scientific-data-extraction

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime LLM ingestion path is the extracted scientific document text (e.g., pymupdf4llm.to_markdown(pdf_path) / scipdf_parser.parse_pdf_to_dict(...) / OCR text), which is outsider-authored when the operating user supplies a PDF/HTML/image from outside their organization; that readable text is then embedded into LLM prompts for “LLM-enhanced extraction” and “LLM verification” (e.g., Text: {text} in the prompt).