@su-record/vibe 2.6.34 → 2.6.41

package/commands/vibe.spec.md

@@ -32,6 +32,8 @@ When `ultrawork` (or `ulw`) is included, automatically chains:
 
 **No manual intervention between steps.**
 
+> **⏱️ Timer**: Call `getCurrentTime` tool at the START. Record the result as `{start_time}`.
+
 ## Input Mode Detection (Auto-detect)
 
 **Input priority:**
@@ -90,11 +92,34 @@ User: 1 hour, disallow, confirm
 | PDF | `.pdf` | Planning docs, design documents |
 | Image | `.png`, `.jpg`, `.jpeg`, `.webp` | Wireframes, UI design, screenshots |
 
+**Image input analysis:**
+
+When image files (`.png`, `.jpg`, `.jpeg`, `.webp`) are provided as input, analyze them using the best available method:
+
+- **Gemini Enabled**: `llm-orchestrate.js gemini analyze-image` (Gemini 3 Flash - best image recognition)
+- **Gemini Disabled**: Claude Opus Read tool (built-in multimodal, existing behavior)
+
+**Gemini enabled - analyze via llm-orchestrate.js:**
+
+```bash
+# [LLM_SCRIPT] = resolved path from: node -e "console.log(require('path').join(process.env.APPDATA || require('os').homedir() + '/.config', 'vibe/hooks/scripts/llm-orchestrate.js'))"
+node "[LLM_SCRIPT]" gemini analyze-image "./designs/login-wireframe.png" "Analyze this UI design image. Identify all UI elements, layout structure, colors, typography, and component hierarchy. Output a structured breakdown."
+```
+
+Parse the JSON result: `{ success: true, analysis: "..." }` → use `analysis` field content.
+
+If `success: false`, fall back to Claude Read tool.
+
+**Gemini disabled - analyze via Claude Read tool:**
+
+Use the Read tool directly on the image file. Claude can read images natively.
+
 **Image input example:**
 ```
 /vibe.spec "designs/login-wireframe.png"
 
 🖼️ Image analysis: designs/login-wireframe.png
+   (via Gemini 3 Flash / Claude Opus)
 
 📋 Detected UI elements:
   - Email input field
@@ -378,41 +403,44 @@ Read ~/.claude/vibe/languages/typescript-react.md
 3. ✅ Language guide copied (step 2.5)
 4. ✅ Core requirements collected
 
-**→ IMMEDIATELY run these 4 Bash commands IN PARALLEL (all at once):**
+**→ IMMEDIATELY run these 6 Bash commands IN PARALLEL (all at once):**
 
 **🚨 MANDATORY: Copy the EXACT path below. DO NOT modify or use alternative paths.**
 
+**Step 0: Resolve script path (once per session, using Bash tool):**
 ```bash
-# Cross-platform path (works on Windows/macOS/Linux)
-# ⚠️ COPY THIS EXACTLY - DO NOT USE ~/.claude/ or any other path!
-VIBE_SCRIPTS="$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts"
+node -e "console.log(require('path').join(process.env.APPDATA || require('os').homedir() + '/.config', 'vibe/hooks/scripts/llm-orchestrate.js'))"
+```
+Save the output as `[LLM_SCRIPT]` for use in all subsequent calls.
 
+**Run all 4 in PARALLEL (each as separate Bash tool call):**
+```bash
 # 1. GPT: Best practices
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+node "[LLM_SCRIPT]" gpt orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
 # 2. GPT: Security
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json "Security vulnerabilities for [FEATURE] with [STACK]. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
+node "[LLM_SCRIPT]" gpt orchestrate-json "Security vulnerabilities for [FEATURE] with [STACK]. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
 
