The Agent Skills Directory

[Indirect Prompt Injection] (MEDIUM): The skill processes untrusted PDF documents provided by users and converts them into images for visual analysis by the agent (scripts/convert_pdf_to_images.py). This creates an attack surface where a malicious PDF could contain text or visual cues designed to influence or hijack the agent's instructions.
Ingestion points: scripts/convert_pdf_to_images.py and scripts/extract_form_field_info.py read PDF data.
Boundary markers: None identified. The agent is instructed to visually analyze the generated images.
Capability inventory: Local file read/write operations for PDF and image processing. No network or arbitrary command execution capabilities were found.
Sanitization: No sanitization or filtering of the PDF content or extracted metadata is performed.
[Dynamic Execution] (LOW): The script scripts/fill_fillable_fields.py performs runtime monkeypatching of the pypdf library's DictionaryObject.get_inherited method. While used to fix a specific bug in the library's handling of selection lists, runtime modification of external dependencies is a security risk.
[Command Execution] (LOW): The skill relies on the execution of several internal Python scripts (scripts/check_fillable_fields.py, scripts/extract_form_field_info.py, etc.) to perform its functions. These scripts are local and utilize standard, well-known libraries.

pdf