reverse-trace

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). High: the required runtime workflow runs rt_trace.py, which invokes rt_vision.py (Google Vision Web Detection) and rt_gemini.py (Gemini multimodal LLM) and then synthesizes/prints the returned web page titles/URLs and Gemini “reasoning” into the final report; those fields originate from external services and can include attacker-controlled free text (e.g., page titles/snippets) that is then placed into the agent’s LLM context via the orchestrator’s synthesis/reporting path.