 # 3. Gemini: Best practices
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+node "[LLM_SCRIPT]" gemini orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
 # 4. Gemini: Security
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json "Security advisories for [FEATURE] with [STACK]. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
+node "[LLM_SCRIPT]" gemini orchestrate-json "Security advisories for [FEATURE] with [STACK]. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
 ```
 
 **Concrete example - run all 4 in parallel:**
 ```bash
 # GPT best practices
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+node "[LLM_SCRIPT]" gpt orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
 # GPT security
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "Security vulnerabilities for passkey authentication with React, Supabase. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
+node "[LLM_SCRIPT]" gpt orchestrate-json "Security vulnerabilities for passkey authentication with React, Supabase. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
 
 # Gemini best practices
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+node "[LLM_SCRIPT]" gemini orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
 # Gemini security
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json "Security advisories for passkey authentication with React, Supabase. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
+node "[LLM_SCRIPT]" gemini orchestrate-json "Security advisories for passkey authentication with React, Supabase. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
 ```
 
 **ALSO run Claude research agents in parallel using Task tool:**
@@ -428,6 +456,71 @@ node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/v
 
 **🚨 GPT/Gemini MUST be called via Bash with llm-orchestrate.js! 🚨**
 
+#### 3.0.1 Agent Teams — Research Collaboration
+
+> **Agent Teams**: 개별 리서치 결과를 실제 팀으로 교차 검증하여 더 깊은 인사이트를 도출합니다.
+> 설정: `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` + `teammateMode: in-process` (`~/.claude/settings.json` 전역 — postinstall 자동 설정)
+
+**팀 구성:**
+
+| 팀원 | 역할 |
+|------|------|
+| best-practices (리더) | 패턴/안티패턴 종합, 충돌 해결, 최종 요약 |
+| security-advisory | 보안 관점 검증, 보안 리스크 식별 |
+| codebase-patterns | 기존 코드와의 일관성 검증 |
+| framework-docs | 최신 문서/API 변경사항 확인 |
+
+**실행 순서:**
+
+1. `TeamCreate(team_name="research-{feature}")` — 팀 + 공유 태스크 리스트 생성
+2. 4개 팀원 병렬 생성 — 각각 `Task(team_name=..., name=..., subagent_type=...)` 으로 spawn
+3. 팀원들이 공유 TaskList에서 작업을 claim하고, SendMessage로 상호 검증
+4. 리더(best-practices)가 최종 통합 요약 작성 → SPEC Context에 반영
+5. 모든 팀원 shutdown_request → TeamDelete로 정리
+
+**팀원 spawn 패턴:**
+
+```text
+TeamCreate(team_name="research-{feature}", description="Research collaboration for {feature}")
+
+# 4개 병렬 spawn
+Task(team_name="research-{feature}", name="best-practices", subagent_type="best-practices-agent",
+  prompt="리서치 팀 리더. 10개 병렬 리서치 결과를 종합하세요.
+  리서치 결과: {all_research_results}
+  역할: 패턴/안티패턴 종합, 팀원 간 충돌 해결, 최종 통합 요약 작성.
+  TaskList를 확인하고 작업을 claim하세요. 팀원에게 SendMessage로 검증을 요청하세요.
+  모든 작업 완료 후 최종 요약을 작성하세요.")
+
+Task(team_name="research-{feature}", name="security-advisory", subagent_type="security-advisory-agent",
+  prompt="리서치 팀 보안 담당. 리서치 결과: {all_research_results}
+  역할: 보안 관점에서 모든 권장사항 검증.
+  보안 리스크 발견 시 best-practices에게 SendMessage로 알리세요.
+  TaskList에서 보안 관련 작업을 claim하세요.")
+
