sandboxer

What This Skill Does

The sandboxer skill enables the OpenClaw AI agent to interface directly with Claude Code sessions hosted within tmux sessions on your local machine. By leveraging an HTTP API provided by the Sandboxer daemon, the agent can monitor, manage, and interact with terminal-based AI workspaces in real-time. This creates a powerful bridge where the agent isn't just generating code in a chat window, but executing it, reviewing output, and managing long-running processes directly in your terminal environment. It abstracts the complexity of manual tmux management, allowing for automated workflows where tasks are dispatched, monitored for completion, and verified through terminal output streams.

Installation

To integrate this skill into your OpenClaw environment, execute the following command: clawhub install openclaw/skills/skills/chriopter/sandboxer. Ensure that the Sandboxer backend is running on your host machine. You can verify this by checking the status of the sandboxer systemd service or by running the health check curl command provided in the technical documentation. The agent relies on local port 8081; ensure no other services are occupying this port. Refer to the GitHub repository at https://github.com/chriopter/sandboxer for advanced configurations and systemd setup scripts.

Use Cases

This skill is ideal for developers managing multiple AI-assisted projects simultaneously. Use it to: 1. Automate the lifecycle of dev environments by programmatically creating and destroying sessions based on project requirements. 2. Chain AI tasks by sending commands to a session and parsing the output to decide subsequent steps. 3. Perform remote debugging by capturing tmux panes to inspect the logs of a long-running process without manually switching windows. 4. Provide a dashboard overview of all active AI sessions to monitor system resource utilization and task progress.

Example Prompts

1. "OpenClaw, list all currently active Claude sessions and tell me which one is running the authentication project."
2. "Please send the command 'npm run test' to the session named 'backend-service' and let me know if it passes."
3. "I need to start a new Claude session in the /home/user/projects/dashboard directory, please set that up using sandboxer."

Tips & Limitations

Always run the health check before attempting to execute commands to avoid failure due to disconnected sockets. Note that this skill requires the tmux utility installed on the host system to function correctly. Avoid sending commands to sessions that are currently in an interactive loop unless the session is specifically designed to handle stdin automation. Because this tool executes shell commands, ensure that your environment security policies allow the agent to invoke tmux and perform local network requests.