The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell scripts (scripts/tq and scripts/new-canvas) to communicate with the tldraw application and perform file operations. These scripts use standard system tools like curl, python3, and open to interact with the local environment and are restricted to the user's ~/canvas directory.
[REMOTE_CODE_EXECUTION]: The skill enables the execution of JavaScript within the local tldraw application via an HTTP bridge. This functionality is used exclusively to translate the agent's diagramming instructions into visual elements on the canvas.
[PROMPT_INJECTION]: The skill processes user-supplied text to generate diagram models. While this involves ingesting untrusted input, the skill's capabilities are limited to visual rendering within the sandboxed tldraw environment, posing no threat of privilege escalation or data exfiltration.

architecture-diagram