The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external command-line utilities using Python's subprocess module to perform document conversion and version comparison.
scripts/thumbnail.py executes soffice to generate intermediate PDFs and pdftoppm to produce slide thumbnails.
references/ooxml/scripts/pack.py calls soffice headlessly to validate the integrity of OOXML packages.
references/ooxml/scripts/validation/redlining.py uses git diff to identify and validate tracked changes within document XML.
[DYNAMIC_EXECUTION]: The skill utilizes browser automation through Playwright in scripts/html2pptx.js to render HTML templates into slide layouts.
It employs page.evaluate() to run internal JavaScript for measuring dimensions and extracting CSS properties like gradients and shadows.
This runtime execution in a browser environment is a necessary part of its layout engine but represents a dynamic execution vector.
[INDIRECT_PROMPT_INJECTION]: The skill architecture facilitates the ingestion of data from external files, which can contain adversarial instructions targeting the LLM.
Ingestion points: Data is ingested through scripts/inventory.py (reading slide content from .pptx) and scripts/replace.py (reading text from .json maps).
Boundary markers: There are no explicit instructions or delimiters used to separate processed document text from agent instructions.
Capability inventory: The skill possesses capabilities including file system writes, browser automation, and execution of system binaries.
Sanitization: Extracted content is processed as raw text without filtering for embedded instructional markers, leaving the surface exposed to indirect injection.

powerpoint