A pattern for safely executing dangerous operations like firewall rules on remote servers - verify everything before the point of no return.

How to use OS-level firewall rules scoped to specific users to restrict what your AI agent can access