+Task(team_name="research-{feature}", name="codebase-patterns", subagent_type="codebase-patterns-agent",
+  prompt="리서치 팀 코드베이스 담당. 리서치 결과: {all_research_results}
+  역할: 권장사항이 기존 코드 패턴과 호환되는지 검증.
+  비호환 발견 시 best-practices에게 SendMessage로 알리세요.
+  TaskList에서 호환성 관련 작업을 claim하세요.")
+
+Task(team_name="research-{feature}", name="framework-docs", subagent_type="framework-docs-agent",
+  prompt="리서치 팀 문서 담당. 리서치 결과: {all_research_results}
+  역할: 최신 API/문서와 권장사항 대조 검증.
+  폐기/변경된 API 발견 시 best-practices에게 SendMessage로 알리세요.
+  TaskList에서 문서 검증 관련 작업을 claim하세요.")
+```
+
+**팀원 간 통신 예시:**
+
+```text
+security-advisory → best-practices: "JWT 라이브러리 권장사항에 CVE-2024-xxxx 취약점. 대안 필요"
+codebase-patterns → best-practices: "기존 코드가 class 패턴인데 함수형 권장은 비호환. 점진적 마이그레이션 제안"
+framework-docs → best-practices: "React 19에서 useEffect 패턴 변경됨. 리서치 결과의 패턴은 구버전"
+best-practices → broadcast: "최종 합의: JWT는 jose 라이브러리로 교체, 함수형 전환은 신규 파일만 적용"
+```
+
+**토론 결과는 SPEC의 Context 섹션에 반영됩니다.**
+
 #### 3.1 Result Merge Rules
 
 | Area | Merge Strategy |
@@ -446,6 +539,52 @@ node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/v
 
 **Research results are reflected in SPEC's Context section.**
 
+### 3.2 UI/UX Design Intelligence (Auto-triggered)
+
+> **조건**: SPEC 컨텍스트에 UI/UX 키워드 포함 시 자동 실행 (website, landing, dashboard, app, e-commerce, portfolio, SaaS, mobile app, web app, UI, UX, frontend, 디자인)
+> **비활성화**: `.claude/vibe/config.json`에 `"uiUxAnalysis": false` 설정
+
+**UI/UX 키워드 감지 시, 리서치와 병렬로 3개 에이전트 순차 실행:**
+
+```
+[Parallel Research] GPT + Gemini + Claude agents
+        ↓ (동시 실행)
+[UI/UX Intelligence]
+  ① ui-industry-analyzer (Haiku) → 산업 분석 + 디자인 전략
+        ↓
+  ②③ 병렬 실행:
+    ② ui-design-system-gen (Sonnet) → MASTER.md 생성
+    ③ ui-layout-architect (Haiku) → 레이아웃 설계
+```
+
+**실행 방법:**
+
+1. **① ui-industry-analyzer** — Task(haiku) 에이전트로 실행:
+```text
+Task(subagent_type="ui-industry-analyzer",
+  prompt="Analyze product: [USER_DESCRIPTION]. Use core_ui_search to detect category, style priority, color mood, typography mood. Save result to .claude/vibe/design-system/{project}/analysis-result.json")
+```
+
+2. **②③ 병렬 실행** — ①의 결과를 입력으로:
+```text
+# ② 디자인 시스템 생성 (Sonnet)
+Task(subagent_type="ui-design-system-gen",
+  prompt="Generate design system from analysis-result.json for project '{project}'. Use core_ui_search for style/color/typography, then core_ui_generate_design_system and core_ui_persist_design_system.")
+
+# ③ 레이아웃 설계 (Haiku) — 병렬 실행
+Task(subagent_type="ui-layout-architect",
+  prompt="Design layout from analysis-result.json for project '{project}'. Use core_ui_search for landing patterns and dashboard layout.")
+```
+
+3. **결과를 SPEC Context에 주입:**
+```markdown
+### Design System (Auto-generated)
+- Category: {①의 category}
+- Style: {①의 style_priority}
+- MASTER.md: .claude/vibe/design-system/{project}/MASTER.md
+- Layout: {③의 pattern + sections}
+```
+
 ### 4. Write SPEC Document (PTCF Structure)
 
 #### 4.0 Large Scope Detection & Auto-Split (MANDATORY)
@@ -854,6 +993,8 @@ Output the handoff message:
 📋 SPEC: .claude/vibe/specs/{feature-name}.md
 📋 Feature: .claude/vibe/features/{feature-name}.feature
 📊 Quality Score: {score}/100
+⏱️ Started: {start_time}
+⏱️ Completed: {getCurrentTime 결과}
 
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 ⚠️ NEXT STEP: Run SPEC review
@@ -1008,7 +1149,7 @@ Claude: Thank you. SPEC has been refined.
 📊 Quality score: 95/100 (A) ← Improved after review
 ```
 
