Tag
The paper proposes SWE-Pruner Pro, which uses the coding agent's own internal representations to prune irrelevant tool-output lines, reducing token use by up to 39% without a separate pruning model.