │
+│  - Identify files to modify/create                              │
+└─────────────────────────────────────────────────────────────────┘
+                              │
+                              ↓
+┌─────────────────────────────────────────────────────────────────┐
+│  STEP 3: IMPLEMENT + BACKGROUND AGENTS (PARALLEL)               │
+│                                                                 │
+│  Main Agent (sonnet):                                           │
+│  └─→ Execute current phase implementation                       │
+│                                                                 │
+│  Background Agents (haiku, run_in_background=true):             │
+│  ├─→ Task: "Prepare Phase N+1 - analyze required files"         │
+│  ├─→ Task: "Pre-generate test cases for current implementation" │
+│  └─→ Task: "Search for related types/interfaces needed"         │
+│                                                                 │
+│  [ULTRAWORK] All 4 agents run simultaneously!                   │
+└─────────────────────────────────────────────────────────────────┘
+                              │
+                              ↓ (main completes, check backgrounds)
+┌─────────────────────────────────────────────────────────────────┐
+│  STEP 4: TEST + PHASE PIPELINING                                │
+│                                                                 │
+│  Current Phase:                                                 │
+│  └─→ Task(haiku): Write tests using pre-generated cases         │
+│                                                                 │
+│  Next Phase Prep (from background results):                     │
+│  └─→ Already have file analysis, ready to start immediately     │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+### Parallel Task Call Pattern (MUST FOLLOW)
+
+**Correct - Single message with multiple parallel Tasks:**
+```
+<message>
+  Task(haiku, "Analyze src/components/ for existing patterns")
+  Task(haiku, "Check package.json dependencies")
+  Task(haiku, "Find usage of similar features in codebase")
+</message>
+→ All 3 run simultaneously, ~3x faster
+```
+
+**WRONG - Sequential calls (DO NOT DO THIS):**
+```
+<message>Task(haiku, "Analyze...")</message>
+<message>Task(haiku, "Check...")</message>
+<message>Task(haiku, "Find...")</message>
+→ 3x slower, wastes time
+```
+
+### Background Agent Pattern (ULTRAWORK) via Orchestrator
+
+**Launch background agents for next phase via Orchestrator:**
+```bash
+# Start background agent (doesn't block)
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => o.runAgent('Phase 2 prep: Analyze auth API endpoints', 'phase2-prep').then(r => console.log(r.content[0].text)))"
+
+# Multiple backgrounds in parallel
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(async o => {
+  await Promise.all([
+    o.runAgent('Phase 2 prep: Analyze auth API endpoints', 'phase2-prep'),
+    o.runAgent('Pre-generate test cases for login form', 'test-prep'),
+    o.runAgent('Find existing validation patterns', 'pattern-finder')
+  ]);
+  console.log('All background agents started');
+})"
+```
+
+**Check background agent status:**
+```bash
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => console.log(o.status().content[0].text))"
+```
+
+**Get result when ready:**
+```bash
+node -e "import('{{VIBE_PATH_URL}}/node_modules/@su-record/vibe/dist/infra/orchestrator/index.js').then(o => o.getResult('SESSION_ID').then(r => console.log(r.content[0].text)))"
+```
+
+**Why Background Agents Matter:**
+
+| Without Background | With Background |
+|--------------------|-----------------|
+| Phase 1: 60s | Phase 1: 60s (+ backgrounds running) |
+| Phase 2 prep: 20s | Phase 2 prep: 0s (already done!) |
+| Phase 2: 60s | Phase 2: 60s |
+| **Total: 140s** | **Total: 120s** |
+
+For 5 phases: 4 × 20s saved = **80s faster**
+
+### Why Parallel Matters
+
+| Approach | Time | Cache Benefit |
+|----------|------|---------------|
+| Sequential (3 Tasks) | ~30s | Cache cold on each |
+| **Parallel (3 Tasks)** | **~10s** | **Cache warmed once, shared** |
+
+core ProjectCache (LRU) caches ts-morph parsing results. Parallel calls share the warmed cache.
