Agentic Research
中文
English Edition/Tools

OpenClaw Installation and Claude Code Bridge Skill Configuration

2026/05/0917 min readBryan Chan閱讀中文原文
TopicsOpenClawClaude CodeAgentTools

What is OpenClaw?

OpenClaw is an AI Agent orchestration framework and the second layer in a three-layer architecture. It is responsible for:

  • Task understanding and decomposition
  • Skills management (reusable capability modules)
  • Calling other tools through the Bridge mechanism (Claude Code, terminal, etc.)

Step 1: Install OpenClaw

# Install via npm
npm install -g openclaw

# Or Install from Source
git clone https://github.com/openclaw/openclaw.git
cd openclaw
pnpm install
pnpm build

Verification:

openclaw --version

Step 2: Initialize Configuration

openclaw init

This creates the ~/.openclaw/ directory:

~/.openclaw/
├── openclaw.json        # Main configuration
├── agents/              # Sub-agent definitions
├── skills/              # Skills directory
├── workspace/           # Workspace
├── plugins/             # Plugins
├── memory/              # Long-term memory
└── logs/                # Logs

Step 3: Configure the LLM Backend

Edit ~/.openclaw/openclaw.json:

{
  "models": {
    "default": "deepseek",
    "providers": {
      "deepseek": {
        "api_key": "sk-xxxxxxxxxxxxxxxxxxxxxxxx",
        "base_url": "https://api.deepseek.com/v1",
        "models": ["deepseek-chat", "deepseek-reasoner"]
      },
      "modelstudio": {
        "api_key": "sk-xxxxxxxxxxxxxxxxxxxxxxxx",
        "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
        "models": ["qwen-max", "qwen-plus", "qwen-coder-plus"]
      }
    }
  }
}

Step 4: Install the Claude Code Bridge Skill

This is the critical bridge connecting OpenClaw and Claude Code.

4.1 Install the Skill

# Install from the Skills Marketplace
openclaw skill install claude-code-bridge

# Or Manual Installation
cd ~/.openclaw/skills
git clone https://github.com/openclaw/skills-claude-code-bridge.git claude-code-bridge

4.2 Configure the Bridge

openclaw skill config claude-code-bridge

Edit the configuration file that opens:

{
  "claude_code": {
    "binary_path": "/opt/homebrew/bin/claude",
    "default_model": "claude-sonnet-4-20250514",
    "timeout_seconds": 300,
    "max_concurrent_sessions": 3,
    "workspace": "~/.openclaw/workspace/claude-code",
    "auto_compact": true,
    "compact_threshold": 80000
  },
  "fallback": {
    "enabled": true,
    "provider": "deepseek",
    "model": "deepseek-chat"
  }
}

4.3 Verify the Bridge

openclaw skill test claude-code-bridge

Expected output:

✓ Claude Code binary found at /opt/homebrew/bin/claude
✓ Claude Code version: 1.x.x
✓ API connection verified
✓ Bridge ready

Step 5: Configure the Hermes ↔ OpenClaw Connection

Enable the Hermes Agent to invoke OpenClaw:

Configure on the Hermes Side

~/.hermes/config.yaml:

integrations:
  openclaw:
    enabled: true
endpoint: "http://localhost:3001"  # OpenClaw API port
    api_key: "oc_xxxxxxxxxxxx"

Configure on the OpenClaw Side

~/.openclaw/openclaw.json:

{
  "server": {
    "port": 3001,
    "host": "127.0.0.1",
    "api_keys": ["oc_xxxxxxxxxxxx"]
  }
}

Step 6: Start OpenClaw

# Development Mode
openclaw serve --dev

# Production Mode
openclaw serve

# Running in the Background
openclaw serve --daemon

Step 7: End-to-End Testing

Send the following to Hermes via Feishu:

Invoke Claude Code via OpenClaw to create a README_CN.md file in the OpenClaw project directory, containing a Chinese introduction to the project.

Complete flow:

  1. Feishu → Hermes receives
  2. Hermes → OpenClaw API passes the task
  3. OpenClaw → Claude Code Bridge → Claude Code executes
  4. Claude Code creates the file → returns the result
  5. OpenClaw → Hermes sends back
  6. Hermes → Feishu replies to the user

Common Management Commands

# View Running Status
openclaw status

# View Installed Skills
openclaw skill list

# Check Claude Code Bridge Status
openclaw skill status claude-code-bridge

# View Logs
openclaw logs --tail 50

# Restart Service
openclaw restart

Usage Scenarios

Scenario 1: Agent-Based Automated Coding

Send commands to Hermes via Feishu. OpenClaw invokes Claude Code to automatically create files, modify code, and commit to Git, enabling a coding workflow without manual intervention.

Scenario 2: Multi-Agent Coordination

OpenClaw acts as an agent orchestration framework, coordinating multiple Claude Code instances to process different tasks in parallel (such as frontend development + backend development + documentation generation).

Scenario 3: Automated Code Review

Claude Code Bridge automatically performs code review tasks, checking code style, potential bugs, and security vulnerabilities, generating a review report and returning it to Hermes.

Reference Resources

More in Tools