SWE-Cycle: Benchmarking Code Agents across the Complete Issue Resolution Cycle

As autonomous code agents move toward end-to-end software development, evaluating their practical autonomy becomes critical. Current benchmarks hide friction by testing agents in pre-configured environments, and their static evaluation pipelines frequently fail when parsing fully autonomous trajecto...