See and interact with the macOS desktop through a vision-action loop. Captures screenshots, analyzes them with Claude vision, and executes mouse clicks, keyboard input, and app navigation via cliclick and AppleScript. Use when the user asks to interact with GUI applications, click buttons, open or navigate native macOS apps (Figma, Finder, System Settings, etc.), export files from visual interfaces, fill forms in desktop apps, or perform any task requiring seeing and acting on the screen. Triggers: 'look at my screen', 'click on X', 'open app Y and do Z', 'export from Figma', 'navigate to settings', or any multi-step desktop GUI workflow.