The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes raw JavaScript code within a Node.js environment via the artifacts tool to build and render slides.
[EXTERNAL_DOWNLOADS]: The instructions describe fetching images from external URLs or local paths to embed them into slides, though no specific untrusted remote execution patterns were found.
[DATA_EXFILTRATION]: The skill uses node:fs/promises to read files from the disk (e.g., images for slides) and write presentation files (.pptx) and previews (.png) to the artifacts/ directory.
[PROMPT_INJECTION]: The skill processes untrusted data (user-provided text, images, and imported PPTX files) which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Data enters via user prompts for slide content and through the PresentationFile.importPptx method which loads external files.
Boundary markers: None identified in the provided instructions to delimit user-provided data from system instructions.
Capability inventory: Access to the local filesystem via node:fs/promises and the ability to execute JavaScript via the artifacts tool.
Sanitization: No explicit sanitization or validation of imported content or user text is mentioned.

slides