+
+### UI/UX Design Intelligence (Auto-triggered before Phase 1)
+
+> **조건**: SPEC 또는 Feature에 UI/UX 키워드 포함 시 자동 실행
+> **비활성화**: `.claude/vibe/config.json`에 `"uiUxAnalysis": false` 설정
+
+**Phase 1 시작 전, 2개 에이전트 자동 실행:**
+
+| Agent | Condition | Role |
+|-------|-----------|------|
+| ④ ui-stack-implementer | **항상 실행** | 프레임워크별 컴포넌트 가이드라인 제공 |
+| ⑤ ui-dataviz-advisor | **조건부** (chart/dashboard/visualization 키워드) | 차트/시각화 라이브러리 추천 |
+
+**실행 방법:**
+
+```text
+# ④ 항상 실행 (Haiku)
+Task(subagent_type="ui-stack-implementer",
+  prompt="Provide implementation guidelines for project '{project}' using {detected_stack}. Use core_ui_stack_search for framework-specific patterns.")
+
+# ⑤ 조건부 실행 (Haiku) — SPEC에 차트/대시보드/시각화 키워드 포함 시
+Task(subagent_type="ui-dataviz-advisor",
+  prompt="Recommend data visualization approach for project '{project}'. Use core_ui_search for chart types and react-performance patterns.")
+```
+
+**디자인 시스템 자동 참조:**
+- `.claude/vibe/design-system/{project}/MASTER.md` 존재 시 자동 로드
+- 구현 에이전트가 CSS 변수, 폰트, 색상 팔레트를 직접 참조
+- 페이지별 오버라이드 `pages/{page}.md` 존재 시 우선 적용
+
+### Phase Execution Flow (ULTRAWORK Pipeline)
+
+```
+Phase N Start
+    │
+    ├─→ [PARALLEL] Task(haiku) × 3: Exploration
+    │       - Related code analysis
+    │       - Dependency check
+    │       - Pattern discovery
+    │
+    ↓ (all complete)
+    │
+    ├─→ Opus: Synthesize and decide
+    │
+    ├─→ [PARALLEL PIPELINE] ←── KEY SPEED OPTIMIZATION
+    │       │
+    │       ├─→ Main: Task(sonnet) Implementation
+    │       │
+    │       └─→ Background (run_in_background=true):
+    │               ├─→ Task(haiku): Phase N+1 file analysis
+    │               ├─→ Task(haiku): Test case preparation
+    │               └─→ Task(haiku): Type/interface lookup
+    │
+    ↓ (main completes)
+    │
+    ├─→ Task(haiku): Tests (uses pre-generated cases)
+    │
+    ↓
+Phase N Complete
+    │
+    ↓ (Background results ready - NO WAIT for Phase N+1 exploration!)
+    │
+Phase N+1 Start (IMMEDIATE - exploration already done!)
+```
+
+**Speed Comparison:**
+
+| Mode | Phase Time | 5 Phases Total |
+|------|------------|----------------|
+| Sequential | ~2min/phase | ~10min |
+| Parallel Exploration | ~1.5min/phase | ~7.5min |
+| **ULTRAWORK Pipeline** | **~1min/phase** | **~5min** |
+
+**Why Pipeline is Faster:**
+- Background agents prepare next phase WHILE current phase implements
+- No idle time between phases
+- Test cases pre-generated during implementation
+- Cache stays warm across parallel tasks
+
+---
+
+### Agent Teams — Dev Team
+
+> **팀 정의**: `agents/teams/dev-team.md` 참조
+> 설정: `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` + `teammateMode: in-process` (`~/.claude/settings.json` 전역 — postinstall 자동 설정)
+
+**활성화 조건 (Dev Team Full — 4명):**
+- ULTRAWORK 모드 + 3개 이상 시나리오
+- 또는 복잡도 점수 20+ (High)
+
+### Agent Teams — Lite Team (Normal Mode)
+
+> **팀 정의**: `agents/teams/lite-team.md` 참조
+
+**활성화 조건 (Lite Team — 3명):**
+- 일반 모드 + 3개 이상 시나리오
+- 복잡도 점수 8-19 (Medium)
+- 단순 구현(1-2 파일, 시나리오 2개 이하)에서는 기존 병렬 모드 유지
+
+**팀 선택 기준:**
+
+| 조건 | 팀 |
+|------|-----|
+| 시나리오 1-2개, 파일 1-2개 | 기존 병렬 모드 (팀 없음) |
+| 시나리오 3개+, 일반 모드 | **Lite Team (3명)** |
+| ULTRAWORK 또는 복잡도 20+ | Dev Team Full (4명) |
+
+### Agent Teams — Review Team
+
