1 paper
Mingxuan Wang, Hongyue Chen, Yinglong Guo +6
Long-horizon agents continuously accumulate interaction history during task execution, yet the importance of past interactions changes as the agent state evolves. Existing context…