@su-record/vibe 2.8.49 → 2.8.51

package/.env.example

@@ -1,37 +1,37 @@
-# ─── VIBE Configuration ─────────────────────────────────────────────
-#
-# All credentials, channels, and model overrides are stored in:
-#   ~/.vibe/config.json
-#
-# Use CLI commands to configure:
-#   vibe setup                    Interactive setup wizard
-#   vibe gpt auth                 GPT OAuth (Plus/Pro)
-#   vibe gpt key <key>            GPT API key
-#   vibe gemini auth              Gemini OAuth
-#   vibe gemini key <key>         Gemini API key
-#   vibe claude key <key>         Claude API key
-#   vibe telegram setup <token>   Telegram bot
-#   vibe slack setup <bot> <app>  Slack bot
-#   vibe status                   Check all status
-#
-# Config structure (~/.vibe/config.json):
-# {
-#   "version": "1",
-#   "credentials": {
-#     "gpt": { "apiKey": "sk-..." },
-#     "gemini": { "apiKey": "AIza..." }
-#   },
-#   "channels": {
-#     "telegram": { "botToken": "123:ABC", "allowedChatIds": ["-100..."] },
-#     "slack": { "botToken": "xoxb-...", "appToken": "xapp-..." }
-#   },
-#   "models": {
-#     "gpt": "gpt-5.4",
-#     "gemini": "gemini-3.1-pro-preview",
-#     "claudeBackground": "claude-sonnet-4-5"
-#   }
-# }
-#
-# Environment variables are still supported as fallback:
-#   OPENAI_API_KEY, GEMINI_API_KEY, GPT_MODEL, etc.
-# ────────────────────────────────────────────────────────────────────
+# ─── VIBE Configuration ─────────────────────────────────────────────
+#
+# All credentials, channels, and model overrides are stored in:
+#   ~/.vibe/config.json
+#
+# Use CLI commands to configure:
+#   vibe setup                    Interactive setup wizard
+#   vibe gpt auth                 GPT OAuth (Plus/Pro)
+#   vibe gpt key <key>            GPT API key
+#   vibe gemini auth              Gemini OAuth
+#   vibe gemini key <key>         Gemini API key
+#   vibe claude key <key>         Claude API key
+#   vibe telegram setup <token>   Telegram bot
+#   vibe slack setup <bot> <app>  Slack bot
+#   vibe status                   Check all status
+#
+# Config structure (~/.vibe/config.json):
+# {
+#   "version": "1",
+#   "credentials": {
+#     "gpt": { "apiKey": "sk-..." },
+#     "gemini": { "apiKey": "AIza..." }
+#   },
+#   "channels": {
+#     "telegram": { "botToken": "123:ABC", "allowedChatIds": ["-100..."] },
+#     "slack": { "botToken": "xoxb-...", "appToken": "xapp-..." }
+#   },
+#   "models": {
+#     "gpt": "gpt-5.4",
+#     "gemini": "gemini-3.1-pro-preview",
+#     "claudeBackground": "claude-sonnet-4-5"
+#   }
+# }
+#
+# Environment variables are still supported as fallback:
+#   OPENAI_API_KEY, GEMINI_API_KEY, GPT_MODEL, etc.
+# ────────────────────────────────────────────────────────────────────

package/CLAUDE.md

@@ -1,169 +1,169 @@
-# VIBE
-
-## Project Nature
-
-> **This is the source code for `@su-record/vibe` npm package.**
-> Modify files in THIS repository only — never the installed copy (`~/.claude/`, `~/.vibe/`).