+> **팀 정의**: `agents/teams/review-debate-team.md` 참조
+
+**활성화 조건:**
+
+- `/vibe.review` 실행 후 P1 또는 P2 이슈 2개 이상 발견 시
+- Agent Teams 환경변수 활성화 상태
+
+### Agent Teams — Debug Team
+
+> **팀 정의**: `agents/teams/debug-team.md` 참조
+
+**활성화 조건:**
+
+- 동일 빌드/테스트 실패 3회 이상
+- UltraQA `architecture_question` 상태 진입 시
+
+### Agent Teams — Research Team
+
+> **팀 정의**: `agents/teams/research-team.md` 참조
+
+**활성화 조건:**
+
+- `/vibe.spec` Step 3 리서치 단계
+- Agent Teams 환경변수 활성화 상태
+
+### Agent Teams — Security Team
+
+> **팀 정의**: `agents/teams/security-team.md` 참조
+
+**활성화 조건:**
+
+- auth, payment, user-data, crypto 관련 파일 변경 감지 시
+- 또는 수동으로 `security` 키워드 지정 시
+
+### Agent Teams — Migration Team
+
+> **팀 정의**: `agents/teams/migration-team.md` 참조
+
+**활성화 조건:**
+
+- package.json 주요 의존성 버전 변경 감지 시
+- 또는 수동으로 `migration` 키워드 지정 시
+
+### Agent Teams — Fullstack Team
+
+> **팀 정의**: `agents/teams/fullstack-team.md` 참조
+
+**활성화 조건:**
+
+- SPEC에 frontend + backend 파일이 모두 포함된 경우
+- 또는 수동으로 `fullstack` 키워드 지정 시
+
+---
+
+1. **Related code analysis**: Task(haiku) explores `<context>` related code
+2. **File creation/modification**: Task(sonnet) implements per `<output_format>`
+3. **Constraint compliance**: Check `<constraints>`
+4. **Run verification**: Execute verification commands
+
+### 4. Brand Assets Generation (Optional)
+
+When starting a **new project** with brand context in SPEC, auto-generate app icons and favicons:
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎨 BRAND ASSETS GENERATION
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+[Check] Brand assets exist? → Skip if favicon.ico exists
+[Check] Gemini API configured? → Required for image generation
+[Check] SPEC has brand context? → Extract app name, colors, style
+
+[Generate] Creating app icon with Gemini Image API...
+  - Prompt: "App icon for [AppName], [style], [color]..."
+  - Generated: 512x512 master icon
+
+[Resize] Creating platform variants...
+  ✅ favicon.ico (16/32/48)
+  ✅ favicon-16x16.png
+  ✅ favicon-32x32.png
+  ✅ apple-touch-icon.png (180x180)
+  ✅ android-chrome-192x192.png
+  ✅ android-chrome-512x512.png
+  ✅ site.webmanifest
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+✅ Brand assets generated in public/
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**SPEC Brand Context Example:**
+
+```xml
+<context>
+Brand:
+  - App Name: MyApp
+  - Primary Color: #2F6BFF
+  - Style: Modern, minimalist, flat design
+  - Icon Concept: Abstract geometric shape
+</context>
+```
+
+**Trigger Conditions:**
+- First `/vibe.run` execution (no existing icons)
+- SPEC contains brand/design context
+- Gemini API key configured (`vibe gemini key <key>`)
+
+**Manual Generation:**
+```bash
+# [LLM_SCRIPT] = {{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js
+node "[LLM_SCRIPT]" gemini image "App icon for MyApp, primary color #2F6BFF, square format 1:1, simple recognizable design, works well at small sizes, no text or letters, solid or gradient background, modern minimalist" --output "./public/app-icon.png"
+```
+
+---
+
+### 5. Race Code Review (GPT + Gemini) + Auto-Fix (v2.6.9)
+
+After all scenarios are implemented, **GPT and Gemini review in parallel with cross-validation**:
+
+> **ULTRAWORK Default**: In ULTRAWORK mode, race review is automatically enabled.
