A trust layer for AI agents is overdue, and structured reviews from verified users is the right call. How do you stop agent builders from gaming their own ratings?