Audit an AI agent configuration for security holes: prompt injection exposure, missing authentication, over-broad tool permissions and unsafe gateway settings. Check an agent before you trust it or connect to it