The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted PDF files and uses their content, including field IDs, structure, and text, to drive agent actions and script parameters, creating a surface for indirect prompt injection.\n
Ingestion points: The script scripts/extract_form_field_info.py reads internal PDF structure, while scripts/convert_pdf_to_images.py generates images for visual analysis.\n
Boundary markers: The instructions in forms.md lack explicit delimiters or warnings to ignore potential commands embedded in PDF metadata or visible text.\n
Capability inventory: The agent can execute scripts that modify the filesystem, specifically scripts/fill_fillable_fields.py and scripts/fill_pdf_form_with_annotations.py, based on data extracted from processed PDFs.\n
Sanitization: No sanitization or escaping of field IDs or values is performed before they are processed by internal scripts.\n- [REMOTE_CODE_EXECUTION]: The script scripts/fill_fillable_fields.py performs runtime monkeypatching of the pypdf library to resolve a functional bug in selection list handling.\n
Evidence: The monkeypatch_pydpf_method function modifies the DictionaryObject.get_inherited method at runtime. While this is a targeted fix for library compatibility within the skill's primary purpose, runtime modification of library behavior is a dynamic execution pattern that could potentially be used to alter execution flow.

pdf