paper-narrative

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests outsider-authored free text from the runtime-provided manuscript abstract/intro and figure captions (e.g., @manuscript.tex and caption text), which are untrusted and are directly embedded into the LLM prompt in derive_paper_brief(abstract_text, figure_claims) and then used again in narrative_review_task via the derived brief fields.