The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and process content from external PDF files, which is a known vector for untrusted data input.
Ingestion points: PDF content is read using libraries like pypdf, pdfplumber, and pdftotext as implemented in SKILL.md and scripts such as extract_form_structure.py.
Boundary markers: The instructions do not specify the use of clear delimiters or markers to separate the extracted file content from the agent's internal instructions.
Capability inventory: The skill enables operations such as file system access, image processing, and the execution of external command-line utilities.
Sanitization: Content extracted from PDFs is processed directly by the parsing libraries and provided to the agent without an intermediate sanitization layer.
[Command Execution]: The skill's documentation guides the use of various system-level command-line tools for document manipulation.
Evidence: SKILL.md and forms.md describe workflows involving qpdf, pdftotext, pdftk, and magick/convert for tasks like merging files and cropping images.
[Dynamic Execution]: One of the provided scripts modifies the behavior of a dependency at runtime to ensure compatibility.
Evidence: In scripts/fill_fillable_fields.py, the monkeypatch_pydpf_method function is used to alter the pypdf library's DictionaryObject.get_inherited method. This pattern of dynamic library modification is used here to ensure form field attributes are correctly inherited during the filling process.

pdf