The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses utility scripts to perform document operations via subprocess.run. It invokes standard tools like soffice (LibreOffice) for format conversion, pdftoppm (Poppler) for image generation, and git for comparing document revisions. These operations use safe list-based arguments and are consistent with the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill relies on standard dependencies from official registries. Python dependencies include python-pptx, Pillow, defusedxml, lxml, and markitdown. Node.js dependencies include pptxgenjs, playwright, sharp, react, react-dom, and react-icons. It also recommends system-level installations of LibreOffice and Poppler.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided PowerPoint files to extract text content and analyze layout metadata. This creates a risk surface for indirect prompt injection if the input documents contain malicious instructions intended for the AI. This is addressed as follows:
Ingestion points: User-provided .pptx files are unpacked, and their content is extracted via inventory.py and markitdown.
Boundary markers: Extracted content is handled as structured JSON or Markdown text, which helps distinguish it from system instructions.
Capability inventory: The skill can execute system commands for document conversion and rendering via subprocess.run (in pack.py, thumbnail.py, and redlining.py) and utilizes playwright for slide rendering.
Sanitization: The skill uses defusedxml to safely parse XML content and prevent XML External Entity (XXE) attacks.

pptx