@su-record/vibe 2.6.35 → 2.6.41

package/.env

@@ -0,0 +1,62 @@
+# ─── Telegram ────────────────────────────────────────────────────────
+TELEGRAM_BOT_TOKEN=8351914914:AAEwjQHYG7hZA1_xVPpHFTI5uS4t9qF3Tm8
+TELEGRAM_ALLOWED_CHAT_IDS=8382889321
+
+# ─── Discord ─────────────────────────────────────────────────────────
+DISCORD_BOT_TOKEN=
+DISCORD_ALLOWED_GUILD_IDS=
+DISCORD_ALLOWED_CHANNEL_IDS=
+
+# ─── Slack ───────────────────────────────────────────────────────────
+SLACK_BOT_TOKEN=
+SLACK_APP_TOKEN=
+SLACK_ALLOWED_CHANNELS=
+
+
+# ─── Gemini (API key + Search grounding) ────────────────────────────
+GEMINI_API_KEY=AIzaSyAi8fYbd5tlYn7nBYoKcaLJI0u-Z7Nlw1U
+
+# ─── OpenAI ──────────────────────────────────────────────────────────
+OPENAI_API_KEY=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
+
+# ─── GPT OAuth (OpenAI Codex CLI PKCE) ──────────────────────────────
+# client_id: app_EMoamEEZ73f0CkXaXp7hrann
+# authorize: https://auth.openai.com/oauth/authorize
+# token:     https://auth.openai.com/oauth/token
+# redirect:  http://localhost:1455/auth/callback
+# scope:     openid profile email offline_access
+GPT_OAUTH_REFRESH_TOKEN=rt_RW_twFiROvrUP612RH9nL68pZyOZH_Vyz3x4qLEesRE.gX6Hswde_zcubOF1jJpvROSmSr-pmXkL5ef-wg9pihE
+
+# ─── Gemini OAuth (via Gemini CLI — @google/gemini-cli) ──────────────
+# credentials:   ~/.gemini/oauth_creds.json (auto-detected)
+# client_id:     681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com
+# auth order:    gemini-cli → apikey
+GEMINI_OAUTH_REFRESH_TOKEN=1//0eTG9YDGVchVOCgYIARAAGA4SNwF-L9IryFIeIUvne7lpgRK8QFUgRXuvL55MBHMzDnxsXDbltWwy4H38TMY6zVKcQM-3qgILcb4
+
+# ─── Model Overrides (optional - uses hardcoded defaults if not set) ─
+# Claude (역할별 — Claude Code 네이티브)
+CLAUDE_ARCHITECTURE_MODEL=claude-opus-4-6       # 아키텍처, 복잡한 추론
+CLAUDE_RESEARCH_MODEL=claude-haiku-4-5           # 리서치, 탐색 (빠름)
+CLAUDE_REVIEW_MODEL=claude-haiku-4-5             # 리뷰 (빠름)
+CLAUDE_BACKGROUND_MODEL=claude-sonnet-4-5       # 구현, 백그라운드 작업
+
+# GPT (Chat/Reasoning — OAuth: gpt-5.3-codex, OAuth Pro: codex-spark)
+GPT_MODEL=gpt-5.2
+# Plus Plan Use: gpt-5.3-codex
+# Pro Plan Use: gpt-5.3-codex-spark
+
+# Gemini (API Key + Gemini CLI 공용)
+GEMINI_MODEL=gemini-3-pro-preview
+GEMINI_FLASH_MODEL=gemini-3-flash-preview
+GEMINI_SEARCH_MODEL=gemini-3-flash-preview
+
+GEMINI_OAUTH_MODEL=gemini-3-pro-high
+GEMINI_OAUTH_FLASH_MODEL=gemini-3-flash
+
+# Embedding (API Key 전용 — OAuth 미지원)
+EMBEDDING_MODEL=text-embedding-3-large
+GEMINI_EMBEDDING_MODEL=text-embedding-004
+
+# ─── Workspace ───────────────────────────────────────────────────────
+WORKSPACE_DIR=C:/Users/endba/WorkSpace
+