-## Vibe Tools (Semantic Analysis & Memory)
+## Core Tools (Semantic Analysis & Memory)
 
 ### Tool Invocation
 All tools are called via:

package/commands/vibe.spec.review.md

@@ -18,6 +18,8 @@ Review and enhance SPEC with GPT/Gemini cross-validation.
 
 ---
 
+> **⏱️ Timer**: Call `getCurrentTime` tool at the START. Record the result as `{start_time}`.
+
 ## Workflow
 
 ```
@@ -45,7 +47,9 @@ Detect SPEC structure (single file or split folder) and read files:
 **Split folder structure:**
 ```
 .claude/vibe/specs/{feature-name}/_index.md      (+ phase files)
+.claude/vibe/specs/{feature-name}/phase-*.md     (phase-1-xxx.md, phase-2-xxx.md, ...)
 .claude/vibe/features/{feature-name}/_index.feature (+ phase files)
+.claude/vibe/features/{feature-name}/phase-*.feature
 ```
 
 **Detection logic:**
@@ -53,6 +57,11 @@ Detect SPEC structure (single file or split folder) and read files:
 2. Otherwise check `.claude/vibe/specs/{feature-name}.md` → Single mode
 3. If neither exists → Error
 
+**Split mode file loading:**
+1. Read `_index.md` for master SPEC overview
+2. Glob `phase-*.md` files and read all phase SPECs
+3. Read corresponding `_index.feature` and `phase-*.feature` files
+
 **Output:**
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -174,33 +183,44 @@ Score: 96/100 ✅ PASSED
 3. **DO NOT** simulate or fake review results
 4. **YOU MUST** run all 3 rounds sequentially (each round uses updated SPEC)
 
-> Race Mode reviews SPEC with both GPT and Gemini in parallel, then cross-validates findings for higher confidence.
+> Race Mode reviews SPEC with GPT and Gemini in parallel, then cross-validates findings for higher confidence.
 
 ### 3.1 Review Loop (3 Rounds)
 
 **For EACH round (1, 2, 3), run GPT + Gemini in PARALLEL via Bash tool.**
 
-**🚨 IMPORTANT: SPEC content is too large for CLI arguments. Use stdin pipe method.**
+**🚨 IMPORTANT: SPEC content is too large for CLI arguments. Use --input file method (no pipe needed).**
 
 **Procedure for each round:**
 
-**Step A: Save SPEC content to scratchpad temp file (using Write tool):**
-- Write the SPEC content to `[SCRATCHPAD]/spec-content.txt`
+**Step A: Save SPEC content + prompt as JSON to scratchpad temp file (using Write tool):**
+- Write JSON to `[SCRATCHPAD]/spec-review-input.json` with content:
+```json
+{"prompt": "Review this SPEC for completeness, specificity, testability, security, and performance. Round [N]/3. Find issues and improvements. Return JSON: {issues: [{id, title, description, severity, suggestion}]}. SPEC content: [SPEC_CONTENT]"}
+```
+- Where `[SPEC_CONTENT]` is the full SPEC text (properly JSON-escaped inside the prompt string)
+
+**Step B: Resolve script path (one-time per session, using Bash tool):**
+```bash
+node -e "console.log(require('path').join(process.env.APPDATA || require('os').homedir() + '/.config', 'vibe/hooks/scripts/llm-orchestrate.js'))"
+```
+- Save the output as `[LLM_SCRIPT]` for use in all subsequent calls.
 
-**Step B: Run GPT + Gemini in PARALLEL (two separate Bash tool calls at once):**
+**Step C: Run GPT + Gemini in PARALLEL (two separate Bash tool calls at once):**
 
