The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and extracts text from external PowerPoint files, creating a surface for indirect prompt injection if those files contain instructions designed to manipulate the agent.
Ingestion points: SKILL.md (Reading Content section) and editing.md document workflows for extracting slide text using markitdown.
Boundary markers: Not present; the extracted text is not explicitly delimited to warn the agent about potential embedded instructions.
Capability inventory: The skill can write to the filesystem (scripts/add_slide.py), execute shell commands (soffice), and fetch remote images (pptxgenjs.md).
Sanitization: Most XML manipulation utilizes defusedxml to mitigate XXE vulnerabilities, although some helper scripts use standard ElementTree for author inference.
[EXTERNAL_DOWNLOADS]: The documentation includes functionality to fetch images from remote URLs when generating new presentations.
Evidence: pptxgenjs.md provides examples for including images via https:// URLs.
[COMMAND_EXECUTION]: The skill workflow relies on executing several command-line utilities and local Python scripts.
Evidence: SKILL.md and editing.md integrate the use of soffice, pdftoppm, and multiple internal scripts like unpack.py and pack.py.

powerpoint