The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill's primary function involves ingesting and processing untrusted PDF documents. This creates a surface for indirect prompt injection, where malicious instructions hidden in a PDF (e.g., in text, tables, or metadata) could attempt to influence the agent's logic. (Files: SKILL.md, scripts/convert_pdf_to_images.py, scripts/extract_form_field_info.py)\n
Ingestion points: Processes PDF pages via pypdf, pdfplumber, and pdf2image.\n
Boundary markers: No explicit delimiters or safety instructions are used to separate untrusted document content from the agent's system instructions.\n
Capability inventory: The skill allows file system writes and execution of PDF utilities.\n
Sanitization: No sanitization of extracted text is performed before it is presented to the agent.\n- Command Execution (LOW): The skill documentation provides examples for the agent to execute various command-line tools like pdftotext, qpdf, and pdftk for document manipulation. (File: SKILL.md)\n- Dynamic Execution (LOW): The script scripts/fill_fillable_fields.py performs a runtime monkeypatch of the pypdf library to fix a known bug in selection list field handling in version 5.7.0. While this is a form of dynamic code modification, it is hardcoded, well-documented, and localized to a specific version workaround. (File: scripts/fill_fillable_fields.py)

pdf