AI agent security policy-as-code for detecting capability, permission, and least-privile…