The Agent Skills Directory

[S AFE]: The skill demonstra tes no malicious behavior. Al l scripts and instructions serve the legitimate purpo se of PDF manipulation and form filling.
[EXTERN AL_DOWNLOADS]: The skill r eferences several standard and reputable libraries for PDF and image processing, including pypdf, pdfplumbe r, reportlab, pytessera ct, pdf2image, pandas , pypdfium2, pdf-lib, and pdfjs-dist. These ar e standard tools within the industry.
[COMMAND_EXE CUTION]: The skill utilize s common command-line utili ties such as qpdf, pdftk , and poppler-utils (via pdftotext and pdfimages ) for efficient document h andling. These operations a re transparently documented and follow standard usage p atterns.
[DYNAMIC_EXECU TION]: In scripts/fill_fi llable_fields.py, a contro lled monkeypatch is applied to the pypdf library to fix a specific bug regardin g selection lists. This is a targeted technical fix th at does not introduce arbit rary code execution risks.
[INDIRECT_PROMPT_INJECTI ON]: The skill processes u ntrusted PDF files, which p resents a potential attack surface. However, the skill provides a robust multi-ste p workflow in forms.md th at requires the agent to pe rform automated validation (check_bounding_boxes.py) and visual verification (via create_validation_image. py) before final document generation. This structured approach significantly miti gates the risk of the agent being manipulated by conten t within processed document s.

pdf-official