The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it extracts and processes text content from user-provided PDF files via scripts/extract_form_structure.py and scripts/extract_form_field_info.py.
Ingestion points: Data enters the agent context through PDF text extraction in SKILL.md, forms.md, and the scripts folder.
Boundary markers: No explicit boundary markers or 'ignore' instructions are used when interpolating PDF content into prompts.
Capability inventory: The skill can execute subprocesses (magick, qpdf, pdftotext), write files to the system, and perform network operations if certain libraries (like pytesseract) are configured to do so.
Sanitization: There is no evidence of sanitization or filtering of the extracted text before it is presented to the AI agent.
[COMMAND_EXECUTION]: The documentation in forms.md and SKILL.md instructs the agent to execute various CLI tools for PDF and image manipulation.
Evidence: Instructions include the use of qpdf, pdftotext, pdftk, and magick (ImageMagick) to process files.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py performs a runtime modification (monkeypatch) of a library method.
Evidence: The monkeypatch_pydpf_method function replaces pypdf.generic.DictionaryObject.get_inherited with a custom implementation to handle specific field return types.

pdf