The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF documents by extracting text, tables, and form field metadata (via pdfplumber and pypdf). The agent is then instructed in forms.md to analyze this content to determine field purposes and appropriate values. This constitutes a standard injection surface where a malicious document could contain hidden instructions intended to influence the agent's logic. However, the instructions are scoped to document automation, and the risk is inherent to the skill's primary utility.
Ingestion points: Data enters the agent's context through extract_form_field_info.py (reading field names and choices) and various text extraction methods described in SKILL.md.
Boundary markers: No specific delimiters or warnings for embedded instructions are used when displaying extracted content to the agent.
Capability inventory: The skill possesses capabilities for local file read/write (PDFs, images, and JSON metadata) and execution of provided local Python scripts.
Sanitization: No explicit sanitization or filtering is performed on the data extracted from PDF objects or text streams before processing.
[EXTERNAL_DOWNLOADS]: The documentation references installation of standard, reputable PDF libraries including pypdf, pdfplumber, reportlab, pytesseract, and pdf2image. These are well-established packages in the Python ecosystem and do not represent a security risk.
[COMMAND_EXECUTION]: The skill includes several Python scripts designed to automate complex PDF workflows. The agent is guided to execute these scripts (e.g., fill_fillable_fields.py, convert_pdf_to_images.py) in a specific sequence to achieve tasks like form filling and visual validation. These scripts perform localized document transformations and do not interact with sensitive system files or network resources.

pdf