ClawShell enforces OS-level isolation for AI agents — so even if the agent is fully compromised, it can't access what it was never given.