package/.env.example

@@ -0,0 +1,54 @@
+# ─── Telegram ────────────────────────────────────────────────────────
+TELEGRAM_BOT_TOKEN=
+TELEGRAM_ALLOWED_CHAT_IDS=
+
+# ─── Discord ─────────────────────────────────────────────────────────
+DISCORD_BOT_TOKEN=
+DISCORD_ALLOWED_GUILD_IDS=
+DISCORD_ALLOWED_CHANNEL_IDS=
+
+# ─── Slack ───────────────────────────────────────────────────────────
+SLACK_BOT_TOKEN=
+SLACK_APP_TOKEN=
+SLACK_ALLOWED_CHANNELS=
+
+# ─── Gemini (API key + Search grounding) ────────────────────────────
+GEMINI_API_KEY=
+
+# ─── OpenAI ──────────────────────────────────────────────────────────
+OPENAI_API_KEY=
+
+# ─── GPT OAuth (OpenAI Codex CLI PKCE) ──────────────────────────────
+# Run: vibe gpt auth
+GPT_OAUTH_REFRESH_TOKEN=
+
+# ─── Gemini OAuth (Google Antigravity PKCE) ─────────────────────────
+# Run: vibe gemini auth
+GEMINI_OAUTH_REFRESH_TOKEN=
+
+# ─── Model Overrides (optional - uses hardcoded defaults if not set) ─
+# Claude (역할별 — Claude Code 네이티브)
+CLAUDE_ARCHITECTURE_MODEL=claude-opus-4-6       # 아키텍처, 복잡한 추론
+CLAUDE_RESEARCH_MODEL=claude-haiku-4-5           # 리서치, 탐색 (빠름)
+CLAUDE_REVIEW_MODEL=claude-haiku-4-5             # 리뷰 (빠름)
+CLAUDE_BACKGROUND_MODEL=claude-sonnet-4-5       # 구현, 백그라운드 작업
+
+# GPT (Chat/Reasoning — OAuth: gpt-5.3-codex, OAuth Pro: codex-spark)
+GPT_MODEL=gpt-5.2
+# Plus Plan Use: gpt-5.3-codex
+# Pro Plan Use: gpt-5.3-codex-spark
+
+# Gemini
+GEMINI_MODEL=gemini-3-pro-preview
+GEMINI_FLASH_MODEL=gemini-3-flash-preview
+GEMINI_SEARCH_MODEL=gemini-3-flash-preview
+
+GEMINI_OAUTH_MODEL=gemini-3-pro-high
+GEMINI_OAUTH_FLASH_MODEL=gemini-3-flash
+
+# Embedding (API Key 전용 — OAuth 미지원)
+EMBEDDING_MODEL=text-embedding-3-large
+GEMINI_EMBEDDING_MODEL=text-embedding-004
+
+# ─── Workspace ───────────────────────────────────────────────────────
+WORKSPACE_DIR=

package/CLAUDE.md

@@ -1,43 +1,41 @@
 # VIBE
 
-SPEC-driven AI Coding Framework (Claude Code Exclusive)
+Personalized AI Agent (Claude Code Exclusive) — v0.1.0
+
+## Project Nature
+
+> **This project is the source code for the `@su-record/vibe` npm package.**
+> Since users install this via `npm install`, all bug fixes and improvements MUST be made in **this project source** — not in installed locations (e.g., AppData).
+> Always modify files in this repository, never the installed copy.
 
 ## Philosophy
 
-> **vibe = 바이브코딩을 쉽게 + 최소 품질 보장**
+> **core = Easy vibe coding + Minimum quality guaranteed**
 
 | Principle | Description |
 |-----------|-------------|
-| **Easy Vibe Coding** | 빠른 흐름, 직관적 개발, AI와 협업하며 생각하기 |
-| **Minimum Quality Guaranteed** | 타입 안전성, 코드 품질, 보안 - 자동으로 하한선 확보 |
-| **Iterative Reasoning (6번 유형)** | AI에게 답을 맡기지 말고, 문제를 쪼개고 질문하며 함께 추론 |
-
-### How vibe Guarantees Quality
-
-```
-┌─────────────────────────────────────────────────────────────┐
-│  VIBE QUALITY GUARDRAILS (자동 품질 보장)                    │
-├─────────────────────────────────────────────────────────────┤
-│  1. Type Safety      → Quality Gate (any/Any 차단)          │
-│  2. Code Review      → Race Review (GPT + Gemini 병렬)      │
-│  3. Completion Check → Ralph Loop (100%까지 반복)           │
-│  4. Multi-LLM        → 3개 관점 검증 (Claude+GPT+Gemini)    │
-└─────────────────────────────────────────────────────────────┘
-```
+| **Easy Vibe Coding** | Fast flow, intuitive development, think collaboratively with AI |
+| **Minimum Quality Guaranteed** | Type safety, code quality, security — automatic baseline enforcement |
+| **Iterative Reasoning (Type 6)** | Don't delegate answers to AI — break down problems, ask questions, reason together |
 
