The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to extract and process text from external PowerPoint presentations, which creates a potential surface for indirect prompt injection attacks if an extracted document contains instructions intended to influence the agent's behavior.
Ingestion points: scripts/inventory.py extracts text from presentation shapes, and SKILL.md suggests using the markitdown command for extraction.
Boundary markers: Absent. The instructions do not explicitly mandate the use of delimiters or warnings when passing extracted text to the agent context.
Capability inventory: The skill possesses extensive capabilities including file system read/write, shell command execution (soffice, pdftoppm, git), and the ability to run a headless browser via Playwright.
Sanitization: Present. The skill utilizes the defusedxml library in ooxml/scripts/unpack.py and ooxml/scripts/pack.py to protect against XML-based attacks like XXE.
[COMMAND_EXECUTION]: Several Python scripts execute external system utilities via subprocess.run to handle document conversion and comparison.
ooxml/scripts/pack.py and scripts/thumbnail.py invoke soffice (LibreOffice) to convert presentations to HTML or PDF formats.
scripts/thumbnail.py invokes pdftoppm (Poppler) to convert PDF slides into JPEG thumbnail images.
ooxml/scripts/validation/redlining.py invokes git to perform word-diff operations for document validation.
These executions are implemented using argument lists rather than raw shell strings, which effectively mitigates the risk of command injection.
[EXTERNAL_DOWNLOADS]: The skill requires several third-party libraries and system utilities, as listed in SKILL.md (e.g., markitdown, pptxgenjs, playwright, sharp). These are standard, well-known tools for presentation processing and rendering.

pptx