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🏁 RACE CODE REVIEW (GPT + Gemini)
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+[Step 1] Parallel review execution...
+  ├─ GPT Codex: Reviewing...
+  └─ Gemini: Reviewing...
+
+[Step 2] Cross-validation results:
+  ┌──────────────────────────────────────────────────────────────────┐
+  │ Issue                          │ GPT │ Gemini │ Codex │ Confidence│
+  │────────────────────────────────│─────│────────│───────│───────────│
+  │ Timing attack in password      │ ✅  │ ✅     │ ✅    │ 100% → P1 │
+  │ Rate limiting missing          │ ✅  │ ✅     │ ✅    │ 100% → P1 │
+  │ Magic number usage             │ ✅  │ ❌     │ ❌    │ 50% → P2  │
+  └──────────────────────────────────────────────────────────────────┘
+
+  Summary: 3 issues (P1: 2, P2: 1)
+
+[Step 3] Auto-fixing P1/P2 issues...
+  ✅ auth.service.ts:24 - Applied timingSafeEqual (P1)
+  ✅ auth.controller.ts:15 - Added rate limiter (P1)
+  ✅ auth.service.ts:42 - Extracted constant (P2)
+
+[Step 4] Re-verifying...
+  ✅ Build succeeded
+  ✅ Tests passed
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+✅ Race review complete! 3 improvements (2 P1, 1 P2)
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**Race Review Invocation (GPT + Gemini in parallel via Bash):**
+
+**🚨 Use --input file to avoid CLI argument length limits and Windows pipe issues.**
+
+1. Save code to review into `[SCRATCHPAD]/review-code.txt` (using Write tool)
+2. Write JSON input file `[SCRATCHPAD]/review-input.json` (using Write tool):
+   - `{"prompt": "Review this code for security, performance, and best practices. Return JSON: {issues: [{id, title, description, severity, suggestion}]}. Code: [CODE_CONTENT]"}`
+   - Where `[CODE_CONTENT]` is the code text (properly JSON-escaped inside the prompt string)
+3. Script path: `[LLM_SCRIPT]` = `{{VIBE_PATH}}/hooks/scripts/llm-orchestrate.js`
+4. Run GPT + Gemini in PARALLEL (two Bash tool calls at once):
+
+```bash
+# GPT review (Bash tool call 1)
+node "[LLM_SCRIPT]" gpt orchestrate-json --input "[SCRATCHPAD]/review-input.json"
+```
+
+```bash
+# Gemini review (Bash tool call 2 - run in parallel)
+node "[LLM_SCRIPT]" gemini orchestrate-json --input "[SCRATCHPAD]/review-input.json"
+```
+
+**Confidence-based Priority:**
+
+| Confidence | Priority | Action |
+|------------|----------|--------|
+| 100% (3/3 or 2/2) | P1 | Auto-fix immediately |
+| 67% (2/3) | P1 | Auto-fix immediately |
+| 50% (1/2) or 33% (1/3) | P2 | Auto-fix with review |
+
+**Fallback handling:**
+- If one LLM fails → Use remaining LLM results (reduced confidence)
+- If all fail → Skip and proceed (log warning)
+
+**Review application rules:**
+
+| Feedback Type | Action |
+|---------------|--------|
+| Security vulnerability (P1) | Auto-fix immediately |
+| Performance improvement (P1/P2) | Auto-fix immediately |
+| Best practices (P2) | Auto-fix |
+| Style/preference (P3) | Apply selectively |
+
+**Conditions:**
+- **ULTRAWORK**: Race review enabled by default
+- **Normal mode**: Use `--race` flag to enable
+- Must re-verify build/tests after fixes
+
+### Codex Code Review (Codex 플러그인 활성화 시)
+
+GPT+Gemini race와 **동시에** Codex review 실행:
+
+```
+/codex:review
+```
+
+결과를 race review 교차 검증에 포함 — 3중 리뷰:
+
+```markdown
+| Issue | GPT | Gemini | Codex | Confidence |
+|-------|-----|--------|-------|------------|
+| {이슈} | ✅/❌ | ✅/❌ | ✅/❌ | {%} |
+```
+
+### 6. Quality Report (Auto-generated)
+
+After all scenarios complete + Gemini review, **quality report is auto-generated**:
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  📊 QUALITY REPORT: login                                        │
+├─────────────────────────────────────────────────────────────────┤
+│                                                                 │