-### User's Role (6번 Iterative-Reasoning Type)
+### How CORE Guarantees Quality
 
-연구에 따르면, AI에게 답을 맡기는 것(1번 유형)보다 **문제를 쪼개고 추론하며 협업하는 것(6번 유형)**이 훨씬 좋은 성과를 냅니다.
+| Guardrail | Mechanism |
+|-----------|-----------|
+| Type Safety | Quality Gate — blocks `any` / `Any` / `@ts-ignore` |
+| Code Review | Claude Code native parallel review (13+ agents) |
+| Completion Check | Ralph Loop — iterate until 100% complete (no scope reduction) |
+| Memory RAG | SQLite + FTS5 BM25 persistent memory — gets smarter over time |
 
-| ❌ Avoid | ✅ Do Instead |
-|----------|---------------|
-| "로그인 기능 만들어줘" | "로그인 기능의 요구사항을 분석해보자" |
-| AI 결과 그대로 사용 | "이 접근이 맞나?" 검증 질문 |
-| 완성 코드만 요청 | 단계별로 검토하며 진행 |
+### User's Role (Iterative-Reasoning Type)
 
-## Response Language
+Research shows that **breaking down problems and reasoning collaboratively (Type 6)** produces far better results than simply delegating to AI (Type 1).
 
-**IMPORTANT: Always respond in Korean (한국어) unless the user explicitly requests otherwise.**
+| Avoid | Do Instead |
+|-------|------------|
+| "Build login feature" | "Let's analyze the requirements for login" |
+| Accept AI output as-is | Ask "Is this approach correct?" |
+| Request complete code only | Review step by step |
 
 ## Code Quality Standards (Mandatory)
 
@@ -45,26 +43,27 @@ Follow these standards when writing code. See `~/.claude/vibe/rules/` (global) f
 
 ### Core Principles
 
-- **Modify only requested scope** - Don't touch unrelated code
-- **Preserve existing style** - Follow project conventions
-- **Keep working code** - No unnecessary refactoring
-- **Respect user interrupts** - If user interrupts (Ctrl+C/Escape) and sends a new message, the previous task is CANCELLED. Do NOT resume or continue interrupted work. Respond ONLY to the new message.
+- **Modify only requested scope** — Don't touch unrelated code
+- **Preserve existing style** — Follow project conventions
+- **Keep working code** — No unnecessary refactoring
+- **Edit existing files, never create new ones** — When fixing errors/bugs, ALWAYS modify the original file. NEVER create new files (wrappers, adapters, V2 copies) as a workaround. Fix the problem at its source.
+- **Respect user interrupts** — If user interrupts (Ctrl+C/Escape) and sends a new message, the previous task is CANCELLED. Do NOT resume or continue interrupted work. Respond ONLY to the new message.
 
 ### Code Complexity Limits
 
-| Metric               | Limit                                        |
-|----------------------|----------------------------------------------|
-| Function length      | ≤30 lines (recommended), ≤50 lines (allowed) |
-| Nesting depth        | ≤3 levels                                    |
-| Parameters           | ≤5                                           |
-| Cyclomatic complexity | ≤10                                         |
+| Metric | Limit |
+|--------|-------|
+| Function length | ≤30 lines (recommended), ≤50 lines (allowed) |
+| Nesting depth | ≤3 levels |
+| Parameters | ≤5 |
+| Cyclomatic complexity | ≤10 |
 