 ```bash
 # GPT review (Bash tool call 1)
-node -e "const fs=require('fs');const p=JSON.stringify({prompt:'Review this SPEC for completeness, specificity, testability, security, and performance. Round [N]/3. Find issues and improvements. Return JSON: {issues: [{id, title, description, severity, suggestion}]}. SPEC content: '+fs.readFileSync('[SCRATCHPAD]/spec-content.txt','utf8')});process.stdout.write(p)" | node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json
+node "[LLM_SCRIPT]" gpt orchestrate-json --input "[SCRATCHPAD]/spec-review-input.json"
 ```
 
 ```bash
 # Gemini review (Bash tool call 2 - run in parallel with GPT)
-node -e "const fs=require('fs');const p=JSON.stringify({prompt:'Review this SPEC for completeness, specificity, testability, security, and performance. Round [N]/3. Find issues and improvements. Return JSON: {issues: [{id, title, description, severity, suggestion}]}. SPEC content: '+fs.readFileSync('[SCRATCHPAD]/spec-content.txt','utf8')});process.stdout.write(p)" | node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json
+node "[LLM_SCRIPT]" gemini orchestrate-json --input "[SCRATCHPAD]/spec-review-input.json"
 ```
 
 **🚨 MANDATORY: Replace `[SCRATCHPAD]` with the actual scratchpad directory path.**
 **🚨 Replace `[N]` with the current round number (1, 2, or 3).**
+**🚨 Replace `[LLM_SCRIPT]` with the resolved absolute path from Step B.**
 **🚨 Run GPT and Gemini calls in PARALLEL (two separate Bash tool calls at once).**
 
 - Round 1: Write SPEC → Run GPT + Gemini in parallel → Cross-validate → Apply fixes → Update SPEC file
@@ -211,16 +231,52 @@ node -e "const fs=require('fs');const p=JSON.stringify({prompt:'Review this SPEC
 
 | Agreement | Priority | Action |
 |-----------|----------|--------|
-| Both GPT + Gemini agree (100%) | P1 | Auto-apply immediately |
-| One model only (50%) | P2 | Auto-apply with note |
+| Both models agree (100%) | P1 | Auto-apply immediately |
+| 1 model only (50%) | P2 | Auto-apply with note |
 
 **After each round:**
 
-1. Cross-validate findings (issues found by both → P1, single model → P2)
+1. Cross-validate findings (issues found by 2+ models → P1, single model → P2)
 2. Merge feedback with confidence scores
 3. Auto-apply P1/P2 improvements to SPEC and Feature files (use Edit tool)
 4. Continue to next round with updated SPEC content
 
+### 3.3 User Decision Checkpoint (Round 3 이후)
+
+**🚨 MANDATORY: 3라운드 완료 후 사용자 판단 체크포인트 실행**
+
+> Type 6 (Iterative-Reasoning) 패턴: AI가 혼자 결정하지 않고, 사용자와 함께 판단
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🔍 USER CHECKPOINT: 리뷰 결과 검토