+│  ✅ Scenarios: 4/4 passed                                       │
+│                                                                 │
+│  ┌───────────────────────────────────────────────────────────┐  │
+│  │ # │ Scenario                  │ Status │ Retries │        │  │
+│  │───│───────────────────────────│────────│─────────│        │  │
+│  │ 1 │ Valid login success       │ ✅     │ 0       │        │  │
+│  │ 2 │ Invalid password error    │ ✅     │ 1       │        │  │
+│  │ 3 │ Email format validation   │ ✅     │ 0       │        │  │
+│  │ 4 │ Password reset link       │ ✅     │ 0       │        │  │
+│  └───────────────────────────────────────────────────────────┘  │
+│                                                                 │
+│  📈 Quality score: 94/100                                       │
+│                                                                 │
+│  ┌─────────────────────────────────────────────────────────┐    │
+│  │ Item              │ Result │ Notes                       │    │
+│  │───────────────────│────────│─────────────────────────────│    │
+│  │ Build             │ ✅     │ npm run build succeeded     │    │
+│  │ Tests             │ ✅     │ 12/12 passed                │    │
+│  │ Type check        │ ✅     │ 0 errors                    │    │
+│  │ Complexity        │ ✅     │ All functions ≤30 lines     │    │
+│  │ Security          │ ✅     │ 0 vulnerabilities           │    │
+│  │ Race review       │ ✅     │ 3 improvements applied      │    │
+│  └─────────────────────────────────────────────────────────┘    │
+│                                                                 │
+│  ⏱️ Started: {start_time}                                        │
+│  ⏱️ Completed: {getCurrentTime 결과}                             │
+│                                                                 │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+**What users should check**:
+- Scenario pass rate (4/4 = 100%)
+- Quality score (94/100)
+- Build/test status
+
+**This alone is enough to trust quality.**
+
+### 7. Update Feature File
+
+Auto-update scenario status:
+
+```markdown
+## Coverage
+
+| Scenario | SPEC AC | Status |
+|----------|---------|--------|
+| Valid login success | AC-1 | ✅ |
+| Invalid password error | AC-2 | ✅ |
+| Email format validation | AC-3 | ✅ |
+| Password reset link | AC-4 | ✅ |
+
+**Last verified**: 2024-01-15 14:32
+**Quality score**: 94/100
+```
+
+## TRUST 5 Principles
+
+Follow during implementation:
+
+| Principle | Description |
+|-----------|-------------|
+| **T**est-first | Write tests first |
+| **R**eadable | Clear code |
+| **U**nified | Consistent style |
+| **S**ecured | Consider security |
+| **T**rackable | Logging, monitoring |
+
+## Input
+
+- `.claude/vibe/specs/{feature-name}.md` (PTCF SPEC)
+- `.claude/vibe/features/{feature-name}.feature` (BDD)
+- `CLAUDE.md` (project context)
+
+## Output
+
+- Implemented code files
+- Test files
+- Updated SPEC document (checkmarks)
+
+## Example
+
+```
+User: /vibe.run "login"
+
+Claude:
+📄 Loading SPEC: .claude/vibe/specs/login.md
+📄 Loading Feature: .claude/vibe/features/login.feature
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📋 Scenarios to Implement
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+| # | Scenario | Status |
+|---|----------|--------|
+| 1 | Valid login success | ⬜ |
+| 2 | Invalid password error | ⬜ |
+| 3 | Email format validation | ⬜ |
+| 4 | Password reset link | ⬜ |
+
+Total: 4 scenarios
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 1/4: Valid login success
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Given: User is registered
+When: Login with valid email and password
+Then: Login success + JWT token returned
+
+⚡ [PARALLEL] Exploring...