 ### TypeScript Rules
 
-- No `any` → `unknown` + type guards
-- No `as any` → proper interfaces
-- No `@ts-ignore`
-- Explicit return types
+- No `any` type — use `unknown` + type guards
+- No `as any` casting — define proper interfaces
+- No `@ts-ignore` — fix type issues at root
+- Explicit return types on all functions
 
 ### Error Handling Required
 
@@ -79,24 +78,39 @@ Follow these standards when writing code. See `~/.claude/vibe/rules/` (global) f
 - No commented-out code
 - No incomplete code without TODO
 
+### Superpowers Integration (v0.2.0)
+
+Four quality mechanisms inspired by the Superpowers project:
+
+| Feature | Description | Location |
+|---------|-------------|----------|
+| **Anti-Rationalization** | Blocks 6 categories of AI excuse patterns (SPEC skip, quality skip, verification skip, scope expansion, architecture avoidance, evidence avoidance) | `rules/principles/anti-rationalization.md` |
+| **Evidence Gate** | 5-step verification protocol (IDENTIFY → RUN → READ → VERIFY → CLAIM). No completion claims without evidence | `rules/quality/evidence-gate.md` + UltraQA |
+| **3-Fix Rule** | Same failure 3 times → switch to architecture question. User discussion required before fix #4 | UltraQA `architecture_question` status |
+| **Rule TDD** | 6 quality checks (example/both-sides/rationalization/edge/quality/comprehensive), 70+ pass threshold | RuleBuildSystem `pressureTestRule()` |
+
+**Core principle:** Violating the letter of a rule is violating the spirit of a rule.
+
 ## Workflow
 
 ```text
-/vibe.spec → /new → /vibe.spec.review → /vibe.run → /vibe.trace → (auto) code review → ✅ Done
+/vibe.spec → /new → /vibe.spec.review → /vibe.run → /vibe.trace → (auto) code review → Done
 ```
 
-1. `/vibe.spec` - Write SPEC (requirements + research + draft)
-2. `/new` - Start new session (clean context)
-3. `/vibe.spec.review` - GPT/Gemini review (3-round mandatory)
-4. `/vibe.run` - Implementation + Gemini review
-5. **(auto)** 13+ agent parallel review + P1/P2 auto-fix
+1. `/vibe.spec` — Write SPEC (requirements + research + draft)
+2. `/new` — Start new session (clean context)
+3. `/vibe.spec.review` — SPEC quality review
+4. `/vibe.run` — Implementation + parallel agent review
+5. **(auto)** — 13+ agent parallel review + P1/P2 auto-fix
 
-## Plan Mode vs VIBE
+**All commands call `getCurrentTime` at start/end to display elapsed time.**
 
-| Task Size                    | Recommended  |
-|------------------------------|--------------|
-| Simple changes (1-2 files)   | Plan Mode    |
-| Complex features (3+ files)  | `/vibe.spec` |
+## Plan Mode vs CORE
+
+| Task Size | Recommended |
+|-----------|-------------|
+| Simple changes (1-2 files) | Plan Mode |
+| Complex features (3+ files) | `/vibe.spec` |
 
 After `/vibe.analyze` or `/vibe.review` with dev request → **Ask user for workflow choice**
 
@@ -111,101 +125,303 @@ Include `ultrawork` or `ulw` keyword for maximum performance:
 
