The Agent Skills Directory

[SAFE]: No malicious code, obfuscation, or hardcoded credentials were detected in the skill files or scripts. No sensitive local file access was observed.
[COMMAND_EXECUTION]: The skill utilizes Node.js scripts (located in scripts/*.cjs) to interact with a canvas server via HTTP. These scripts facilitate management tasks such as clearing the canvas, creating, updating, and deleting elements, as well as exporting and importing diagram data.
[DATA_EXFILTRATION]: The tool export_to_excalidraw_url transmits canvas data to the well-known service excalidraw.com to generate shareable links. This behavior is documented and consistent with the skill's purpose.
[PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it ingests and processes untrusted text data from the Excalidraw canvas.
Ingestion points: Tools such as describe_scene, query_elements, and get_element read element metadata and text labels from the canvas server (defaulting to localhost).
Boundary markers: The instructions do not implement specific delimiters or "ignore instructions" warnings when processing text retrieved from diagram labels.
Capability inventory: The agent has the ability to execute shell commands (via the provided scripts), perform file system writes (via export-elements.cjs), and conduct network operations via HTTP.
Sanitization: No validation or sanitization is performed on the element text retrieved from the canvas server before it is returned to the agent context.

excalidraw-skill