The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and extract text from user-provided PDF files, which can lead to indirect prompt injection if those files contain malicious instructions.
Ingestion points: Text, labels, and structure are extracted from external PDFs in scripts/extract_form_structure.py, scripts/extract_form_field_info.py, and scripts/check_fillable_fields.py.
Boundary markers: The skill instructions do not specify the use of delimiters or boundary markers (e.g., XML tags or clear separators) when presenting extracted PDF content to the agent.
Capability inventory: The skill has capabilities to write files to the local system (PdfWriter), create images (pdf2image), and execute local utility scripts via the shell.
Sanitization: The extraction scripts do not include sanitization or filtering logic to strip potential instructions from the extracted PDF text before processing.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py utilizes runtime monkeypatching to modify the behavior of the pypdf library.
Evidence: The monkeypatch_pydpf_method function redefines DictionaryObject.get_inherited at runtime to alter how FieldDictionaryAttributes.Opt values are handled. While this is a common developer technique for task-specific workarounds in Python, it represents dynamic modification of executable code logic.

pdf