 ## Commands
 
-| Command                      | Description                       |
-|------------------------------|-----------------------------------|
-| `/vibe.spec "name"`          | Write SPEC (PTCF) + parallel research |
-| `/vibe.spec.review "name"`   | GPT/Gemini review (new session)   |
-| `/vibe.run "name"`           | Execute implementation            |
-| `/vibe.run "name" ultrawork` | Maximum performance mode          |
-| `/vibe.verify "name"`        | Verification against SPEC         |
-| `/vibe.review`               | 13+ agent parallel code review    |
-| `/vibe.trace "name"`         | Requirements traceability matrix  |
-| `/vibe.reason "problem"`     | Systematic reasoning              |
-| `/vibe.analyze`              | Project analysis                  |
-| `/vibe.utils --e2e`          | E2E testing (Playwright)          |
-| `/vibe.utils --diagram`      | Generate diagrams                 |
-| `/vibe.utils --ui "desc"`    | UI preview                        |
-| `/vibe.utils --continue`     | Session restore                   |
+| Command | Description |
+|---------|-------------|
+| `/vibe.spec "name"` | Write SPEC (PTCF) + parallel research |
+| `/vibe.spec.review "name"` | SPEC quality review |
+| `/vibe.run "name"` | Execute implementation |
+| `/vibe.run "name" ultrawork` | Maximum performance mode |
+| `/vibe.verify "name"` | Verification against SPEC |
+| `/vibe.review` | 13+ agent parallel code review |
+| `/vibe.trace "name"` | Requirements traceability matrix |
+| `/vibe.reason "problem"` | Systematic reasoning |
+| `/vibe.analyze` | Project analysis |
+| `/vibe.utils --e2e` | E2E testing (Playwright) |
+| `/vibe.utils --diagram` | Generate diagrams |
+| `/vibe.utils --ui "desc"` | UI preview |
+| `/vibe.utils --continue` | Session restore |
+| `/vibe.voice` | Voice-to-coding command (Gemini audio) |
+
+## CLI Commands
+
+| Command | Description |
+|---------|-------------|
+| `vibe init [project]` | Project initialization |
+| `vibe setup` | Interactive setup wizard (LLM auth + channels) |
+| `vibe update` | Update project configuration |
+| `vibe upgrade` | Upgrade to latest version |
+| `vibe remove` | Remove from project |
+| `vibe status` | Full status check (LLM, channels) |
+| `vibe version` | Show version |
+| `vibe claude <cmd>` | Claude (key, status, logout) |
+| `vibe gpt <cmd>` | GPT (auth, key, status, logout, remove) |
+| `vibe gemini <cmd>` | Gemini (auth, key, import, status, logout, remove) |
+| `vibe sync <cmd>` | Cloud sync (login, push, pull, status, logout) |
+| `vibe telegram <cmd>` | Telegram (setup, chat, status) |
+| `vibe slack <cmd>` | Slack (setup, channel, status) |
+
+## External Channels
+
+| Channel | Interface | Requirements |
+|---------|-----------|-------------|
+| Telegram | TelegramBot (polling) | `TELEGRAM_BOT_TOKEN` |
+| Slack | SlackBot (Socket Mode) | `SLACK_BOT_TOKEN`, `SLACK_APP_TOKEN` |
+
+### Channel Environment Variables
+
+| Variable | Description |
+|----------|-------------|
+| `VIBE_TELEGRAM_ENABLED` | Enable Telegram channel |
+| `VIBE_SLACK_ENABLED` | Enable Slack channel |
+| `SLACK_BOT_TOKEN` | Slack bot token (xoxb-) |
+| `SLACK_APP_TOKEN` | Slack app-level token (xapp-) |
+| `SLACK_ALLOWED_CHANNELS` | Comma-separated allowed channel IDs |
 
 ## Magic Keywords
 
-| Keyword              | Effect                                        |
-|----------------------|-----------------------------------------------|
-| `ultrawork` / `ulw`  | Parallel + auto-continue + Ralph Loop         |
-| `ralph`              | Iterate until 100% complete (no scope reduction) |
-| `ralplan`            | Iterative planning + persistence              |
-| `verify`             | Strict verification mode                      |
-| `quick`              | Fast mode, minimal verification               |
+| Keyword | Effect |
+|---------|--------|
+| `ultrawork` / `ulw` | Parallel + auto-continue + Ralph Loop |
+| `ralph` | Iterate until 100% complete (no scope reduction) |
+| `ralplan` | Iterative planning + persistence |
+| `verify` | Strict verification mode |
+| `quick` | Fast mode, minimal verification |
 
 ## PTCF Structure
 
 SPEC documents use: `<role>` `<context>` `<task>` `<constraints>` `<output_format>` `<acceptance>`
 
-## Built-in Tools
+## Built-in Tools (41+)
+
+### Memory & Session
+
+| Tool | Purpose |
+|------|---------|
+| `core_start_session` | Restore previous session context |
+| `core_auto_save_context` | Save current state |
+| `core_save_memory` | Save important decisions |
+| `core_recall_memory` | Recall saved memories |
+| `core_list_memories` | List all memories |
+| `core_search_memories` | Search memories |
+| `core_search_memories_advanced` | Advanced memory search |
+| `core_link_memories` | Link related memories |
+| `core_get_memory_graph` | Visualize memory graph |
+| `core_create_memory_timeline` | Create memory timeline |
+| `core_restore_session_context` | Restore session context |
+| `core_prioritize_memory` | Prioritize memory items |
 
