Submitted by Xiyang Wu 7 Failure Recovery via Belief Updates and World Modeling for Long-Horizon Agents Agent Intelligence-Lab π¦ 2