Claw Sheath

Run AI agents in a secure shell

Claw Sheath is a lightweight protection layer for AI coding agents. It wraps the shell and intercepts risky commands like rm, dd, mkfs, kill, and kubectl delete (can be anything), asking the agent to justify the action before it runs. It works with OpenClaw, Claude Code, Cursor, and Antigravity, and also offers a strict mode with an LLM judge for extra review. It is not a sandbox - it is a simple first layer of defense against accidental damage from hallucinating agents.