The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface through the ingestion of untrusted PDF content.
Ingestion points: PDF metadata and page content are extracted and processed using scripts/extract_form_field_info.py and scripts/convert_pdf_to_images.py (which renders pages to images for vision analysis).
Boundary markers: No specific delimiters or safety instructions are provided to help the agent distinguish between data and instructions when analyzing extracted PDF content or images.
Capability inventory: The skill allows for file writing (PdfWriter.write) and execution of system commands for document manipulation.
Sanitization: No sanitization of text or image-based content is performed before processing.
[COMMAND_EXECUTION]: The skill instructions and scripts involve executing several system-level PDF utilities.
Evidence: The toolkit provide examples and scripts for using external binaries like qpdf, pdftotext, pdftk, and tesseract for document processing operations.
[EXTERNAL_DOWNLOADS]: The skill documentation references and depends on several external packages.
Evidence: Documentation in SKILL.md and reference.md recommends installing Python packages like pypdf, pdfplumber, reportlab, pdf2image, and pytesseract, and references JavaScript libraries such as pdf-lib and pdfjs-dist in the advanced section.

pdf