Has anyone used this? Do strict LLM permissions not suffice? Seemed like a nice idea. But it's just an 8000-line shell script written by some guy. So... 🤔 https://agent-safehouse.dev