pptx

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on several external system utilities executed via subprocess.run to handle presentation processing and conversion:
  • soffice (from LibreOffice) is invoked in ooxml/scripts/pack.py and scripts/thumbnail.py to validate document structures and convert presentations to PDF format.
  • pdftoppm (from Poppler-utils) is utilized in scripts/thumbnail.py to generate image thumbnails for slide previews.
  • git is used in ooxml/scripts/validation/redlining.py to perform word-level diffing between document versions for change tracking.
  • [COMMAND_EXECUTION]: The scripts/html2pptx.js library uses the playwright framework to launch a headless Chromium browser instance. This is a core part of the workflow for rendering HTML slide layouts and calculating element positions for precise PowerPoint conversion.
  • [PROMPT_INJECTION]: The skill extracts text content from user-provided .pptx files through markitdown and scripts/inventory.py. This content is then inserted into the agent's context, which creates an indirect prompt injection surface. A malicious presentation could contain text designed to trick the agent into ignoring its safety guidelines or executing unintended actions.
  • [SAFE]: The skill demonstrates excellent security hygiene by using the defusedxml library for parsing potentially untrusted Office Open XML files, effectively mitigating risks from XML External Entity (XXE) and billion laughs attacks.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 06:18 PM
Security Audit — agent-trust-hub — pptx