The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted .docx and .pptx files, creating an ingestion surface for malicious instructions embedded in document content.
Ingestion points: ooxml/scripts/unpack.py and ooxml/scripts/validation/docx.py.
Boundary markers: None identified in the processing scripts.
Capability inventory: Executes soffice via subprocess in ooxml/scripts/pack.py and performs file-system modifications in scripts/rearrange.py.
Sanitization: Utilizes defusedxml to protect against XML External Entity (XXE) and expansion attacks during parsing.
Command Execution (LOW): The skill invokes the system binary soffice (LibreOffice) to perform document validation and format conversion.
Evidence: subprocess.run call in ooxml/scripts/pack.py within the validate_document function. The command uses a fixed binary name and hardcoded filter names, though it operates on user-provided file paths.

pptx