The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess.run function to execute system binaries essential for presentation processing. This includes calling soffice (LibreOffice) for format conversion, pdftoppm for image generation, and git for validating tracked changes. These calls are located in scripts/thumbnail.py, ooxml/scripts/pack.py, and ooxml/scripts/validation/redlining.py.
[EXTERNAL_DOWNLOADS]: The skill manages dependencies from official, well-known registries. It uses npm to install packages such as pptxgenjs, playwright, and sharp, and also automates the installation of the Chromium browser for rendering tasks. Python dependencies like python-pptx, Pillow, and markitdown are resolved via uv run as specified in the script headers and SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, user-controlled data from PowerPoint files, creating an attack surface for indirect prompt injection where malicious instructions could be embedded in slide content.
Ingestion points: Untrusted content from .pptx files is read into the agent's context through scripts/inventory.py and ooxml/scripts/unpack.py.
Boundary markers: The scripts do not explicitly apply delimiters to the extracted text content before processing.
Capability inventory: The skill has capabilities for file system read/write operations, ZIP archive extraction, and the execution of specific document-processing system tools.
Sanitization: To mitigate XML-based vulnerabilities such as XML External Entity (XXE) injection, the skill correctly employs the defusedxml library in its XML processing scripts (ooxml/scripts/unpack.py and ooxml/scripts/pack.py).

pptx