Task 1 — Browser Agent
Submit a natural-language task. The agent runs an explicit state machine — PLAN → LOCATE → ACT → VERIFY per step, with DIAGNOSE-driven recovery on failure (not try/except).
Live progress
No events yet. Submit a task above and watch the state machine run step-by-step.