The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Python and Node.js scripts to perform its core functions. These scripts are invoked via the command line to process files and automate browser interactions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as .excalidraw board files, image files, and web content from slide decks. This represents an ingestion surface for potentially malicious instructions.
Ingestion points: Board files read by inspect_scene.py and split_scene.py; slide deck content and URLs accessed by shoot_deck.mjs.
Boundary markers: The scripts do not implement specific delimiters for the data they process, relying on standard file reading and browser automation.
Capability inventory: The skill has the capability to write files (whiteboard scenes and screenshots) and perform network requests (via Puppeteer browser automation).
Sanitization: Scripts perform integrity and structural checks (e.g., verifying aspect ratios and file hashes) but do not filter or sanitize text content for potential injection patterns.
[EXTERNAL_DOWNLOADS]: The shoot_deck.mjs tool uses the well-known puppeteer library to visit and capture images from slide deck URLs provided by the user. This involves standard browser network operations.

excalidraw-use