The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell scripts to perform its tasks, including scripts/tq and scripts/new-canvas.
Evidence: These scripts invoke curl for API communication, python3 -c for parsing JSON configuration from the local filesystem, and open -a (on macOS) to launch the tldraw application.
[DATA_EXPOSURE]: The skill reads a local configuration file to obtain authentication credentials for the tldraw offline app.
Evidence: scripts/tq reads server.json from standard application support directories to retrieve a per-session token and port. This information is used exclusively for authorized communication with the local application via localhost.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential vulnerability surface because it processes external canvas data during the export process.
Ingestion points: Data retrieved via editor.toImageDataUrl and other Editor API calls.
Boundary markers: The skill instructions specify clear workflows but do not explicitly define delimiters for canvas content.
Capability inventory: The skill has the capability to write files to the local filesystem and execute shell commands via scripts.
Sanitization: The skill treats the canvas export as raw image/vector data to be written to disk, which limits the potential for instruction injection to affect the agent's logic.

canvas-export