The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several well-known and standard system utilities for PDF processing, such as pdftotext, qpdf, pdftoppm, and pdfimages. These are invoked via shell commands to perform legitimate document transformations.
[EXTERNAL_DOWNLOADS]: The skill requires standard Python packages (pypdf, pdfplumber, reportlab, pytesseract, pdf2image) and system packages (poppler-utils, qpdf, tesseract-ocr) from official repositories. These are standard prerequisites for the described functionality.
[DYNAMIC_EXECUTION]: In scripts/fill_fillable_fields.py, the skill applies a runtime monkeypatch to the pypdf library's DictionaryObject.get_inherited method. This is used to correct specific behavior regarding field options in the PDF specification and does not involve executing untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data by extracting text and structure from user-provided PDF files. While this creates a surface for indirect prompt injection (e.g., instructions hidden within a PDF's text), this is a standard risk for any document processing tool, and no malicious intent was found in the skill's own logic.
Ingestion points: PDF files are read using pdfplumber.open() and pypdf.PdfReader() across multiple scripts.
Boundary markers: None identified in the extraction scripts.
Capability inventory: The skill has the ability to write files and execute shell commands (qpdf, pdftotext).
Sanitization: No specific sanitization of extracted text is performed before it is presented to the agent.

pdf