The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to interface with local system utilities required for document conversion and analysis. These include LibreOffice (soffice) for converting presentations to PDF/HTML, Poppler (pdftoppm) for rendering slide thumbnails, and git for generating text-based diffs for change tracking validation. These operations are conducted using explicit argument lists and target internal or validated file paths.
Evidence: scripts/thumbnail.py, ooxml/scripts/pack.py, ooxml/scripts/validation/redlining.py.
[PROMPT_INJECTION]: Documentation within the skill employs emphatic instructional language (e.g., 'CRITICAL', 'MANDATORY', 'NEVER') to guide the AI agent's workflow. These patterns are used legitimately to ensure full context loading of technical schemas and to maintain layout integrity, rather than to override model safety filters or bypass system constraints.
Evidence: SKILL.md, html2pptx.md.
[SAFE]: The skill actively mitigates XML-related vulnerabilities like XML External Entity (XXE) injection by utilizing the defusedxml library for all internal XML parsing and formatting tasks. This is a best-practice security measure for handling Office Open XML (OOXML) file structures.
Evidence: ooxml/scripts/unpack.py, ooxml/scripts/pack.py.

powerpoint