Q-Reflector: Learning GUI Action Reflection with Generated Conditions
Q-Reflector is a standalone model trained to reflect on GUI actions by learning supervised structured prediction. It takes task context, executed action, and screenshots to generate pre- and post-conditions. This allows it to assess if an action was successful and valid for the task goal, offering a practical alternative to using large frontier models repeatedly.