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+3라운드 리뷰에서 발견된 주요 변경사항:
+
+| # | 변경 내용 | 출처 | 신뢰도 |
+|---|----------|------|--------|
+| 1 | {변경1} | GPT+Gemini | 100% |
+| 2 | {변경2} | GPT only | 50% |
+| ... | ... | ... | ... |
+
+질문:
+1. 위 변경사항 중 제외하고 싶은 항목이 있나요?
+2. 추가로 명시해야 할 요구사항이 있나요?
+3. 기술적 접근 방식에 동의하시나요?
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**Checkpoint 행동 규칙:**
+
+| 상황 | 행동 |
+|------|------|
+| `ultrawork` 모드 | 체크포인트 스킵, 자동 진행 |
+| 일반 모드 | 반드시 사용자 응답 대기 |
+| 사용자가 변경 요청 | 수정 후 다시 체크포인트 |
+| 사용자가 승인 | Step 4로 진행 |
+
 **Output format:**
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -276,6 +332,66 @@ Cross-Validated Issues: None
 
 ---
 
+## Step 3.5: Review Debate Team (Agent Teams)
+
+> **조건**: Agent Teams 활성화 + 3라운드 완료 후 P1/P2 이슈 2개 이상 발견 시
+> 여러 관점에서 교차 검증하여 오탐 제거 및 우선순위 조정
+
+**팀 구성:**
+
+| 팀원 | 역할 |
+|------|------|
+| security-reviewer (리더) | 보안 관점 우선순위 결정, 교차 검증 주도 |
+| architecture-reviewer | 구조적 영향 평가, 설계 일관성 확인 |
+| performance-reviewer | 성능 영향 평가, 불필요한 최적화 식별 |
+| simplicity-reviewer | 과잉 설계 여부, YAGNI 위반 식별 |
+
+**활성화 조건:**
+
+| 상황 | 행동 |
+|------|------|
+| P1/P2 이슈 2개 이상 | 자동 활성화 |
+| P1/P2 이슈 1개 이하 | 스킵 → Step 4로 진행 |
+| Agent Teams 비활성화 | 스킵 → Step 4로 진행 |
+
+**spawn 패턴:**
+
+```text
+TeamCreate(team_name="spec-debate-{feature}", description="SPEC review debate for {feature}")
+
+Task(team_name="spec-debate-{feature}", name="security-reviewer", subagent_type="security-reviewer",
+  prompt="SPEC 리뷰 팀 리더. GPT/Gemini가 발견한 P1/P2 이슈를 교차 검증하세요.
+  SPEC: {spec_content}
+  발견된 이슈: {p1_p2_issues}
+  역할: 각 이슈가 진짜인지(오탐 아닌지) 검증. 실제 영향도 기준으로 우선순위 조정.
+  분쟁이 있는 이슈는 해당 리뷰어에게 SendMessage로 확인 요청하세요.")
+
+Task(team_name="spec-debate-{feature}", name="architecture-reviewer", subagent_type="architecture-reviewer",
+  prompt="SPEC 리뷰 팀 아키텍처 전문가. 아키텍처 관련 이슈를 검증하세요.
+  SPEC: {spec_content}
+  역할: 구조적 일관성, SOLID 원칙, 레이어 경계 검증.
+  우선순위 변경 필요 시 security-reviewer에게 SendMessage로 알리세요.")
+
+Task(team_name="spec-debate-{feature}", name="performance-reviewer", subagent_type="performance-reviewer",
+  prompt="SPEC 리뷰 팀 성능 전문가. 성능 관련 이슈를 검증하세요.
+  SPEC: {spec_content}
+  역할: 성능 요구사항 현실성 평가, 불필요한 최적화 식별.
+  P2→P1 승격이 필요하면 security-reviewer에게 SendMessage로 알리세요.")
+
+Task(team_name="spec-debate-{feature}", name="simplicity-reviewer", subagent_type="simplicity-reviewer",
+  prompt="SPEC 리뷰 팀 단순성 전문가. 과잉 설계 여부를 검증하세요.
+  SPEC: {spec_content}
+  역할: YAGNI 위반, 불필요한 복잡성 식별.
+  P1→P2 강등이 필요하면 security-reviewer에게 SendMessage로 알리세요.")
+```
+
+**결과 통합:**
+- 팀 합의 결과를 SPEC에 반영 (P1 즉시 적용, P2 노트 추가)
+- 팀원 shutdown_request → TeamDelete로 정리
+- Step 4 (Final Summary)로 진행
+
+---
+
 ## Step 4: Final Summary
 
