read-docx-review

Pass

Audited by Gen Agent Trust Hub on Aug 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses dotnet-script to execute tasks, which automatically downloads the DocumentFormat.OpenXml package from NuGet. This is a well-known library maintained by Microsoft used for processing Word documents.
  • [COMMAND_EXECUTION]: The Python component (scripts/bridge_lib/docx_bridge_client.py) invokes shell commands via subprocess.run to run C# scripts. It uses list-based argument passing which mitigates shell injection risks, although it passes user-provided file paths to the scripts.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content (comments and revisions) from Word files and renders them into Markdown. This creates a surface where malicious instructions in a document could influence an agent reading the output.
  • Ingestion points: scripts/csharp/tasks/list_comments.csx, scripts/csharp/tasks/list_revisions.csx, and scripts/csharp/tasks/extract_views.csx using WordprocessingDocument.Open.
  • Boundary markers: Absent; content is placed directly into Markdown tables.
  • Capability inventory: subprocess.run execution of external scripts in scripts/bridge_lib/docx_bridge_client.py.
  • Sanitization: Absent; the code only escapes pipe characters for Markdown table compatibility.
  • [DYNAMIC_EXECUTION]: The skill employs a polyglot approach, dynamically compiling and running C# (.csx) scripts at runtime to handle complex OpenXML parsing that Python libraries often miss.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 29, 2026, 03:05 AM
Security Audit — agent-trust-hub — read-docx-review