garry00107/SandboxOS ? reverse-engineered prompt

Build me a sandboxed terminal OS in Python that can safely run AI agents without letting them touch the host machine.

I want a simple command line experience with a fake file system, common shell commands, process handling, and a clear boot flow. It should block path escapes, dangerous imports, shell escapes, and network access by default, while also enforcing real resource limits like CPU, memory, file size, open files, and forks. Every action and security violation should be logged so I can inspect what happened later.

Please include a stricter locked down mode for Linux or macOS if possible, plus commands to check security status, quotas, and audit logs. Also add a built in Gemini powered coding agent that can read, write, create, list, and run Python files inside the sandbox, and can iterate on tasks like making a small CLI app and tests. Keep it Python stdlib only unless the Gemini client is absolutely needed, and look up current docs online if you need to.