The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted PDF data, creating a surface for indirect prompt injection where malicious documents could attempt to influence the agent's logic. This is a common characteristic of document manipulation tools.
Ingestion points: PDF files are parsed in scripts/extract_form_field_info.py, scripts/fill_fillable_fields.py, and scripts/check_fillable_fields.py.
Boundary markers: Absent; extracted text is not delimited with specific warnings to the agent.
Capability inventory: The skill can write PDF files using pypdf and execute command-line tools.
Sanitization: No specific filtering or escaping is applied to extracted text before it is returned to the agent context.
[COMMAND_EXECUTION]: The toolkit references and provides instructions for using standard external command-line utilities for PDF operations, such as qpdf, pdftotext, pdftoppm, and pdftk. These are industry-standard tools used for their intended purpose.
[REMOTE_CODE_EXECUTION]: The script scripts/fill_fillable_fields.py implements a targeted monkeypatch of the pypdf library to resolve a specific handling issue with selection list fields. This runtime modification is limited to internal library logic and does not involve executing untrusted or remote code.

pdf