The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke standard system utilities for document processing. This includes soffice (LibreOffice) for PDF conversion and validation in scripts/thumbnail.py and ooxml/scripts/pack.py, pdftoppm (Poppler) for image generation in scripts/thumbnail.py, and git for calculating text differences in ooxml/scripts/validation/redlining.py. These operations are transparently documented as core features for presentation handling.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of .pptx files, which constitutes a potential surface for indirect prompt injection.
Ingestion points: Untrusted text is extracted from PowerPoint files using markitdown and scripts/inventory.py as part of the analysis workflow.
Boundary markers: Not explicitly defined in the prompts for separating user content from instructions.
Capability inventory: The environment allows file system access and shell command execution via documented scripts (thumbnail.py, pack.py, redlining.py).
Sanitization: The skill effectively mitigates XML-based attacks (such as XXE) by using the defusedxml library for all XML parsing operations in ooxml/scripts/unpack.py and ooxml/scripts/pack.py.
[SAFE]: The skill demonstrates security awareness by employing defusedxml and relies on standard, well-known libraries like python-pptx, playwright, and sharp. All detected behaviors align with the skill's stated purpose of PowerPoint manipulation.

pptx