About William2333ZZ/trustshell
Every AI agent that can *act* is a new attack surface, and automated scans only catch the easy layer. The real risks — prompt injection, memory poisoning, sandbox escape, tool/action abuse, poisoned skills — need an adversary.