The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill extracts and processes text content from external PowerPoint files, creating a potential surface for indirect prompt injection if malicious instructions are embedded in the slide data.
Ingestion points: Text is extracted from presentations in scripts/inventory.py and via the markitdown utility mentioned in SKILL.md.
Boundary markers: Extracted text is not explicitly wrapped in delimiters or accompanied by warnings to ignore embedded instructions.
Capability inventory: The skill possesses capabilities for command execution via subprocess.run (in pack.py, thumbnail.py, redlining.py) and network-adjacent rendering via Playwright in scripts/html2pptx.js.
Sanitization: Extracted text content is processed without filtering or sanitization before being provided to the agent context.
[COMMAND_EXECUTION]: The skill uses system utilities via the subprocess module to perform document conversions and validation tasks.
Evidence: ooxml/scripts/pack.py, ooxml/scripts/validation/redlining.py, and scripts/thumbnail.py call soffice, pdftoppm, and git diff using subprocess.run with list-based arguments, minimizing command injection risk.
[DYNAMIC_EXECUTION]: The skill utilizes a headless browser environment to render HTML and JavaScript content for the purpose of slide generation.
Evidence: scripts/html2pptx.js uses Playwright to render local HTML files and calculate element positioning, which involves executing JavaScript within a sandboxed browser context.

pptx