Files
Claude-Code-Workflow/README.md
catlog22 10b3ace917 docs: 添加自定义 CLI 注册说明
- 通过 API Settings 注册任意 API 为自定义 CLI
- 注册后可语义调用自定义 CLI
- 支持自定义 CLI 与内置 CLI 协同编排
2026-01-17 11:02:20 +08:00

10 KiB

Claude Code Workflow (CCW)


CCW is a JSON-driven multi-agent development framework with intelligent CLI orchestration. It provides 4-level workflow system from rapid execution to full brainstorming, transforming AI development into powerful orchestration.

Key Features

Feature Description
🎯 4-Level Workflows From lite-lite-lite (instant) to brainstorm (multi-role analysis)
🔄 Multi-CLI Orchestration Gemini, Qwen, Codex, Claude - auto-select or manual
Dependency-Aware Parallelism Agent parallel execution without worktree complexity
🔧 Issue Workflow Post-development maintenance with optional worktree isolation
📦 JSON-First State .task/IMPL-*.json as single source of truth
🖥️ Dashboard Visual session management, CodexLens search, graph explorer

📖 New? See Workflow Guide for the complete 4-level workflow system.


🚀 Quick Start

Install CCW

npm install -g claude-code-workflow
ccw install -m Global

Choose Your Workflow Level

Level Command Use Case
1 /workflow:lite-lite-lite Quick fixes, config changes
📝 2 /workflow:lite-plan Clear single-module features
🔧 2 /workflow:lite-fix Bug diagnosis and fix
🔍 2 /workflow:multi-cli-plan Multi-perspective analysis
📊 3 /workflow:plan Multi-module development
🧪 3 /workflow:tdd-plan Test-driven development
🧠 4 /workflow:brainstorm:auto-parallel New features, architecture design

Workflow Examples

# Level 1: Instant execution
/workflow:lite-lite-lite "Fix typo in README"

# Level 2: Lightweight planning
/workflow:lite-plan "Add JWT authentication"
/workflow:lite-fix "User upload fails with 413 error"

# Level 3: Standard planning with session
/workflow:plan "Implement payment gateway integration"
/workflow:execute

# Level 4: Multi-role brainstorming
/workflow:brainstorm:auto-parallel "Design real-time collaboration system" --count 5
/workflow:plan --session WFS-xxx
/workflow:execute

🛠️ CLI Tool Installation

CCW supports multiple CLI tools for code analysis and generation. Install as needed:

🔷 Gemini CLI

Google's official Gemini CLI:

# Install
npm install -g @anthropic-ai/gemini-cli

# Configure API Key
export GEMINI_API_KEY="your-api-key"

# Verify
gemini --version

🟢 Codex CLI

OpenAI Codex CLI (recommended for long autonomous coding):

# Install
npm install -g @openai/codex

# Configure API Key
export OPENAI_API_KEY="your-api-key"

# Verify
codex --version

🟠 OpenCode CLI

Open-source multi-model CLI:

# Install
npm install -g opencode-ai

# Configure (supports multiple models)
export OPENCODE_API_KEY="your-api-key"

# Verify
opencode --version

🟣 Qwen CLI

Alibaba Cloud Qwen CLI:

# Install
pip install qwen-cli

# Configure
export QWEN_API_KEY="your-api-key"

# Verify
qwen --version

🎭 Semantic CLI Invocation

Users can semantically specify CLI tools in prompts - the system automatically invokes the corresponding CLI for analysis.

