The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from external, untrusted PDF files, which creates an attack surface for indirect prompt injection.
Ingestion points: PDFs are processed through multiple tools including pypdf, pdfplumber, pdftotext, and pytesseract (OCR) as described in SKILL.md and forms.md.
Boundary markers: The skill instructions in forms.md require the agent to generate intermediate JSON metadata and validation images for visual inspection, providing a manual oversight step before final document generation.
Capability inventory: The skill has file-writing capabilities via pypdf, pdfplumber, and reportlab, and executes local Python scripts for validation and conversion.
Sanitization: No explicit string sanitization is present, but the workflow utilizes structured data validation via scripts/check_bounding_boxes.py to verify bounding box logic.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py performs runtime modification of an external library method.
Evidence: The monkeypatch_pydpf_method function replaces pypdf.generic.DictionaryObject.get_inherited with a custom implementation. This is used as a functional workaround to resolve a specific data type conflict in the pypdf library's form-writing logic.

pdf