StateAct is a code-first, multi-agent harness for long-horizon computer-use tasks. Its main agent operates on program state, including files, application backends, and the DOM, while a GUI subagent handles only screenshot-and-click subtasks. An independent finish gate verifies structural outcomes such as missing, unsaved, or misplaced files, and fresh subagents help contain context growth. On OSWorld 2.0, Claude Opus 4.8 improves from 20.6% to 26.9% binary success and from 54.8% to 61.6% partial success, with roughly 9x lower task cost than screenshot-only control.
No heat snapshots are available in the last 24 hours.