The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted presentation files (.pptx), creating a surface for indirect prompt injection where malicious content in slides could attempt to influence agent behavior.
Ingestion points: PPTX files are converted to Markdown or unpacked for XML analysis (SKILL.md).
Boundary markers: Not explicitly implemented in the instructions for separating extracted content from agent logic.
Capability inventory: Full shell access, Python script execution, and Node.js script execution (SKILL.md).
Sanitization: Mitigates common XML vulnerabilities by using the defusedxml library for parsing operations.
[COMMAND_EXECUTION]: The workflow relies on executing system utilities and local scripts for document processing.
Evidence: Uses soffice for PDF conversion, pdftoppm for slide image extraction, and local Python/JS scripts for OOXML packing and HTML rendering (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill documents several external dependencies from public registries that are standard for the described tasks.
Evidence: References playwright, sharp, markitdown, and pptxgenjs (SKILL.md, html2pptx.md).

pptx