paper-beamer-deck

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

• Third-party content exposure detected (high risk: 0.75). The skill ingests outsider-authored free text from the source PDF itself (e.g., figure/table captions and labels extracted by Docling or pdffigures2 in scripts/figure_extraction/extract_figures.py via docling_caption_text() / parse_pdffigures2_json()), which then gets placed into the LLM context when the agent uses the extracted figures.json/captions to author slides.