The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation and scripts reference several external dependencies that must be present in the environment.
Python libraries: pypdf, pdfplumber, reportlab, pytesseract, pdf2image, pillow, pandas, and pypdfium2.
JavaScript libraries: pdf-lib and pdfjs-dist.
System utilities: qpdf, pdftk, and poppler-utils (for pdftotext and pdfimages).
[INDIRECT_PROMPT_INJECTION]: The skill processes external PDF documents, which are a source of untrusted data. Instructions in forms.md guide the agent to perform visual analysis on these documents to determine form field purposes.
Ingestion points: Files are read and processed in scripts/convert_pdf_to_images.py and scripts/extract_form_field_info.py.
Boundary markers: The instructions do not specify any delimiters or warnings for the agent to ignore instructions embedded within the PDF content during analysis.
Capability inventory: The skill includes file-writing capabilities via pypdf and provides examples for executing system commands via various PDF utilities.
Sanitization: There is no evidence of sanitization or filtering of the extracted text or visual data before it is presented to the agent.
[DYNAMIC_EXECUTION]: The file scripts/fill_fillable_fields.py contains a runtime monkeypatch of the pypdf.generic.DictionaryObject.get_inherited method. This is used to fix a specific bug in the pypdf library regarding selection list fields. It does not incorporate untrusted external input into the logic of the patch.

pdf