 ```
@@ -286,6 +402,8 @@ Cross-Validated Issues: None
 Quality Score: 96/100 ✅
 Review Rounds: 3/3 ✅
 Total Improvements: 4
+⏱️ Started: {start_time}
+⏱️ Completed: {getCurrentTime 결과}
 
 Updated files:
   📋 .claude/vibe/specs/{feature-name}.md (or split folder)
@@ -344,6 +462,33 @@ If no issues, proceed with /vibe.run "{feature-name}".
 - Wait for user confirmation after review (unless ultrawork mode)
 - In ultrawork mode: output summary then auto-proceed to `/vibe.run`
 
+### 5.1 Final User Checkpoint
+
+**🚨 MANDATORY: `/vibe.run` 진행 전 최종 사용자 확인**
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+✅ SPEC 리뷰 완료 - 최종 확인
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+다음 질문에 답변해 주세요:
+
+1. **요구사항 정확성**: 위 SPEC이 원래 의도한 기능을 정확히 설명하고 있나요?
+2. **범위 적절성**: 구현 범위가 너무 크거나 작지 않나요?
+3. **기술 스택**: 선택된 기술 스택에 동의하시나요?
+4. **우선순위**: Phase 순서와 우선순위가 맞나요?
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+💡 "ok" 또는 "진행"으로 승인 / 수정 사항이 있으면 말씀해 주세요
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+**Why This Checkpoint Matters:**
+
+> AI가 더 많이 해준다고 좋은 게 아닙니다.
+> 사용자가 AI와 함께 생각하고 판단할 때 최고의 결과가 나옵니다.
+> 이 체크포인트는 Type 6 (Iterative-Reasoning) 패턴을 유도합니다.
+
 ---
 
 ## Next Step

package/commands/vibe.trace.md

@@ -10,6 +10,8 @@ Generate and display Requirements Traceability Matrix (RTM).
 /vibe.trace "feature-name" --save       # Save to file
 ```
 
+> **⏱️ Timer**: Call `getCurrentTime` tool at the START. Record the result as `{start_time}`.
+
 ## Description
 
 Requirements Traceability Matrix (RTM) tracks the relationship between:
@@ -22,12 +24,55 @@ Requirements Traceability Matrix (RTM) tracks the relationship between:
 
 ### 1. Load Files
 
+**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)
+├── phase-1-{name}.md      → Phase 1 SPEC
+└── ...
+
+📁 .claude/vibe/features/{feature-name}/
+├── _index.feature         → Master Feature (read first)
+├── phase-1-{name}.feature → Phase 1 scenarios
+└── ...
+
+→ Load all phase files, generate RTM across all phases
+```
+
+**Single file detected:**
 ```
 📄 .claude/vibe/specs/{feature-name}.md      → SPEC
 📄 .claude/vibe/features/{feature-name}.feature → Feature
+```
+
+**Additional (auto-detect):**
+```
 📄 src/**/*.test.ts                           → Tests (auto-detect)
 ```
 
+**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.
+```
+
 ### 2. Extract Mappings
 
 For each requirement ID (REQ-xxx-xxx):
@@ -85,6 +130,9 @@ The following requirements lack full coverage:
 - Add Feature scenarios for uncovered requirements
 - Add test cases for uncovered requirements
 
+⏱️ Started: {start_time}
+⏱️ Completed: {getCurrentTime 결과}
+
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 ```
 
@@ -129,7 +177,7 @@ Loading files:
 
 ## Tool Integration
 
-The RTM generation uses vibe tools:
+The RTM generation uses core tools:
 
 ```bash
 # Generate RTM

package/commands/vibe.utils.md

@@ -33,7 +33,7 @@ Read and follow `agents/ui-previewer.md` for UI preview generation.
 