-
-## Philosophy
-
-> **vibe = Easy vibe coding + Minimum quality guaranteed**
-
-- **Easy Vibe Coding** — Fast flow, think collaboratively with AI
-- **Minimum Quality Guaranteed** — Type safety, code quality, security — automatic baseline
-- **Iterative Reasoning** — Break down problems, ask questions, reason together (don't delegate)
-
-## Constraints (Hard Rules)
-
-### Code Behavior
-
-- **Modify only requested scope** — Don't touch unrelated code
-- **Preserve existing style** — Follow project conventions
-- **Edit existing files, never create new ones** — Fix problems at source
-- **Respect user interrupts** — Ctrl+C/Escape = previous task CANCELLED
-
-### Complexity Limits
-
-| Metric | Limit |
-|--------|-------|
-| Function length | ≤50 lines |
-| Nesting depth | ≤3 levels |
-| Parameters | ≤5 |
-| Cyclomatic complexity | ≤10 |
-
-### TypeScript (Enforced by Quality Gate hooks)
-
-- No `any` — use `unknown` + type guards
-- No `as any` — define proper interfaces
-- No `@ts-ignore` — fix type issues at root
-- Explicit return types on all functions
-
-### Convergence Principle (Over-Diagnosis Prevention)
-
-- **P1 only is mandatory, P2/P3 are best-effort** — P1=0 means done
-- **Repeated reviews narrow scope** — Round 2: P1+P2 only, Round 3+: P1 only
-- **Convergence detection** — Same results as previous round → stop immediately
-- **3 failures = escalation** — No more retries, move to TODO
-- **Changed files only** — Never scan the entire project
-
-### Forbidden Patterns
-
-- No `console.log` in commits
-- No hardcoded strings/numbers (use constants)
-- No commented-out code
-- No incomplete code without TODO
-
-## Architecture (What's NOT Obvious)
-
-### Module System
-
-- ESM only (`"type": "module"`) — imports need `.js` extension
-- TypeScript strict mode → `dist/` via `tsc`
-- Build before test: `npm run build && npx vitest run`
-
-### Config Locations (Don't Confuse)
-
-| Path | Purpose |
-|------|---------|
-| `~/.vibe/config.json` | Global credentials, channels, models (0o600) |
-| `.claude/vibe/config.json` | Project stacks, capabilities, details |
-| `.claude/settings.local.json` | Project hooks (auto-generated, don't commit) |
-
-### Installation Flow
-
-- `postinstall` → global assets (`~/.claude/agents/`, `~/.claude/skills/`, `~/.claude/commands/`)
-- `vibe init` → project setup (detect stacks → config → constitution → rules → hooks → local skills + languages)
-- `vibe update` → re-detect stacks, refresh project config (preserves user capabilities)
-
-### Key Directories (Where to Find Things)
-
-| What | Where |
-|------|-------|
-| CLI commands | `src/cli/commands/` |
-| Postinstall modules | `src/cli/postinstall/` |
-| Stack detection | `src/cli/detect.ts` |
-| Auth (all LLMs) | `src/cli/auth.ts` |
-| Setup helpers | `src/cli/setup.ts`, `src/cli/setup/` |
-| Hooks scripts | `hooks/scripts/` |
-| Agent definitions | `agents/` |
-| Skill definitions | `skills/` |
-| Language rules | `languages/` |
-| SPEC/rule templates | `vibe/rules/`, `vibe/templates/` |
-| Infra (daemon, memory, policy) | `src/infra/` |
-| Telemetry (local JSONL) | `src/infra/lib/telemetry/` |
-| Test helpers (shared) | `src/test-helpers/` |
-| Skill catalog generator | `scripts/gen-skill-docs.ts` |
-
-### Gotchas
-
-- `better-sqlite3` WAL mode — DB module uses synchronous API
-- `crypto.timingSafeEqual` requires same-length buffers — check length first
-- `GLOBAL_SKILLS_CORE`, `GLOBAL_SKILLS_STANDARD`, `GLOBAL_SKILLS_OPTIONAL`, `STACK_TO_SKILLS`, `CAPABILITY_SKILLS` in `src/cli/postinstall/constants.ts` — single source of truth for stack→asset mapping
-- Hook dispatch chain: `prompt-dispatcher.js` → `keyword-detector.js` → `llm-orchestrate.js` (order matters)
-- Pre-existing test failures in `prdParser.test.ts` (7) and `traceabilityMatrix.test.ts` (1) — unrelated to current work
-
-## Workflow
-
-```text
-/vibe.spec → /new → /vibe.spec.review → /vibe.run → /vibe.trace → (auto) code review → Done
-```
-
-| Task Size | Approach |
-|-----------|----------|
-| 1-2 files | Plan Mode |
-| 3+ files | `/vibe.spec` |
-
-## Magic Keywords
-
-| Keyword | Effect |
-|---------|--------|
-| `ultrawork` / `ulw` | Parallel agents + 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 |
-
-## Skill Tier System
-
-Skills are classified into 3 tiers to prevent context overload (Curse of Instructions):
-
-| Tier | Loading | Purpose | Count |
-|------|---------|---------|-------|
-| **core** | Always active | Bug/mistake prevention safety nets | ~9 |
-| **standard** | Project setup selects | Workflow support by stack/capability | ~21 |
-| **optional** | Explicit `/skill` only | Reference/wrapper — not auto-loaded | ~4 |
-
-**Constants**: `GLOBAL_SKILLS_CORE` + `GLOBAL_SKILLS_STANDARD` in `src/cli/postinstall/constants.ts`
-
-## Proactive Skill Suggestions
-
-When the user's context matches a pattern below, suggest the relevant skill **once** per session. If the user says "stop suggesting", disable for the rest of the session.
