literature-research

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

• Third-party content exposure detected (high risk: 0.80). This skill directly fetches and ingests open public content (PubMed/PMC full texts, ClinicalTrials.gov entries, and Unpaywall results) as part of its required workflow (see scripts/literature_deep_review.py and pubmed_literature_search.py where it downloads PMC full texts and queries ClinicalTrials.gov/Unpaywall), and that untrusted third-party content is parsed and used to drive classification, filtering, and downstream decisions—so external content could indirectly influence agent behavior.