openclaw-claude-code-plugin

OpenClaw plugin to orchestrate Claude Code

Orchestrate Claude Code sessions as managed background processes from any OpenClaw channel.

Launch, monitor, and interact with multiple Claude Code SDK sessions directly from Telegram, Discord, or any OpenClaw-supported platform — without leaving your chat interface.

Two parallel Claude Code agents building an X clone and an Instagram clone simultaneously from Telegram.

Quick Start

1. Install the plugin

openclaw plugins install @betrue/openclaw-claude-code-plugin
openclaw gateway restart

2. Configure notifications (minimal)

Add to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "openclaw-claude-code-plugin": {
        "enabled": true,
        "config": {
          "fallbackChannel": "telegram|my-bot|123456789",
          "maxSessions": 5
        }
      }
    }
  }
}

3. Launch your first session

Ask your agent: "Fix the bug in auth.ts"

On first launch, the plugin runs 4 safety checks and guides you through one-time setup:

1. Answer an autonomy question — tell the agent how much freedom Claude Code gets
2. Run a heartbeat config command — paste the jq one-liner the agent provides
3. Restart the gateway — openclaw gateway restart

That's it. Future launches skip setup entirely.

Full first-launch walkthrough: docs/safety.md

Features

• Multi-session management — Run multiple concurrent sessions, each with a unique ID and human-readable name
• Foreground / background model — Sessions run in background by default; bring any to foreground to stream output in real time, with catchup of missed output
• Real-time notifications — Get notified on completion, failure, or when Claude asks a question
• Multi-turn conversations — Send follow-up messages, interrupt, or iterate with a running agent
• Session resume & fork — Resume any completed session or fork it into a new conversation branch
• 4 pre-launch safety checks — Autonomy skill, heartbeat config, HEARTBEAT.md, and channel mapping
• Multi-agent support — Route notifications to the correct agent/chat via workspace-based channel mapping
• Automatic cleanup — Completed sessions garbage-collected after 1 hour; IDs persist for resume

Tools

All tools are also available as chat commands (/claude, /claude_fg, etc.) and most as gateway RPC methods.

Full parameter tables and response schemas: docs/API.md

Quick Usage

# Launch a session
/claude Fix the authentication bug in src/auth.ts
/claude --name fix-auth Fix the authentication bug

# Monitor
/claude_sessions
/claude_fg fix-auth
/claude_bg fix-auth

# Interact
/claude_respond fix-auth Also add unit tests
/claude_respond --interrupt fix-auth Stop that and do this instead

# Lifecycle
/claude_kill fix-auth
/claude_resume fix-auth Add error handling
/claude_resume --fork fix-auth Try a different approach
/claude_stats

Notifications

The plugin sends real-time notifications to your chat based on session lifecycle events:

Foreground sessions stream full output in real time. Background sessions only send lifecycle notifications.

Notification architecture and delivery model: docs/NOTIFICATIONS.md

Configuration

Set values in ~/.openclaw/openclaw.json under plugins.entries["openclaw-claude-code-plugin"].config.

Essential parameters

Example

{
  "plugins": {
    "entries": {
      "openclaw-claude-code-plugin": {
        "enabled": true,
        "config": {
          "maxSessions": 3,
          "defaultBudgetUsd": 10,
          "defaultModel": "sonnet",
          "permissionMode": "bypassPermissions",
          "fallbackChannel": "telegram|main-bot|123456789",
          "agentChannels": {
            "/home/user/agent-seo": "telegram|seo-bot|123456789",
            "/home/user/agent-main": "telegram|main-bot|123456789"
          }
        }
      }
    }
  }
}

Skill Example

The plugin is a transparent transport layer — business logic lives in OpenClaw skills:

---
name: Coding Agent Orchestrator
description: Orchestrates Claude Code sessions with auto-response rules.
metadata: {"openclaw": {"requires": {"plugins": ["openclaw-claude-code-plugin"]}}}
---

# Coding Agent Orchestrator

## Auto-response rules

When a Claude Code session asks a question, analyze and decide:

### Auto-respond (use `claude_respond` immediately):
- Permission requests for file reads, writes, or bash commands -> "Yes, proceed."
- Confirmation prompts like "Should I continue?" -> "Yes, continue."

### Forward to user:
- Architecture decisions (Redis vs PostgreSQL, REST vs GraphQL...)
- Destructive operations (deleting files, dropping tables...)
- Anything involving credentials, secrets, or production environments

## Workflow
1. User sends a coding task -> `claude_launch(prompt, ...)`
2. Session runs in background. Monitor via wake events.
3. On wake event -> `claude_output` to read the question, then auto-respond or forward.
4. On completion -> summarize the result and notify the user.

A comprehensive orchestration skill is available at skills/claude-code-orchestration/SKILL.md.

Documentation

License

MIT — see package.json for details.