desktop-use
Installation
SKILL.md
Desktop Use
One controller operates an exact GUI target, observes each effect, and stops when the user's postcondition is proved.
Act
| Goal | Command / tool |
|---|---|
| Discover | cua-driver list_apps; use launch_app when launch is requested, then select the intended window from its response or list_windows |
| Observe | get_window_state with observed pid, window_id, and the run's session; use query, max_elements, or max_depth to bound large trees |
| Act | click / type_text with an exact target and fresh element_token; use x,y only from a valid target screenshot |
| Menu / geometry | Prefer invoke_menu with an observed menu path, or set_window_frame; verify the resulting window state |
| Verify | Fresh get_window_state / get_desktop_state, or verify_state for an expressible exact-window postcondition |
| Finish | Finalize any owned recording, then end_session; leave personal apps and shared services running unless closure was requested |
Use the CLI by default or an existing MCP connection. Replace example IDs, tokens, coordinates, and RUN_ID with current observations and a unique run label: