The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local Python scripts and command-line utilities to perform its core functions. Evidence includes the use of pdftotext, qpdf, pdftk, and pdfimages (poppler-utils) for text and image extraction, and magick or convert (ImageMagick) for image processing and cropping.
[EXTERNAL_DOWNLOADS]: Documentation recommends installing several well-known third-party libraries using package managers. Evidence includes instructions in SKILL.md and reference.md for installing Python packages like pypdf, pdfplumber, reportlab, pytesseract, and others from official registries.
[PROMPT_INJECTION]: Indirect prompt injection surface exists as the skill ingests untrusted data from PDF files. 1. Ingestion points: scripts/extract_form_structure.py and scripts/extract_form_field_info.py read text and structure directly from PDF documents. 2. Boundary markers: Absent. The skill instructions do not specify any delimiters or warnings to treat extracted content as untrusted. 3. Capability inventory: The skill possesses capabilities to write files and execute shell commands, which could be targeted by adversarial content in a PDF. 4. Sanitization: Absent. Extracted labels and text are used to generate coordinates and form content without explicit sanitization logic.

pdf