pptgen-drawio

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes local Python scripts and external command-line utilities to handle document processing.\n
  • scripts/analyze_pptx.py parses user-provided PowerPoint templates to extract style metadata.\n
  • scripts/postprocess_pptx.py applies fixes to generated PPTX files, such as forcing word wrap for Chinese text and injecting speaker notes.\n
  • scripts/render_check.py invokes subprocess.run to call soffice (LibreOffice) and pdftoppm for converting slides to images. On Windows, it uses win32com.client to interface with the local PowerPoint application for rendering verification.\n- [EXTERNAL_DOWNLOADS]: The instructions recommend installing the drawio2pptx package from PyPI. It suggests using the Tsinghua University open-source mirror (pypi.tuna.tsinghua.edu.cn), which is a well-known and trusted service provided by a reputable academic institution.\n- [INDIRECT_PROMPT_INJECTION]: The skill functions as a generator that interpolates user-provided content into structured XML data, creating an attack surface for indirect injection.\n
  • Ingestion points: User-supplied paper text and report outlines are processed in the Step 1 phase of the SKILL.md.\n
  • Boundary markers: The instructions advise on using XML entity escaping (e.g., &, <) and specific line-ending characters (
) to ensure data is correctly rendered within the Draw.io format.\n
  • Capability inventory: The skill maintains the ability to write local files and execute the processing scripts described in the command execution section.\n
  • Sanitization: The skill relies on the agent to correctly escape user data before inserting it into the XML templates, minimizing the risk of document corruption or unexpected behavior during rendering.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 02:46 AM