+✅ Exploration complete (2.1s)
+
+🔨 Implementing...
+  ✅ auth.service.ts - Added login()
+  ✅ auth.controller.ts - POST /login
+
+🔍 Verifying...
+  ✅ Given: OK
+  ✅ When: OK
+  ✅ Then: OK
+
+✅ Scenario 1 passed!
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 2/4: Invalid password error
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+🔨 Implementing...
+  ✅ auth.service.ts - Password validation logic
+
+🔍 Verifying...
+  ✅ Given: OK
+  ✅ When: OK
+  ❌ Then: "Invalid credentials" error message
+     Actual: "Error" returned
+
+🔄 Auto-fix [iteration 1]...
+  Fix: auth.service.ts line 42
+
+🔍 Re-verifying...
+  ✅ Then: OK
+
+✅ Scenario 2 passed! (1 fix)
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 3/4: Email format validation
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+🔨 Implementing...
+🔍 Verifying...
+✅ Scenario 3 passed!
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎯 Scenario 4/4: Password reset link
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+🔨 Implementing...
+🔍 Verifying...
+✅ Scenario 4 passed!
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🔍 GEMINI CODE REVIEW
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📤 Sending code to Gemini...
+📝 Gemini feedback:
+  1. [Security] Need timing attack prevention → Fixing...
+  2. [Performance] Unnecessary DB call → Fixing...
+
+✅ 2 improvements auto-applied
+🔍 Re-verifying... ✅ Passed
+
+┌─────────────────────────────────────────────────────────────────┐
+│  📊 QUALITY REPORT: login                                        │
+├─────────────────────────────────────────────────────────────────┤
+│                                                                 │
+│  ✅ Scenarios: 4/4 passed                                       │
+│                                                                 │
+│  | # | Scenario              | Status | Retries |               │
+│  |---|───────────────────────|───────|─────────|               │
+│  | 1 | Valid login success   | ✅    | 0       |               │
+│  | 2 | Invalid password error| ✅    | 1       |               │
+│  | 3 | Email format validation| ✅   | 0       |               │
+│  | 4 | Password reset link   | ✅    | 0       |               │
+│                                                                 │
+│  📈 Quality score: 94/100                                       │
+│  Build: ✅ | Tests: ✅ | Types: ✅ | Gemini: ✅ (2 applied)     │
+│                                                                 │
+│  ⏱️ Started: {start_time}                                        │
+│  ⏱️ Completed: {getCurrentTime 결과}                             │
+└─────────────────────────────────────────────────────────────────┘
+
+🎉 Implementation complete! All scenarios passed + Gemini review applied.
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🔍 AUTO REVIEW (13+ Agents)
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+⚡ [PARALLEL] 13 expert agents reviewing...
+  - security-reviewer ✅
+  - performance-reviewer ✅
+  - architecture-reviewer ✅
+  - ...
+
+📋 Review results:
+  - P1 Critical: 0
+  - P2 Important: 2
+  - P3 Nice-to-have: 1
+
+🔧 Auto-fixing P2 issues...
+  1. [PERF] N+1 query → Fixed
+  2. [ARCH] Circular dependency → Fixed
+
+✅ Auto Review complete! 2 issues auto-resolved.
+```
+
+### Phase-specific Execution
+
+```
+User: /vibe.run "brick-game" --phase 2
+
+Claude:
+📄 Reading SPEC: .claude/vibe/specs/brick-game.md
+🎯 Executing Phase 2 only.
+
+Phase 2: Game Logic
+1. [ ] Paddle movement implementation
+2. [ ] Ball physics engine
+3. [ ] Brick collision handling
+4. [ ] Score system
+5. [ ] Game over conditions
+
+⚡ Launching parallel exploration...
+[Task(haiku) × 3 launched in parallel]
+
+🚀 Starting implementation...
