The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py uses monkeypatching to modify the pypdf library at runtime. It replaces the pypdf.generic.DictionaryObject.get_inherited method with a custom implementation to resolve a specific issue with selection list fields in pypdf version 5.7.0. This represents a dynamic code modification of an external dependency.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF documents, creating a surface for indirect prompt injection where document content could attempt to influence the agent's behavior.
Ingestion points: PDF documents are ingested and processed using pypdf, pdfplumber, and pdf2image (in scripts/convert_pdf_to_images.py).
Boundary markers: No explicit delimiters or warnings are used to instruct the agent to ignore potentially malicious instructions embedded within the PDF content.
Capability inventory: The skill provides scripts that can read from and write to the local file system.
Sanitization: There is no evidence of sanitization or filtering of the text or visual content extracted from PDFs before it is presented to the agent for analysis.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts reference a large number of third-party libraries and system tools. These include Python packages such as pypdf, pdfplumber, reportlab, pandas, pytesseract, pdf2image, pypdfium2, Pillow, and numpy, as well as Node.js packages like pdf-lib and pdfjs-dist. It also depends on command-line utilities including qpdf, poppler-utils, and pdftk. While these are established tools, they represent a significant dependency footprint.

pdf