The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external PDF files, which is a known surface for indirect prompt injection attacks where malicious content in a document could influence agent behavior.
Ingestion points: PDF reading operations in SKILL.md and scripts such as scripts/extract_form_field_info.py and scripts/extract_form_structure.py.
Boundary markers: None identified; the skill does not explicitly instruct the agent to ignore instructions embedded within the processed PDF content.
Capability inventory: The skill can execute various CLI tools (qpdf, pdftotext, magick) and write files to the filesystem.
Sanitization: No sanitization of the extracted PDF text or metadata is implemented before the data is processed by the agent.
[COMMAND_EXECUTION]: The skill documentation provides instructions for the agent to use several command-line utilities for PDF manipulation.
Evidence: Instructions for using pdftotext, qpdf, pdftk, and magick (ImageMagick) are documented in SKILL.md and forms.md for PDF processing and image refinement.
[EXTERNAL_DOWNLOADS]: The skill references several well-known libraries for PDF processing, which are treated as safe resources.
Evidence: Mentions of standard packages including pypdf, pdfplumber, reportlab, pytesseract, pdf2image, pypdfium2, pdf-lib, and pdfjs-dist.

pdf