1 paper
Jingjing Wang, Xiwen Chen, Wenhui Zhu +6
LLM-powered coding agents spend the majority of their token budget reading repository files, yet much of the retrieved code is irrelevant to the task at hand. Existing learned prun…