llm-relevance-scoring

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

• Third-party content exposure detected (high risk: 0.70). The required runtime path is batch_score() → score_with_ollama(item["content"], ...) which injects the corpus item’s free-form text into the LLM prompt; if that corpus includes outsider-authored messages (e.g., comments/posts/issues) then their text becomes LLM-readable context, creating indirect prompt-injection risk.