Automated security audits of AI agent configurations. Identifies missing authentication, prompt injection, and unsafe permissions.