The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the shell sandbox to execute Python scripts and shell commands (soffice, pdftoppm) for processing PowerPoint files and performing format conversions. These are standard operations for the stated purpose of document manipulation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external .pptx and .ppt files which could contain untrusted data.
Ingestion points: Reads content from .pptx files using Presentation("deck.pptx") and iterates through slides and shapes (SKILL.md).
Boundary markers: None explicitly defined in the prompts to distinguish between instructions and data during extraction.
Capability inventory: Uses exec shell to run Python, soffice for conversion, and zipfile for raw XML manipulation. It has the ability to read and write files in the workspace.
Sanitization: The skill recommends using lxml or defusedxml for XML parsing to avoid namespace issues, but does not explicitly mention protection against XML External Entity (XXE) attacks or prompt injection during text extraction. However, the risk is assessed as low as these are standard data processing tasks.

pptx