docx-processing-openai

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION] (HIGH): The skill provides shell templates for soffice and pdftoppm that use unquoted/unsanitized variables like $INPUT_DOCX and $OUTDIR. If an agent processes a file with a malicious name (e.g., ";touch_exploit;" .docx), it could lead to arbitrary command execution on the host system.
  • [PROMPT_INJECTION] (HIGH): This skill is highly susceptible to indirect prompt injection. It instructs the agent to read and 'visually inspect' external DOCX files and then perform edits based on those results. A malicious document could contain hidden or visible instructions that trick the agent into performing unauthorized file modifications, exfiltrating data, or disregarding its safety guidelines during the 'render-inspect-fix' loop.
  • [EXTERNAL_DOWNLOADS] (LOW): The skill requires pip install python-docx. While python-docx is a standard library and the metadata claims 'OpenAI' as the author (a trusted source), runtime package installation remains a supply chain risk factor. Per [TRUST-SCOPE-RULE], this is rated LOW as the package is well-known.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:20 PM