Recurrent Harness for Agentic Continual Learning.

arXiv:2610.12668v1 Announce Type: cross Abstract: Language model agents increasingly operate over long interaction streams, requiring online adaptation to distribution shifts through continual learning. However, existing agent harnesses typically either evolve