+```
+
+## Error Handling
+
+On failure:
+1. Check error message
+2. Review `<constraints>`
+3. Fix code and retry
+4. If continues to fail, report to user
+
+---
+
+## Quality Gate (Mandatory)
+
+### Implementation Quality Checklist
+
+Before marking any scenario as complete, ALL items must pass:
+
+| Category | Check Item | Weight |
+|----------|------------|--------|
+| **Functionality** | All Given/When/Then conditions verified | 20% |
+| **Functionality** | Edge cases handled per scenario | 10% |
+| **Code Quality** | No `any` types in TypeScript | 10% |
+| **Code Quality** | Functions ≤30 lines, nesting ≤3 levels | 10% |
+| **Code Quality** | No hardcoded values (use constants) | 5% |
+| **Security** | Input validation implemented | 10% |
+| **Security** | Authentication/authorization checked | 5% |
+| **Error Handling** | Try-catch or error states present | 10% |
+| **Error Handling** | User-friendly error messages | 5% |
+| **Testing** | Unit tests exist for core logic | 10% |
+| **Performance** | No N+1 queries or unnecessary loops | 5% |
+
+### Quality Score Calculation
+
+```
+Score = Σ(checked items × weight) / 100
+
+Grades:
+- 95-100: ✅ EXCELLENT - Ready to merge
+- 90-94:  ⚠️ GOOD - Minor improvements required before merge
+- 80-89:  ⚠️ FAIR - Significant improvements required
+- 0-79:   ❌ POOR - Major fixes needed
+```
+
+### Quality Gate Thresholds
+
+| Gate | Minimum Score | Condition |
+|------|---------------|-----------|
+| **Scenario Complete** | 95 | Each scenario must score ≥95 |
+| **Phase Complete** | 95 | Average of all scenarios ≥95 |
+| **Feature Complete** | 95 | All phases complete + Gemini review |
+
+### Auto-Fix Triggers
+
+| Issue Type | Auto-Fix Action |
+|------------|-----------------|
+| Missing error handling | Add try-catch wrapper |
+| Hardcoded values | Extract to constants file |
+| Missing input validation | Add validation schema |
+| Function too long | Suggest split points |
+| N+1 query detected | Add eager loading |
+
+### Auto-Fix 실패 시 Codex Rescue (Codex 플러그인 활성화 시)
+
+P1 auto-fix가 **3회 실패** 시, Codex에 위임:
+
+```
+/codex:rescue "Fix P1 issue: {issue-description}. File: {file-path}. Error: {error-message}"
+```
+
+Codex 수정 완료 후 재검증. Codex도 실패 시 TODO 파일에 기록.
+
+### Forbidden Patterns (Block Merge)
+
+| Pattern | Why Forbidden | Detection |
+|---------|---------------|-----------|
+| `console.log` | Debug code in production | Regex scan |
+| `// TODO` without issue | Untracked work | Comment scan |
+| `any` type | Type safety bypass | TypeScript check |
+| `@ts-ignore` | Type error suppression | TypeScript check |
+| Empty catch blocks | Silent error swallowing | AST analysis |
+| Commented-out code | Dead code | Comment scan |
+
+---
+
+## Auto-Retrospective (Post-Implementation)
+
+After ALL phases complete successfully, **automatically** perform a brief retrospective:
+
+### Retrospective Template
+
+```
+## Retrospective: {feature-name}
+
+### What Worked
+- [List effective patterns, tools, approaches used]
+
+### What Didn't
+- [List issues, failures, unexpected blockers]
+
+### Key Decisions
+- [Important architectural or implementation decisions made during this run]
+
+### Lessons Learned
+- [Principle format: "When X, do Y because Z"]
+```
+
+### Execution Steps
+
+1. Generate retrospective based on the implementation session
+2. Save to `.claude/vibe/retros/{feature-name}.md`
+3. Save key lessons via `core_save_memory` (for cross-session recall)
+4. Update `claude-progress.txt` with final status
+
+**Important:**
+
+- Keep it concise (under 20 lines)
+- Focus on **project-specific** insights, not generic knowledge
+- Only save to memory if the lesson is actionable and non-obvious
+
+---
+
+## Next Step
+
+```
+/vibe.verify "brick-game"
+```
+
+---
+
+ARGUMENTS: $ARGUMENTS