Vibe

# Vibe

**SPEC-Scenario-driven AI coding framework for Claude Code with Multi-LLM orchestration**  

[![npm version](https://img.shields.io/npm/v/@su-record/vibe.svg)](https://www.npmjs.com/package/@su-record/vibe)
[![npm downloads](https://img.shields.io/npm/dt/@su-record/vibe)](https://www.npmjs.com/package/@su-record/vibe)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

## Install

```bash
npm install -g @su-record/vibe
vibe init
```

## Commands

### Terminal

| Command | Description |
|---------|-------------|
| `vibe init` | Initialize project |
| `vibe update` | Update settings |
| `vibe status` | Check status |
| `vibe gpt auth` | GPT OAuth authentication |
| `vibe gpt key <KEY>` | GPT API key setup |
| `vibe gpt logout` | GPT logout |
| `vibe gemini auth` | Gemini OAuth authentication |
| `vibe gemini key <KEY>` | Gemini API key setup |
| `vibe gemini logout` | Gemini logout |
| `vibe help` | Help |
| `vibe version` | Version info |

### Claude Code

| Command | Description |
|---------|-------------|
| `/vibe.spec "feature"` | Create SPEC + parallel research |
| `/vibe.run "feature"` | Execute implementation |
| `/vibe.run "feature" ultrawork` | Maximum performance mode |
| `/vibe.verify "feature"` | BDD verification |
| `/vibe.review` | 13+ agent parallel review |
| `/vibe.analyze` | Code analysis |
| `/vibe.reason "problem"` | Systematic reasoning |
| `/vibe.utils` | Utilities (--e2e, --diagram, etc.) |

## Workflow

```
/vibe.spec → /vibe.run → /vibe.verify → /vibe.review
```

## Key Features

| Feature | Description |
|---------|-------------|
| **Multi-LLM Research (v2.5)** | Claude + GPT + Gemini 3-way validation during SPEC research |
| **Multi-model orchestration** | Claude + GPT-5.2 + Gemini 3 Pro |
| **13+ parallel review agents** | Security, performance, architecture |
| **Iterative SPEC review** | GPT + Gemini convergence loop (max 3 rounds) |
| **Large SPEC auto-split** | 5+ phases auto-split into phase files |
| **BDD auto verification** | Given/When/Then scenario verification |
| **ULTRAWORK mode** | One keyword enables all optimizations |
| **25 built-in tools** | Code analysis, memory management, quality validation |
| **Auto context management** | 80%+ auto-save, session restore |
| **23 language presets** | TypeScript, Python, Go, Rust, Swift, Kotlin, C#, Ruby, Dart, GDScript |

## ULTRAWORK Mode

Enable maximum performance with `ultrawork` or `ulw`:

```bash
/vibe.run "feature" ultrawork
/vibe.run "feature" ulw        # Same
```

**Enabled features:**
- Parallel subagent exploration (3+ concurrent)
- Boulder Loop (auto-progress until all scenarios complete)
- Auto-retry on errors (max 3)
- Continuous execution without phase confirmation
- Auto-save at 80%+ context

## Multi-model Orchestration

**Hook-based automatic routing** - Keywords in your prompt trigger the right LLM automatically:

| Keyword Pattern | Routes to | Use Case |
|-----------------|-----------|----------|
| `architecture`, `design` | GPT | Architecture review |
| `UI`, `UX` | Gemini | UI/UX feedback |
| `debugging`, `find bug` | GPT | Bug analysis |
| `analyze code` | Gemini | Code review |

**Smart Routing features:**
- Automatic LLM selection based on task type
- Exponential backoff retry (3 attempts, 2s → 4s → 8s)
- Auto fallback chain: GPT ↔ Gemini (if primary fails)
- 5-minute availability cache
- Web search via Gemini (Google Search grounding)

**API usage:**

```javascript
import('@su-record/vibe/lib/gpt').then(g => g.ask('question'))
import('@su-record/vibe/lib/gemini').then(g => g.webSearch('search query'))
```

## Multi-LLM Research (v2.5)

During `/vibe.spec`, research agents use **3 LLM perspectives** for quality assurance:

```text
┌─────────────────────────────────────────────────────────────┐
│  MULTI-LLM PARALLEL RESEARCH                                │
├─────────────────────────────────────────────────────────────┤
│  best-practices-agent                                       │
│    ├── Claude: Core patterns, anti-patterns                 │
│    ├── GPT: Architecture patterns, code conventions         │
│    └── Gemini: Latest trends, framework updates             │
├─────────────────────────────────────────────────────────────┤
│  security-advisory-agent                                    │
│    ├── Claude: OWASP Top 10, security patterns              │
│    ├── GPT: CVE database, vulnerability details             │
│    └── Gemini: Latest security advisories, patches          │
└─────────────────────────────────────────────────────────────┘
```

> vibe = Quality Assurance Framework

## Parallel Review Agents

`/vibe.review` runs 13+ agents concurrently:

| Category | Agents |
|----------|--------|
| Security | security-reviewer, data-integrity-reviewer |
| Performance | performance-reviewer, complexity-reviewer |
| Architecture | architecture-reviewer, simplicity-reviewer |
| Language | python, typescript, rails, react reviewers 

... (truncated)