Show HN: Scan your AI agents for dangerous capabilities