SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?

ArXi:2603.15401v1 Announce Type: cross Agent skills, structured procedural knowledge packages injected at inference time, are increasingly used to augment LLM agents on software engineering tasks. However, their real utility in end-to-end development settings remains unclear. We present SWE-Skills-Bench, the first requirement-driven benchmark that isolates the marginal utility of agent skills in real-world software engineering