Use when testing an authorized LLM application for prompt injection, system-prompt exposure, unsafe tool use, or RAG data-boundary failures.