-
-| User Context | Suggested Skill | Tier | Signal |
-|-------------|-----------------|------|--------|
-| Writing a new SPEC / planning | `/vibe.spec` | core | "let's build", "new feature", "requirements" |
-| SPEC exists, ready to implement | `/vibe.run` | core | SPEC file present, no implementation started |
-| Implementation done, verifying | `/vibe.trace` | core | Code changes match SPEC phases |
-| Technical debt accumulating | `/techdebt` | core | Multiple `any` types, console.log, unused imports |
-| Multi-file refactoring planned | `/exec-plan` | core | 3+ files to change, complex dependencies |
-| Complex feature, unknown tech | `/parallel-research` | standard | "how should we", "which library", architecture questions |
-| Debugging errors repeatedly | `/vibe.trace` | core | 3+ error cycles without resolution |
-| Session ending, work incomplete | `/handoff` | standard | 70%+ context, incomplete tasks |
-| Building new UI/UX | `/design-teach` | standard | "new page", "landing", "dashboard", Figma URL |
-| UI code review | `/design-audit` | standard | After implementing UI components |
-| New project setup / CLAUDE.md | `/claude-md-guide` | standard | "new project", "write CLAUDE.md" |
-| Before shipping UI | `/design-polish` | standard | "ready to deploy", "final check" |
-| External API/SDK code needed | `/chub-usage` | optional | "Stripe 연동", "OpenAI API", SDK/API code request |
-
-## Context Management
-
-- **Exploration/Search**: Haiku | **Implementation**: Sonnet | **Architecture**: Opus
-- At 70%+ context: `save_memory` → `/new` → `/vibe.utils --continue`
-
-## Git Commit Rules
-
-**Must include:** `.claude/vibe/specs/`, `.claude/vibe/features/`, `.claude/vibe/todos/`, `.claude/vibe/config.json`, `CLAUDE.md`
-
-**Exclude:** `~/.claude/vibe/rules/`, `~/.claude/commands/`, `~/.claude/agents/`, `~/.claude/skills/`, `.claude/settings.local.json`
-
-<!-- VIBE:END -->
+# VIBE
+
+## Project Nature
+
+> **This is the source code for `@su-record/vibe` npm package.**
+> Modify files in THIS repository only — never the installed copy (`~/.claude/`, `~/.vibe/`).
