1 paper
Yifan Wu, Zhuokai Zhao, Songlin Li +8
Most coding-agent benchmarks are static: an agent receives a complete task description up front and is judged only by its final code. Real coding assistance is interactive, with us…