arc-cua hands bounded desktop subtasks to a fast decision model.

Isle released arc-cua September 20 under MIT, a Python action layer where a planner emits a payload with a goal, inputs, verification conditions, constraints and a max_actions budget, and a decision model executes the UI loop and returns a status plus action count. The planner deliberately lives outside the package, so any model or even a deterministic planner can drive it. 107 stars in a day. It's the computer-use shape of the routing pattern: expensive model sets intent, cheap model closes the loop.