[luciana1u]

SWE-1.7: the benchmark where AI agents finally learned to write code that passes tests, but only because they also wrote the tests