 ### Semantic & Quality
 
-| Tool                          | Purpose                          |
-|-------------------------------|----------------------------------|
-| `vibe_find_symbol`            | Find symbol definitions          |
-| `vibe_find_references`        | Find references                  |
-| `vibe_analyze_complexity`     | Analyze complexity               |
-| `vibe_validate_code_quality`  | Validate quality                 |
+| Tool | Purpose |
+|------|---------|
+| `core_find_symbol` | Find symbol definitions |
+| `core_find_references` | Find references |
+| `core_analyze_dependency_graph` | Analyze dependency graph |
+| `core_analyze_complexity` | Analyze complexity |
+| `core_validate_code_quality` | Validate quality |
+| `core_check_coupling_cohesion` | Check coupling/cohesion |
+| `core_suggest_improvements` | Suggest improvements |
+| `core_apply_quality_rules` | Apply quality rules |
 
-### Memory & Session
+### UI & Utility
+
+| Tool | Purpose |
+|------|---------|
+| `core_preview_ui_ascii` | UI preview in ASCII |
+| `core_get_current_time` | Get current time |
+
+### Channel Tools
+
+| Tool | Purpose |
+|------|---------|
+| `send_slack` | Send message to Slack channel |
 
-| Tool                          | Purpose                          |
-|-------------------------------|----------------------------------|
-| `vibe_start_session`          | Restore previous session context |
-| `vibe_auto_save_context`      | Save current state               |
-| `vibe_save_memory`            | Save important decisions         |
+### SPEC & Testing
 
-### Session RAG (v2.6.27)
+| Tool | Purpose |
+|------|---------|
+| `core_spec_generator` | Generate SPEC documents |
+| `core_prd_parser` | Parse PRD documents |
+| `core_traceability_matrix` | Generate traceability matrix |
+| `core_e2e_test_generator` | Generate E2E tests |
 
-구조화된 세션 컨텍스트를 저장/검색하는 시스템. SQLite + FTS5 BM25 하이브리드 검색.
+### Session RAG
 
-| Tool                          | Purpose                          |
-|-------------------------------|----------------------------------|
-| `save_session_item`           | Decision/Constraint/Goal/Evidence 저장 |
-| `retrieve_session_context`    | 하이브리드 검색 (BM25 + recency + priority) |
-| `manage_goals`                | Goal 생명주기 관리 (list/update/complete) |
+Structured session context storage/retrieval system. SQLite + FTS5 BM25 hybrid search.
 
-**4가지 엔티티:**
+| Tool | Purpose |
+|------|---------|
+| `save_session_item` | Store Decision/Constraint/Goal/Evidence |
+| `retrieve_session_context` | Hybrid search (BM25 + recency + priority) |
+| `manage_goals` | Goal lifecycle management (list/update/complete) |
+
+**4 Entity Types:**
 
 | Entity | Description | Key Fields |
 |--------|-------------|------------|
-| Decision | 사용자 확인 결정사항 | title, rationale, alternatives, impact, priority |
-| Constraint | 명시적 제약조건 | title, type (technical/business/resource/quality), severity |
-| Goal | 현재 목표 스택 (계층 지원) | title, status, priority, progressPercent, successCriteria |
-| Evidence | 검증/테스트 결과 | title, type (test/build/lint/coverage), status, metrics |
+| Decision | User-confirmed decisions | title, rationale, alternatives, impact, priority |
+| Constraint | Explicit constraints | title, type (technical/business/resource/quality), severity |
+| Goal | Current goal stack (hierarchical) | title, status, priority, progressPercent, successCriteria |
+| Evidence | Verification/test results | title, type (test/build/lint/coverage), status, metrics |
 
