The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run across multiple scripts to execute system utilities for document processing. In scripts/thumbnail.py and ooxml/scripts/pack.py, it calls soffice (LibreOffice) for PDF and HTML conversions and pdftoppm (Poppler) for image generation. In ooxml/scripts/validation/redlining.py, it uses git to perform character-level diffs on XML content. These are legitimate uses of external tools to support the skill's primary purpose. Additionally, the skill includes setup instructions involving sudo apt-get for tool installation.
[EXTERNAL_DOWNLOADS]: The skill depends on several widely-used third-party libraries for its operations. This includes Python packages like markitdown, defusedxml, and python-pptx, as well as Node.js packages like pptxgenjs, playwright, and sharp. All dependencies are standard in the respective domains of document automation and web rendering and are sourced from official registries (PyPI and NPM).
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes untrusted content from PowerPoint presentations provided by users.
Ingestion points: XML content is extracted and parsed in ooxml/scripts/unpack.py and scripts/inventory.py.
Boundary markers: Extracted text is processed without specific boundary markers or instructions to ignore embedded commands.
Capability inventory: The skill has the capability to write to the file system, execute shell commands via subprocesses, and access the network via Playwright.
Sanitization: The skill implements robust sanitization by using defusedxml to parse all XML files, which effectively mitigates XML External Entity (XXE) and billion laughs attacks.

pptx