opensandbox-group/OpenSandbox — reverse-engineered prompt

Build me an open source sandbox platform for AI agents called OpenSandbox. It should let developers safely spin up isolated environments, run commands, execute code, move files in and out, and clean everything up afterward.

I want it to support local Docker use first, with a path for Kubernetes later. Include a simple server, a command line tool, and SDKs or clear client examples so people can create a sandbox from Python and JavaScript. The basic demo should create a Python sandbox, run a shell command, write and read a text file, then run a small Python code interpreter example.

Please include sensible security defaults, sandbox timeouts, API key configuration, and clear docs for getting started. It should also be possible to connect it to tools like Claude Code or Cursor through an MCP server so agents can create sandboxes, run commands, and edit text files safely.

Look up current docs online if you need to, and make the first version clean, tested, and easy to run locally.