space-multi-design-ppt
Warn
Audited by Gen Agent Trust Hub on Aug 5, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches brand design tokens using the command 'npx -y getdesign@latest' from the NPM registry. It also requires the installation of multiple Python packages including playwright, python-pptx, img2pdf, and Pillow, along with system-level browser binaries.
- [REMOTE_CODE_EXECUTION]: The execution of the 'getdesign' package via npx at runtime constitutes remote code execution. The package is fetched without version pinning (@latest), which introduces supply chain risks.
- [COMMAND_EXECUTION]: The skill invokes several command-line utilities including python3, soffice (LibreOffice), pdftoppm, and npx. It also executes local Python scripts (build_deck.py, export_deck.py) to manage slide assembly and format conversion.
- [PROMPT_INJECTION]: The 'Cowrite 一键 Slide 模式' instructions direct the agent to skip user confirmations for design styles and outline approval. This increased autonomy reduces human oversight and could be exploited if the agent is processing untrusted or malicious input.
- [DATA_EXFILTRATION]: The skill processes user-provided content and files (md, docx, pdf, txt) to generate slides. While no malicious exfiltration was detected, the agent's network access via package managers and NPM, combined with the capability to read local files and write to temporary directories, creating a potential attack surface for data exposure.
Audit Metadata