+
+## Philosophy
+
+> **vibe = Easy vibe coding + Minimum quality guaranteed**
+
+- **Easy Vibe Coding** — Fast flow, think collaboratively with AI
+- **Minimum Quality Guaranteed** — Type safety, code quality, security — automatic baseline
+- **Iterative Reasoning** — Break down problems, ask questions, reason together (don't delegate)
+
+## Constraints (Hard Rules)
+
+### Code Behavior
+
+- **Modify only requested scope** — Don't touch unrelated code
+- **Preserve existing style** — Follow project conventions
+- **Edit existing files, never create new ones** — Fix problems at source
+- **Respect user interrupts** — Ctrl+C/Escape = previous task CANCELLED
+
+### Complexity Limits
+
+| Metric | Limit |
+|--------|-------|
+| Function length | ≤50 lines |
+| Nesting depth | ≤3 levels |
+| Parameters | ≤5 |
+| Cyclomatic complexity | ≤10 |
+
+### TypeScript (Enforced by Quality Gate hooks)
+
+- No `any` — use `unknown` + type guards
+- No `as any` — define proper interfaces
+- No `@ts-ignore` — fix type issues at root
+- Explicit return types on all functions
+
+### Convergence Principle (Over-Diagnosis Prevention)
+
+- **P1 only is mandatory, P2/P3 are best-effort** — P1=0 means done
+- **Repeated reviews narrow scope** — Round 2: P1+P2 only, Round 3+: P1 only
+- **Convergence detection** — Same results as previous round → stop immediately
+- **3 failures = escalation** — No more retries, move to TODO
+- **Changed files only** — Never scan the entire project
+
+### Forbidden Patterns
+
+- No `console.log` in commits
+- No hardcoded strings/numbers (use constants)
+- No commented-out code
+- No incomplete code without TODO
+
+## Architecture (What's NOT Obvious)
+
+### Module System
+
+- ESM only (`"type": "module"`) — imports need `.js` extension
+- TypeScript strict mode → `dist/` via `tsc`
+- Build before test: `npm run build && npx vitest run`
+
+### Config Locations (Don't Confuse)
+
+| Path | Purpose |
+|------|---------|
+| `~/.vibe/config.json` | Global credentials, channels, models (0o600) |
+| `.claude/vibe/config.json` | Project stacks, capabilities, details |
+| `.claude/settings.local.json` | Project hooks (auto-generated, don't commit) |
+
+### Installation Flow
+
+- `postinstall` → global assets (`~/.claude/agents/`, `~/.claude/skills/`, `~/.claude/commands/`)
+- `vibe init` → project setup (detect stacks → config → constitution → rules → hooks → local skills + languages)
+- `vibe update` → re-detect stacks, refresh project config (preserves user capabilities)
+
+### Key Directories (Where to Find Things)
+
+| What | Where |
+|------|-------|
+| CLI commands | `src/cli/commands/` |
+| Postinstall modules | `src/cli/postinstall/` |
+| Stack detection | `src/cli/detect.ts` |
+| Auth (all LLMs) | `src/cli/auth.ts` |
+| Setup helpers | `src/cli/setup.ts`, `src/cli/setup/` |
+| Hooks scripts | `hooks/scripts/` |
+| Agent definitions | `agents/` |
+| Skill definitions | `skills/` |
+| Language rules | `languages/` |
+| SPEC/rule templates | `vibe/rules/`, `vibe/templates/` |
+| Infra (daemon, memory, policy) | `src/infra/` |
+| Telemetry (local JSONL) | `src/infra/lib/telemetry/` |
+| Test helpers (shared) | `src/test-helpers/` |
+| Skill catalog generator | `scripts/gen-skill-docs.ts` |
+
+### Gotchas
+
+- `better-sqlite3` WAL mode — DB module uses synchronous API
+- `crypto.timingSafeEqual` requires same-length buffers — check length first
+- `GLOBAL_SKILLS_CORE`, `GLOBAL_SKILLS_STANDARD`, `GLOBAL_SKILLS_OPTIONAL`, `STACK_TO_SKILLS`, `CAPABILITY_SKILLS` in `src/cli/postinstall/constants.ts` — single source of truth for stack→asset mapping
+- Hook dispatch chain: `prompt-dispatcher.js` → `keyword-detector.js` → `llm-orchestrate.js` (order matters)
+- Pre-existing test failures in `prdParser.test.ts` (7) and `traceabilityMatrix.test.ts` (1) — unrelated to current work
+
+## Workflow
+
+```text
+/vibe.spec → /new → /vibe.spec.review → /vibe.run → /vibe.trace → (auto) code review → Done
+```
+
+| Task Size | Approach |
+|-----------|----------|
+| 1-2 files | Plan Mode |
+| 3+ files | `/vibe.spec` |
+
+## Magic Keywords
+
+| Keyword | Effect |
+|---------|--------|
+| `ultrawork` / `ulw` | Parallel agents + 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 |
+
+## Skill Tier System
+
+Skills are classified into 3 tiers to prevent context overload (Curse of Instructions):
+
+| Tier | Loading | Purpose | Count |
+|------|---------|---------|-------|
+| **core** | Always active | Bug/mistake prevention safety nets | ~9 |
+| **standard** | Project setup selects | Workflow support by stack/capability | ~21 |
+| **optional** | Explicit `/skill` only | Reference/wrapper — not auto-loaded | ~4 |
+
+**Constants**: `GLOBAL_SKILLS_CORE` + `GLOBAL_SKILLS_STANDARD` in `src/cli/postinstall/constants.ts`
+
+## Proactive Skill Suggestions
+
+When the user's context matches a pattern below, suggest the relevant skill **once** per session. If the user says "stop suggesting", disable for the rest of the session.
