The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted PDF files through multiple ingestion points, including text extraction and image conversion, which exposes the agent to adversarial content embedded in documents.
Ingestion points: scripts/extract_form_field_info.py, scripts/extract_form_structure.py, and scripts/convert_pdf_to_images.py read and parse PDF data.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions when presenting extracted text to the agent.
Capability inventory: The skill has significant local capabilities including writing files, filling PDF forms, and executing CLI tools like magick and qpdf.
Sanitization: There is no evidence of sanitization or filtering of text extracted from PDF documents before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructions in forms.md direct the agent to execute shell commands for image manipulation, specifically using the magick or convert utility from ImageMagick.
Evidence: magick <page_image> -crop <width>x<height>+<x>+<y> +repage <crop_output.png> in forms.md.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py performs runtime modification of the pypdf library to resolve specific field attribute handling issues.
Evidence: The script monkeypatches DictionaryObject.get_inherited within the pypdf namespace to alter how form field options are retrieved.
[EXTERNAL_DOWNLOADS]: Documentation and scripts reference multiple third-party Python and JavaScript libraries for PDF handling.
Evidence: SKILL.md and reference.md mention pypdf, pdfplumber, reportlab, weasyprint, pdf-lib, and pdfjs-dist.

pdf