By leveraging inference-time scaling and a novel "reflection" mechanism, ALE-Agent solves the context-drift problems that ...