docx
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document content, creating a standard surface for indirect prompt injection. This risk is mitigated by its intended design and secure implementation.\n
- Ingestion points: ooxml/scripts/unpack.py and ooxml/scripts/validation/redlining.py ingest XML data from documents.\n
- Boundary markers: Absent for document text content, though structural XML elements are processed.\n
- Capability inventory: subprocess.run is used for soffice and git commands in ooxml/scripts/pack.py and ooxml/scripts/validation/redlining.py.\n
- Sanitization: Robust use of defusedxml across all parsing operations prevents XML-based attacks.\n- [COMMAND_EXECUTION]: Legitimate use of system binaries for document structure processing.\n
- Invokes soffice for conversion and validation tasks.\n
- Invokes git diff for tracked change validation.\n
- Invokes pandoc for markdown conversion.\n
- All invocations utilize list-based arguments to prevent command injection.
Audit Metadata