execution-infrastructure

HarnessDev evaluates LLMs by their ability to build and evolve execution harnesses, revealing significant variations in performance and poor transferability across models.