The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted PDF files from the user's environment, creating a surface for indirect prompt injection.
Ingestion points: PDF content is read via pypdf.PdfReader in scripts/extract_form_field_info.py, pdfplumber.open in scripts/extract_form_structure.py, and convert_from_path in scripts/convert_pdf_to_images.py.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious instructions embedded within the PDF text or metadata.
Capability inventory: The skill possesses file writing capabilities (via pypdf, PIL, and pdfplumber) and can output extracted text to the agent's context.
Sanitization: No sanitization or filtering is performed on the extracted text or form data before it is presented to the agent.
[DYNAMIC_EXECUTION]: The skill utilizes runtime modification of library behavior to handle specific PDF form structures.
Evidence: The script scripts/fill_fillable_fields.py contains a monkeypatch_pydpf_method function that replaces DictionaryObject.get_inherited in the pypdf library with a custom implementation at runtime. While this is used for functional correction of choice field handling, dynamic execution and library patching are notable patterns.

pdf