extract-from-pdfs

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

• Third-party content exposure detected (medium risk: 0.30). Step 3 “Extract Data from PDFs” reads user-supplied PDF binaries from --pdf-dir/--source-derived pdf_path and ingests their extracted content via Claude’s PDF vision (base64 upload) in scripts/03_extract_from_pdfs.py, so outsider-authored text can be supplied by posting/adding poison PDFs into the pipeline inputs.