Comment by bloppe

Generating big chunks of code is rarely what I want from an agent. They really shine for stuff like combing through logs or scanning dozens of source files to explain a test failure. Which benchmark covers that? I want the debugging benchmark that tests mastery of build systems, CLIs, etc.