Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
Add power-user disclaimer
What This Skill Does
Sandboxer is a local HTTP API (localhost:8081) that spawns and manages tmux sessions for AI coding agents and terminal tools. You can create sessions for Claude Code, Gemini, OpenCode, bash, or lazygit in any workspace repo, send them tasks, stream their output, and kill them when done. It also manages a structured workspace directory with per-agent config, memory, and git-cloned project repos.
Controlling AI coding sessions over a simple HTTP API lets other agents or scripts orchestrate multi-agent workflows without any manual terminal interaction.
When to use it
- Spawning a Claude Code session to fix failing tests in a specific repo
- Running parallel AI agent sessions across multiple projects without switching terminals
- Sending a follow-up prompt to an in-progress agent session
- Checking the last 20 lines of terminal output to see if a task is still running
- Auto-committing workspace changes after an agent finishes a batch of work
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Have Claude fix the failing tests in the payments repo
- 1Call GET /api/workspaces to find the payments repo path
- 2Call GET /api/create?type=claude&dir=/root/workspaces/AGENT/data/repos/payments to spawn a Claude Code session
- 3Call GET /api/send with the session name and task text to dispatch the instruction
- 4Poll GET /api/session-monitor every 30 seconds to track status and recent output
- 5Call GET /api/kill once status shows done
Claude Code completes the task inside the tmux session; full output retrievable via /api/capture