+
+| User Context | Suggested Skill | Tier | Signal |
+|-------------|-----------------|------|--------|
+| Writing a new SPEC / planning | `/vibe.spec` | core | "let's build", "new feature", "requirements" |
+| SPEC exists, ready to implement | `/vibe.run` | core | SPEC file present, no implementation started |
+| Implementation done, verifying | `/vibe.trace` | core | Code changes match SPEC phases |
+| Technical debt accumulating | `/techdebt` | core | Multiple `any` types, console.log, unused imports |
+| Multi-file refactoring planned | `/exec-plan` | core | 3+ files to change, complex dependencies |
+| Complex feature, unknown tech | `/parallel-research` | standard | "how should we", "which library", architecture questions |
+| Debugging errors repeatedly | `/vibe.trace` | core | 3+ error cycles without resolution |
+| Session ending, work incomplete | `/handoff` | standard | 70%+ context, incomplete tasks |
+| Building new UI/UX | `/design-teach` | standard | "new page", "landing", "dashboard", Figma URL |
+| UI code review | `/design-audit` | standard | After implementing UI components |
+| New project setup / CLAUDE.md | `/claude-md-guide` | standard | "new project", "write CLAUDE.md" |
+| Before shipping UI | `/design-polish` | standard | "ready to deploy", "final check" |
+| External API/SDK code needed | `/chub-usage` | optional | "Stripe 연동", "OpenAI API", SDK/API code request |
+
+## Context Management
+
+- **Exploration/Search**: Haiku | **Implementation**: Sonnet | **Architecture**: Opus
+- At 70%+ context: `save_memory` → `/new` → `/vibe.utils --continue`
+
+## Git Commit Rules
+
+**Must include:** `.claude/vibe/specs/`, `.claude/vibe/features/`, `.claude/vibe/todos/`, `.claude/vibe/config.json`, `CLAUDE.md`
+
+**Exclude:** `~/.claude/vibe/rules/`, `~/.claude/commands/`, `~/.claude/agents/`, `~/.claude/skills/`, `.claude/settings.local.json`
+
+<!-- VIBE:END -->

package/LICENSE

@@ -1,21 +1,21 @@
-MIT License
-
-Copyright (c) 2025 Su
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+MIT License
+
+Copyright (c) 2025 Su
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.ko.md

@@ -0,0 +1,190 @@
+# VIBE
+
+**AI가 코드를 쓴다. Vibe는 그 코드가 좋은지 확인한다.**
+
+[![npm](https://img.shields.io/npm/v/@su-record/vibe)](https://www.npmjs.com/package/@su-record/vibe)
+[![Node.js](https://img.shields.io/badge/node-%3E%3D18-green)](https://nodejs.org/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-yellow.svg)](LICENSE)
+
+> 한 번 설치하면 에이전트 56개, 스킬 45개, 멀티 LLM 오케스트레이션, 자동 품질 게이트가 기존 AI 코딩 워크플로우에 추가됩니다.
