rag-eval

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

• Third-party content exposure detected (medium risk: 0.65). The required runtime path ingests dataset-provided corpus/ files and train.json (including contexts[].text/question/answer) into the evaluation and RAGAS judge context; if those dataset files/JSON were authored by an outsider, their free text becomes LLM-readable input via the evaluator’s dataset loading and per-query prompt construction.