docx-toolkit
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external
.docxfiles and extracts heading text for the agent to review. This creates a surface for indirect prompt injection if a processed document contains adversarial instructions within its heading text.\n - Ingestion points: The
scripts/docx_auditor.pyscript reads paragraph and heading data fromword/document.xmlwithin the ZIP-formatted.docxarchive.\n - Boundary markers: The audit report labels the extracted text under a 'Heading hierarchy' section, providing structural context to the agent, though no explicit 'ignore instructions' warnings are added to the output.\n
- Capability inventory: The skill is limited to read-only file access and standard library XML parsing. It lacks capabilities for network access, file modification, or command execution.\n
- Sanitization: The script does not sanitize or filter the extracted heading text before it is presented to the agent.
Audit Metadata