The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes PDF documents which may contain untrusted data, creating an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through scripts/extract_form_field_info.py and scripts/check_fillable_fields.py via the pypdf library, as well as various extraction examples in SKILL.md.
Boundary markers: None identified in the provided scripts to distinguish between extracted content and agent instructions.
Capability inventory: The skill supports file system writes (PdfWriter.write), form manipulation, and metadata extraction across multiple utility scripts.
Sanitization: There is no explicit sanitization or validation of the text content extracted from PDFs before it is potentially used by the agent.
[DYNAMIC_EXECUTION]: The skill employs a dynamic monkeypatching technique to address a specific library limitation.
Evidence: scripts/fill_fillable_fields.py modifies the pypdf.generic.DictionaryObject.get_inherited method at runtime to correctly process selection list fields. While this is a targeted fix for a known library issue, it involves runtime modification of class behavior.

pdf