-**자동 주입:** `start_session` 호출 시 활성 Goals, 중요 Constraints, 최근 Decisions가 자동으로 세션 컨텍스트에 포함됨.
+**Auto-injection:** When `start_session` is called, active Goals, critical Constraints, and recent Decisions are automatically included in session context.
 
 ```typescript
 import { saveSessionItem, retrieveSessionContext, manageGoals } from '@su-record/vibe/tools';
 
-// 결정 저장
+// Save decision
 await saveSessionItem({ itemType: 'decision', title: 'Use Vitest', rationale: 'Fast and modern' });
 
-// 제약 저장
+// Save constraint
 await saveSessionItem({ itemType: 'constraint', title: 'No vector DB', type: 'technical', severity: 'high' });
 
-// 목표 저장
+// Save goal
 await saveSessionItem({ itemType: 'goal', title: 'Implement Session RAG', priority: 2 });
 
-// 컨텍스트 검색
+// Search context
 await retrieveSessionContext({ query: 'testing' });
 
-// 목표 관리
+// Manage goals
 await manageGoals({ action: 'list' });
 await manageGoals({ action: 'update', goalId: 1, progressPercent: 80 });
 await manageGoals({ action: 'complete', goalId: 1 });
 ```
 
+## External LLM Integration (v0.1.0)
+
+GPT and Gemini external LLM support via `llm-orchestrate.js` hook script.
+
+### Core Modules
+
+| Module | Purpose |
+|--------|---------|
+| `SmartRouter` | Task-type-specific LLM selection + fallback chain |
+| `LLMCluster` | Parallel multi-LLM calls (GPT + Gemini) |
+| `AgentRegistry` | SQLite-based agent execution tracking (WAL mode) |
+
+### External LLM Usage
+
+| Role | Provider | Method |
+|------|----------|--------|
+| SPEC review (cross-validation) | GPT + Gemini | `llm-orchestrate.js` via Bash |
+| Race code review | GPT + Gemini | `llm-orchestrate.js` via Bash |
+| Research (best practices, security) | GPT + Gemini | `llm-orchestrate.js` via Bash |
+| User questions | GPT / Gemini | Hook auto-handles |
+
+### GPT Integration
+
+- Auth: `vibe gpt auth` (OAuth) or `vibe gpt key <key>` (API Key)
+- OAuth endpoint: `https://chatgpt.com/backend-api/codex/responses`
+- API Key endpoint: `https://api.openai.com/v1/responses`
+
+### Gemini Integration
+
+- Auth: `vibe gemini auth` (OAuth) or `vibe gemini key <key>` (API Key)
+- Used for: research, review, image analysis, STT
+
 ## Agents
 
