creating-presentations
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from external PowerPoint presentations and HTML templates, creating a surface for indirect prompt injection where malicious instructions embedded in slides could target the agent's logic.
- Ingestion points: The skill reads .pptx files via markitdown and custom inventory scripts, and processes HTML templates for conversion in references/html2pptx.md.
- Capability inventory: The skill performs file system writes (outputting .pptx and .json files) and executes multiple local scripts and system tools.
- Sanitization: While the skill correctly uses defusedxml to prevent XML External Entity (XXE) attacks, it does not implement specific boundary markers or sanitization for text content extracted from presentation shapes and notes.
- [COMMAND_EXECUTION]: The skill uses several local command-line utilities and custom Python/JavaScript scripts to unpack, analyze, and repack presentation files, including markitdown, libreoffice, and poppler-utils.
- [EXTERNAL_DOWNLOADS]: The skill documentation lists several standard dependencies for installation via pip, npm, brew, and apt-get. All referenced packages (such as playwright, python-pptx, and sharp) are reputable, well-known libraries for document processing and rendering.
Audit Metadata