The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call external system tools such as soffice (LibreOffice), pdftoppm (Poppler), and git. These calls are executed using argument lists rather than shell strings, which is a secure practice that prevents shell injection attacks. These tools are used for legitimate validation and conversion tasks related to the skill's primary purpose.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted content from .pptx files, HTML slide definitions, and JSON replacement data. While it includes mitigations like defusedxml to prevent XML-based attacks, the ingestion of complex document formats inherently creates a surface for indirect prompt injection if the agent interprets content within the slides as instructions. This is a low-risk finding inherent to document processing tasks.
[DYNAMIC_EXECUTION]: The html2pptx.js script uses the playwright library to render HTML slides in a Chromium browser environment. Although used for rendering layout and styles, this mechanism could theoretically execute JavaScript contained within provided HTML content. This behavior is scoped to the rendering workflow and is necessary for the skill's functionality.

pptx