The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted PDF documents, creating a surface for indirect prompt injection where instructions embedded in a PDF could attempt to influence the agent's behavior during analysis or form filling.
Ingestion points: pypdf.PdfReader and pdfplumber.open are used throughout SKILL.md and referenced scripts to read external data.
Boundary markers: While forms.md instructs the agent to visually inspect images, there are no explicit instructions or delimiters used to warn the agent to ignore text-based instructions found within the document.
Capability inventory: The skill has the ability to write files to the local system (writer.write), save images (image.save), and execute shell commands (qpdf, pdftk, poppler-utils).
Sanitization: The script scripts/create_validation_image.py includes a safe_user_path function that resolves and validates CLI paths against the current workspace to prevent path traversal attacks, representing a security best practice.
[EXTERNAL_DOWNLOADS]: The skill references several well-known and standard libraries for PDF processing. These include pypdf, pdfplumber, reportlab, pytesseract, pdf2image, and pandas for Python, as well as pdf-lib and pdfjs-dist for JavaScript environments mentioned in reference.md.
[COMMAND_EXECUTION]: The skill utilizes standard command-line utilities for PDF manipulation, including qpdf, pdftk, and poppler-utils (pdftotext, pdfimages, pdftoppm).
[REMOTE_CODE_EXECUTION]: The script scripts/fill_fillable_fields.py performs a dynamic monkeypatch on the pypdf.generic.DictionaryObject.get_inherited method at runtime to address a specific bug in selection list field handling. While functional, this represents dynamic code modification.

pdf-official