+
+```bash
+npm install -g @su-record/vibe
+vibe init
+```
+
+**Claude Code**, **Codex**, **Cursor**, **Gemini CLI**에서 동작합니다.
+
+---
+
+## 문제
+
+AI 코딩 도구는 동작하는 코드를 생성하지만:
+
+- 타입은 `any`, 리뷰는 생략, 테스트는 잊혀짐
+- 문제를 수동으로 잡아냄 — 이미 피해가 발생한 뒤에
+- 세션 간 컨텍스트가 사라짐
+- 복잡한 작업은 구조 없이 산으로 감
+
+**Vibe는 하네스입니다.** AI 코딩 도구를 감싸고, 코드 생성 전/중/후에 품질을 자동으로 강제합니다.
+
+---
+
+## 작동 원리
+
+```
+Agent = Model + Harness
+```
+
+| | 역할 | 구성 요소 |
+|---|------|----------|
+| **Guides** (피드포워드) | 행동 **전에** 방향 설정 | CLAUDE.md, 에이전트 56개, 스킬 45개, 코딩 규칙 |
+| **Sensors** (피드백) | 행동 **후에** 관찰·교정 | 훅 21개, 품질 게이트, 이볼루션 엔진 |
+
+### 워크플로우
+
+```mermaid
+flowchart LR
+    A["/vibe.spec"] --> B["/vibe.spec.review"]
+    B --> C["/vibe.run"]
+    C --> D["자동 리뷰"]
+    D --> E["완료"]
+```
+
+1. **정의** — `/vibe.spec`이 GPT + Gemini 병렬 리서치로 요구사항 작성
+2. **검증** — `/vibe.spec.review`가 3중 교차 검증 실행 (Claude + GPT + Gemini)
+3. **구현** — `/vibe.run`이 SPEC 기반 구현 + 병렬 코드 리뷰
+4. **리뷰** — 12개 에이전트가 병렬 리뷰, P1/P2 이슈 자동 수정
+
+`ultrawork`를 추가하면 전체 파이프라인이 자동화됩니다:
+
+```bash
+/vibe.run "사용자 인증 추가" ultrawork
+```
+
+---
+
+## 핵심 기능
+
+**품질 게이트** — `any` 타입, `@ts-ignore`, 50줄 초과 함수를 차단합니다. 3계층 방어: 센티넬 가드 → 프리툴 가드 → 코드 체크.
+
+**56개 전문 에이전트** — 탐색, 구현, 아키텍처, 12개 병렬 리뷰어, 8개 UI/UX 에이전트, QA 코디네이터. 각 에이전트는 범용 래퍼가 아닌 목적 특화 설계.
+
+**멀티 LLM 오케스트레이션** — Claude로 오케스트레이션, GPT로 추론, Gemini로 리서치. 가용 모델에 따라 자동 라우팅. 기본값은 Claude 단독 운영.
+
+**Session RAG** — SQLite + FTS5 하이브리드 검색으로 결정, 제약, 목표를 세션 간 지속. 세션 시작 시 컨텍스트 자동 복원.
+
+**24개 프레임워크 감지** — 스택 자동 감지 (Next.js, React, Django, Spring Boot, Rails, Go, Rust 외 17종) 후 프레임워크별 규칙 적용. 모노레포 인식.
+
+**자기 개선** — 이볼루션 엔진이 훅 실행 패턴을 분석하고, 스킬 갭을 감지하고, 새 규칙을 생성. 회귀 발생 시 서킷 브레이커가 롤백.
