docx
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external Word (.docx) files which serves as a data ingestion surface.
- Ingestion points: Word documents are read via
pandocand unpacked usingunpack.py(referenced in SKILL.md). - Boundary markers: No specific delimiters or "ignore instructions" wrappers are mentioned for the document content.
- Capability inventory: The skill utilizes command execution for conversion (
pandoc,soffice,pdftoppm) and file writing capabilities (pack.py,doc.save()). - Sanitization: The skill explicitly instructs the use of
defusedxml(pip package) for XML processing, which is a recognized best practice for preventing XML External Entity (XXE) and billion laughs attacks. - [COMMAND_EXECUTION]: The skill workflow involves standard system utilities for document processing.
- Evidence: Usage of
pandocfor markdown conversion,soffice(LibreOffice) for PDF conversion, andpdftoppm(poppler-utils) for image extraction.
Audit Metadata