docx

Pass

Audited by Gen Agent Trust Hub on May 27, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: Internal validation scripts in the skill use XML parsers that are vulnerable to XML External Entity (XXE) attacks by default.
  • Evidence in ooxml/scripts/validation/base.py (uses lxml.etree) and ooxml/scripts/validation/redlining.py (uses xml.etree.ElementTree).
  • These parsers can be exploited by maliciously crafted document files to read local system files or perform server-side request forgery (SSRF) during the validation phase.
  • While the skill's main editing utilities correctly use the defusedxml library, the inclusion of unsafe parsers in the validation logic creates an exposure risk.
  • [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it is designed to ingest and process data from untrusted Word documents.
  • Ingestion points: Document XML files are unpacked and their text content is read into the agent's context in scripts/document.py and scripts/utilities.py.
  • Boundary markers: The instructions do not define delimiters or provide warnings for the agent to ignore instructions that might be embedded within the document text.
  • Capability inventory: The skill has the capability to execute shell commands (git, soffice) and perform file system writes (doc.save()).
  • Sanitization: The skill uses the defusedxml library in its core manipulation scripts to mitigate XXE risks during primary ingestion.
  • [COMMAND_EXECUTION]: The skill uses subprocess.run to execute external commands for document diffing and validation.
  • Evidence in ooxml/scripts/pack.py (calls soffice) and ooxml/scripts/validation/redlining.py (calls git).
  • These subprocess calls use list-based arguments and target local files, which minimizes the risk of arbitrary command injection.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation identifies several standard third-party tools and libraries required for operation.
  • These include system packages like pandoc, LibreOffice, and poppler-utils.
  • Referenced libraries include the docx Node.js package and the defusedxml Python package.
Audit Metadata
Risk Level
SAFE
Analyzed
May 27, 2026, 02:36 PM
Security Audit — agent-trust-hub — docx