Basic Invocation

User Prompt System Action
"Use Gemini to analyze the auth module" Auto-invoke gemini CLI for analysis
"Let Codex review this code" Auto-invoke codex CLI for review
"Ask Qwen about performance optimization" Auto-invoke qwen CLI for consultation

Multi-CLI Orchestration

Users can semantically orchestrate multiple CLIs in a single prompt:

Pattern User Prompt Example
🔄 Collaborative "Use Gemini and Codex to collaboratively analyze security vulnerabilities"
Parallel "Have Gemini, Codex, and Qwen analyze the architecture in parallel"
🔁 Iterative "Use Gemini to diagnose, then Codex to fix, iterate until resolved"
🔗 Pipeline "Gemini designs the solution, Codex implements, Claude reviews"

Examples

# Single CLI invocation
User: "Use Gemini to analyze the database query performance"
→ System auto-calls: gemini CLI with analysis task

# Collaborative analysis
User: "Use Gemini and Codex to collaboratively review the authentication flow"
→ System auto-calls: gemini + codex CLIs, synthesizes results

# Parallel multi-perspective
User: "Have all available CLIs analyze this architecture design in parallel"
→ System auto-calls: gemini, codex, qwen in parallel → merged report

# Sequential pipeline
User: "Use Gemini to plan the refactoring, then Codex to implement it"
→ System auto-calls: gemini (plan) → codex (implement) sequentially

💡 Key: Just describe which CLI to use and what to do - CCW handles the invocation automatically.

Custom CLI Registration

Register any API as a custom CLI via Dashboard API Settings:

// Example: Register "deepseek" as custom CLI
{
  "tools": {
    "deepseek": {
      "enabled": true,
      "type": "api-endpoint",
      "endpoint": "https://api.deepseek.com/v1/chat",
      "apiKey": "your-api-key"
    }
  }
}

After registration, use it semantically:

User: "Use DeepSeek to analyze this algorithm complexity"
→ System auto-calls: deepseek CLI (your custom endpoint)

User: "Let DeepSeek and Gemini compare their analysis results"
→ System auto-calls: deepseek + gemini in parallel

⚙️ Dashboard: ccw view → Status → API Settings to manage custom CLIs.


🔍 ACE Tool Configuration

ACE (Augment Context Engine) provides powerful semantic code search.

Method Link
📘 Official Augment MCP Documentation
🔧 Proxy Version ace-tool (GitHub)

⚠️ In Development: CodexLens is under iterative optimization. Some features may be unstable.

CodexLens provides local code indexing and search without external APIs:

Search Mode Description
🔤 FTS Full-text search, based on SQLite FTS5
🧠 Semantic Semantic search, using local embedding models
🔀 Hybrid Hybrid search, combining FTS + Semantic + Reranking

Installation

# Enter codex-lens directory
cd codex-lens

# Install dependencies
pip install -e .

# Initialize index
codexlens index /path/to/project

Dashboard Integration

Open Dashboard via ccw view, manage indexes and execute searches in CodexLens Manager.


💻 CCW CLI Commands

ccw install           # Install workflow files
ccw view              # Open dashboard
ccw cli -p "..."      # Execute CLI tools (Gemini/Qwen/Codex)
ccw upgrade -a        # Upgrade all installations

Dashboard Features

Feature Description
📊 Session Overview Track workflow sessions and progress
🔍 CodexLens FTS + Semantic + Hybrid code search
🕸️ Graph Explorer Interactive code relationship visualization
📜 CLI Manager Execution history with session resume

📖 Documentation

Document Description
📘 Workflow Guide 4-level workflow system (recommended)
🚀 Getting Started 5-minute quick start
🖥️ Dashboard Guide Dashboard user guide
FAQ Common questions
📝 Changelog Version history

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     Main Workflow (4 Levels)                    │
│  ⚡ Level 1: lite-lite-lite (instant, no artifacts)             │
│  📝 Level 2: lite-plan / lite-fix / multi-cli-plan (→ execute)  │
│  📊 Level 3: plan / tdd-plan / test-fix-gen (session persist)   │
│  🧠 Level 4: brainstorm:auto-parallel → plan → execute          │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                   Issue Workflow (Supplement)                   │
│  🔍 discover → 📋 plan → 📦 queue → ▶️ execute (worktree)        │
└─────────────────────────────────────────────────────────────────┘

Core Principles:

  • Dependency Analysis solves parallelism - no worktree needed for main workflow
  • 🔧 Issue Workflow supplements main workflow for post-development maintenance
  • 🎯 Select workflow level based on complexity - avoid over-engineering

🤝 Contributing

📄 License

MIT License - see LICENSE