-- **Review (12)**: security, performance, architecture, complexity, simplicity, data-integrity, test-coverage, git-history, python, typescript, rails, react reviewers → `.claude/agents/review/`
-- **Research (4)**: best-practices, framework-docs, codebase-patterns, security-advisory → `.claude/agents/research/`
+### Main Agents (18)
+
+- **Explorer** (high/medium/low) — Codebase exploration
+- **Implementer** (high/medium/low) — Code implementation
+- **Architect** (high/medium/low) — Architecture design
+- **Searcher** — Code search
+- **Tester** — Test generation
+- **Simplifier** — Code simplification
+- **Refactor Cleaner** — Refactoring cleanup
+- **Build Error Resolver** — Build error fixing
+- **Compounder** — Multi-step compound tasks
+- **Diagrammer** — Diagram generation
+- **E2E Tester** — E2E test execution
+- **UI Previewer** — UI preview
+
+### Review Agents (12)
+
+security, performance, architecture, complexity, simplicity, data-integrity, test-coverage, git-history, typescript, python, rails, react → `agents/review/`
+
+### Research Agents (4)
+
+best-practices, framework-docs, codebase-patterns, security-advisory → `agents/research/`
+
+### UI/UX Agents (8)
+
+CSV data-driven design intelligence. BM25 search engine auto-generates industry-specific design strategies from 24 CSV files.
+
+| Phase | Agent | Model | Role |
+|-------|-------|-------|------|
+| SPEC | 1. ui-industry-analyzer | Haiku | Industry analysis + design strategy |
+| SPEC | 2. ui-design-system-gen | Sonnet | MASTER.md design system generation |
+| SPEC | 3. ui-layout-architect | Haiku | Page structure/sections/CTA design |
+| RUN | 4. ui-stack-implementer | Haiku | Framework-specific component guide |
+| RUN | 5. ui-dataviz-advisor | Haiku | Chart/visualization library recommendations |
+| REVIEW | 6. ux-compliance-reviewer | Haiku | UX guideline compliance verification |
+| REVIEW | 7. ui-a11y-auditor | Haiku | WCAG 2.1 AA accessibility audit |
+| REVIEW | 8. ui-antipattern-detector | Haiku | UI anti-pattern detection |
+
+**Execution flow**: 1→2,3 (SPEC, supervisor-worker) → 4,5 (RUN, before Phase start) → 6,7,8 (REVIEW, on UI file changes)
+
+**Disable**: Set `"uiUxAnalysis": false` in `.claude/vibe/config.json`
+
+**MCP tools**: `core_ui_search`, `core_ui_stack_search`, `core_ui_generate_design_system`, `core_ui_persist_design_system`
+
+**Data**: `~/.claude/vibe/ui-ux-data/` (11 domain CSVs + 13 stack CSVs)
+
+→ `agents/ui/`
+
+### Planning Agents (2)
+
+requirements-analyst, ux-advisor → `agents/planning/`
+
+### QA Agents (2)
+
+edge-case-finder, acceptance-tester → `agents/qa/`
+
+### Docs Agents (2)
+
+api-documenter, changelog-writer → `agents/docs/`
+
+### Mentor Agent (1)
+
+junior-mentor — Junior developer mentor with EXPLANATION.md generation
+
+### Agent Teams (9 teams)
+
+> Agents form teams to collaborate via shared task lists with mutual feedback.
+> Setting: `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` (`.claude/settings.local.json` env, postinstall auto-configured)
+
+| Team | Workflow | Members | Activation |
+|------|----------|---------|------------|
+| Lite Team | `/vibe.run` normal | architect, implementer, tester | 3+ scenarios, normal mode |
+| Dev Team | `/vibe.run` ULTRAWORK | architect, implementer, tester, security-reviewer | ULTRAWORK or complexity 20+ |
+| Research Team | `/vibe.spec` Step 3 | best-practices, security-advisory, codebase-patterns, framework-docs | Always in /vibe.spec |
+| Review Debate Team | `/vibe.review` Phase 4.5 | security, architecture, performance, simplicity | P1/P2 issues 2+ |
+| SPEC Debate Team | `/vibe.spec.review` Step 3.5 | security, architecture, performance, simplicity | P1/P2 issues 2+ |
+| Debug Team | `/vibe.run` failure | architect, implementer, tester | Build/test failure 3x+ |
+| Security Team | `/vibe.run` security | security, data-integrity, security-advisory, tester | auth/payment file changes |
+| Migration Team | `/vibe.run` migration | architect, implementer, tester, build-error-resolver | Major dependency version change |
+| Fullstack Team | `/vibe.run` fullstack | architect, implementer-backend, implementer-frontend, tester | SPEC has frontend + backend |
+
+## Hooks System
+
+| Event | Hooks |
+|-------|-------|
+| SessionStart | `session-start.js` |
+| PreToolUse (Bash/Edit/Write) | `pre-tool-guard.js` |
+| PostToolUse (Write/Edit) | `post-edit.js`, `code-check.js`, `post-tool-verify.js` |
+| UserPromptSubmit | `prompt-dispatcher.js`, `skill-injector.js`, `keyword-detector.js`, `hud-status.js` |
+| Notification (context 80/90/95%) | `context-save.js` |
+
+**Additional hooks:** `code-review.js`, `llm-orchestrate.js`, `recall.js`, `complexity.js`, `compound.js`, `skill-requirements.js`, `autonomy-controller.js`, `evolution-engine.js`, `hud-multiline.js`, `sentinel-guard.js`, `stop-notify.js`
+
+## Language Support (25 frameworks)
+
+- **TypeScript**: Next.js, React, Angular, NestJS, Vue, Svelte, Nuxt, Tauri, Electron, React Native, Node, Astro
+- **Python**: Django, FastAPI
+- **Java**: Spring
+- **Kotlin**: Android
+- **Ruby**: Rails
+- **Go**, **Rust**, **Swift** (iOS), **C#** (Unity), **Dart** (Flutter), **GDScript** (Godot)
 
 ## Context Management