literature-review

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow includes searching public academic databases (e.g., Google Scholar/Semantic Scholar/arXiv/PubMed) and then extracting “key data” from included papers’ full text, which would ingest outsider-authored free-form prose from those papers into the agent’s LLM context.