The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF documents using libraries like pypdf and pdfplumber, creating a surface for indirect prompt injection where hidden instructions in the document could influence the agent.
Ingestion points: PDF files are opened in SKILL.md (e.g., PdfReader('document.pdf')), FORMS.md, and scripts like extract_form_field_info.py.
Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions found within the processed PDF data.
Capability inventory: The skill can write files (writer.write, image.save), generate reports, and execute CLI tools (qpdf, pdftk).
Sanitization: There is no evidence of sanitization or validation of the extracted content before further use.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py performs runtime monkeypatching of a library method (pypdf.generic.DictionaryObject.get_inherited) to resolve a specific formatting bug in selection list fields.
[COMMAND_EXECUTION]: Documentation encourages usage of CLI tools like pdftotext, qpdf, pdftk, and pdfimages for document manipulation tasks.
[EXTERNAL_DOWNLOADS]: The skill relies on standard third-party libraries including pypdf, pdfplumber, reportlab, pytesseract, pdf2image, pandas, pypdfium2, and Node.js packages pdf-lib and pdfjs-dist.

pdf