 Generate UI preview from description or design folder.
 
-- **Gemini enabled**: Actual UI mockup image
+- **Gemini enabled**: UI mockup image + component code generation (via `llm-orchestrate.js` → `gemini-api`)
 - **Gemini disabled**: ASCII art fallback
 
 **Input types:**
@@ -84,20 +84,34 @@ Generate Mermaid diagrams for architecture visualization.
 
 ## --image (Image Generation)
 
-Generate images using Gemini Image API (Nano Banana model).
+Generate images using Gemini Image API.
+
+- **Default**: Nano Banana (`gemini-2.5-flash-image`) - 빠르고 가벼운 이미지 생성
+- **--pro**: Nano Banana Pro (`gemini-3-pro-image-preview`) - 고품질 이미지 생성
 
 ### MANDATORY Tool Invocation
 
 **CRITICAL: You MUST use the following command to generate images. Do NOT search for scripts, do NOT use gcloud, do NOT use any other method.**
 
-**General image generation:**
+**Step 0: Resolve script path (once per session):**
+```bash
+node -e "console.log(require('path').join(process.env.APPDATA || require('os').homedir() + '/.config', 'vibe/hooks/scripts/llm-orchestrate.js'))"
+```
+Save the output as `[LLM_SCRIPT]` for use in all subsequent calls.
+
+**General image generation (Nano Banana):**
+```bash
+node "[LLM_SCRIPT]" gemini image "IMAGE_DESCRIPTION" --output "OUTPUT_PATH"
+```
+
+**Pro quality (Nano Banana Pro):**
 ```bash
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "IMAGE_DESCRIPTION" --output "OUTPUT_PATH"
+node "[LLM_SCRIPT]" gemini image "IMAGE_DESCRIPTION" --pro --output "OUTPUT_PATH"
 ```
 
 **With size option:**
 ```bash
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "IMAGE_DESCRIPTION" --size "1920x1080" --output "OUTPUT_PATH"
+node "[LLM_SCRIPT]" gemini image "IMAGE_DESCRIPTION" --size "1920x1080" --output "OUTPUT_PATH"
 ```
 
 ### How to Parse User Request
@@ -107,15 +121,16 @@ node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/v
    - If user specifies a path, use that path
    - Default: `--output "./generated-image.png"`
 3. Extract **size** if specified, otherwise default 1024x1024
+4. If user requests **high quality / pro / 고품질**, add `--pro` flag
 
 ### Examples
 
 ```bash
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "A cute Gemini AI character mascot, colorful, friendly" --output "./gemini-character.png"
+node "[LLM_SCRIPT]" gemini image "A cute Gemini AI character mascot, colorful, friendly" --output "./gemini-character.png"
 
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "Professional website banner, modern design" --size "1920x400" --output "./banner.png"
+node "[LLM_SCRIPT]" gemini image "Professional website banner, modern design" --pro --size "1920x400" --output "./banner.png"
 
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "Modern minimal logo design" --output "./public/logo.png"
+node "[LLM_SCRIPT]" gemini image "Modern minimal logo design" --output "./public/logo.png"
 ```
 
 ### Output Format
@@ -138,10 +153,10 @@ Use the same command with a pre-built icon prompt template:
 
 ```bash
 # --icon "MyApp"
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini image "App icon for MyApp, 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"
+node "[LLM_SCRIPT]" gemini image "App icon for MyApp, 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"
 
 # --icon "MyApp" --color "#2F6BFF"
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" 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"
+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"
 ```
 
 ### Prerequisites
@@ -295,7 +310,7 @@ Restore previous session context for continuity.
 ```
 
 **What it does:**
-1. Calls `vibe_start_session` to load project memories
+1. Calls `core_start_session` to load project memories
 2. Restores previous conversation context
 3. Resumes work from last checkpoint