+
+---
+
+## 포함 내용
+
+| 분류 | 수량 | 예시 |
+|------|------|------|
+| **에이전트** | 56 | Explorer, Implementer, Architect (각 3단계), 리뷰 전문가 12개, UI/UX 8개, QA 코디네이터 |
+| **스킬** | 45 | 3티어 시스템 (core/standard/optional) — 컨텍스트 과부하 방지 |
+| **훅** | 21 | 세션 복원, 파괴적 명령 차단, 자동 포맷, 품질 체크, 컨텍스트 자동 저장 |
+| **프레임워크** | 24 | TypeScript (12), Python (2), Java/Kotlin (2), Rails, Go, Rust, Swift, Flutter, Unity, Godot |
+| **슬래시 커맨드** | 11 | `/vibe.spec`, `/vibe.run`, `/vibe.review`, `/vibe.trace`, `/vibe.figma` 등 |
+
+---
+
+## 멀티 CLI 지원
+
+| CLI | 에이전트 | 스킬 | 지침 파일 |
+|-----|---------|------|----------|
+| Claude Code | `~/.claude/agents/` | `~/.claude/skills/` | `CLAUDE.md` |
+| Codex | `~/.codex/plugins/vibe/` | 플러그인 내장 | `AGENTS.md` |
+| Cursor | `~/.cursor/agents/` | `~/.cursor/skills/` | `.cursorrules` |
+| Gemini CLI | `~/.gemini/agents/` | `~/.gemini/skills/` | `GEMINI.md` |
+
+---
+
+## 멀티 LLM 라우팅
+
+| 프로바이더 | 역할 | 필수 여부 |
+|-----------|------|----------|
+| **Claude** | 오케스트레이션, SPEC, 리뷰 | 필수 (Claude Code) |
+| **GPT** | 추론, 아키텍처, 엣지 케이스 | 선택 (Codex CLI 또는 API Key) |
+| **Gemini** | 리서치, 교차 검증, UI/UX | 선택 (gemini-cli 또는 API Key) |
+
+가용 모델에 따라 자동 전환. 기본값은 Claude 단독 — GPT와 Gemini는 강화 옵션.
+
+Codex CLI나 gemini-cli가 설치되어 있으면 자동 감지됩니다. 직접 API Key를 설정하려면:
+
+```bash
+vibe gpt key <your-api-key>
+vibe gemini key <your-api-key>
+```
+
+---
+
+## Figma → 코드
+
+트리 기반 구조적 매핑으로 디자인을 코드로 변환 (스크린샷 추정이 아님).
+
+```bash
+/vibe.figma "https://figma.com/design/ABC/Project?node-id=1-2"
+
+# 반응형 (모바일 + 데스크탑)
+/vibe.figma "mobile-url" "desktop-url"
+```
+
+Figma REST API로 추출 → 스택 인식 코드 생성 (React/Vue/Svelte/SCSS/Tailwind) → 기존 디자인 시스템에 매핑.
+
+---
+
+## CLI 레퍼런스
+
+```bash
+vibe init                      # 프로젝트 초기화 (스택 감지, 하네스 설치)
+vibe update                    # 스택 재감지, 설정 새로고침
+vibe upgrade                   # 최신 버전 업그레이드
+vibe status                    # 현재 상태 확인
+vibe config show               # 통합 설정 보기
+vibe stats [--week|--quality]  # 사용량 텔레메트리
+
+vibe gpt key|status            # GPT API Key 설정
+vibe gemini key|status         # Gemini API Key 설정
+vibe figma breakpoints         # 반응형 브레이크포인트
+vibe skills add <owner/repo>   # skills.sh에서 스킬 설치
+```
+
+---
+
+## 설정
+
+| 파일 | 용도 |
+|------|------|
+| `~/.vibe/config.json` | 글로벌 — 인증, 채널, 모델 (0o600) |
+| `.claude/vibe/config.json` | 프로젝트 — 스택, 기능, 품질 설정 |
+
+---
+
+## 매직 키워드
+
+| 키워드 | 효과 |
+|--------|------|
+| `ultrawork` | 전체 자동화 — 병렬 에이전트 + 자동 계속 + 품질 루프 |
+| `ralph` | 100% 완료까지 반복 (스코프 축소 없음) |
+| `quick` | 빠른 모드, 최소 검증 |
+
+---
+
+## 요구사항
+
+- Node.js >= 18.0.0
+- Claude Code (필수)
+- GPT, Gemini (선택 — 멀티 LLM 기능용)
+
+## 문서
+
+- [README (English)](README.md)
+- [릴리스 노트](RELEASE_NOTES.md)
+
+## 라이선스
+
+MIT — Copyright (c) 2025 Su