pdf

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The guide includes literal passwords placed directly in command-line arguments and in code (e.g., qpdf --password=mypassword and writer.encrypt("userpassword", "ownerpassword")), which requires embedding secret values verbatim and exposes them in outputs — a high exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.65). The workflow ingests outsider-authored free text contained in a runtime-supplied PDF (e.g., OCR-extracted text or extracted word/annotation text from input.pdf) into LLM-readable JSON/strings/console output—there is no LLM context isolation shown—via scripts like scripts/extract_form_structure.py (reads word["text"]) and then fill_pdf_form_with_annotations.py (uses entry_text["text"]).