The Agent Skills Directory

[PROMPT_INJECTION]: The references/forms.md file contains strong procedural instructions (e.g., 'CRITICAL: You MUST complete these steps in order. Do not skip ahead to writing code.') designed to enforce a specific validation workflow for form filling.
[DYNAMIC_EXECUTION]: The scripts/fill_fillable_fields.py script implements a runtime monkeypatch of the pypdf library (DictionaryObject.get_inherited) to resolve a known bug when handling PDF selection list fields. This modification occurs during script execution to ensure correct data processing.
[COMMAND_EXECUTION]: The skill documentation (SKILL.md and references/reference.md) provides numerous examples and instructions for the agent to execute shell-based PDF utilities including qpdf, pdftotext, pdftk, and pdfimages.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it processes untrusted PDF data.
Ingestion points: PDF content is ingested and parsed via pypdf, pdfplumber, and pytesseract (OCR) in various scripts such as scripts/extract_form_field_info.py and scripts/convert_pdf_to_images.py.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat extracted PDF text as untrusted or to ignore instructions contained within the document.
Capability inventory: The skill possesses the ability to write files to the local system and execute a variety of command-line tools and Python scripts.
Sanitization: No sanitization, filtering, or validation is performed on the text extracted from PDF files before it is made available to the agent.

pdf