Helps founders govern AI agents before they act in production: privilege matrix, sandbox trust boundary, and runtime tool-call decision log