Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills
Jianlyu Chen, Hongjin Qian, Zheng Liu and a large BAAI-led team introduce DisCo and the AREX-Skill Library, distilling 1,000 widely used ML repositories into more than 5,000 verified reusable skills and showing that operational know-how, not the harness, is what research agents are missing.
Ask this paper
Operational knowledge as the missing layer: The claim is precise: a backbone plus a planning and memory harness still leaves domain know-how outside the agent, sitting in repos and papers written for humans and far too large to load at task time.
Two distillation modes: Task-agnostic distillation condenses the open ecosystem into the AREX-Skill Library (20 areas, 178 capability families); task-oriented distillation produces the skills a specific research task calls for.
Gains under a fixed backbone and budget: With GPT-5.5, the same harness and the same execution budget, adding skills yields 134.3 percent higher MLE-bench, 34.4 percent higher PaperBench, 9.2 percent on FrontierCS and 14.0 percent on PassNet.
The controlled comparison is the point: Because the model, harness and compute are all held fixed, the delta is attributable to distilled operating context rather than any of the usual confounds.
Why it matters: A 134 percent MLE-bench jump from context alone is the strongest published evidence yet that skill libraries, not bigger models, are the near-term lever for research agents.
Abstract
Autonomous agents are beginning to carry out machine-learning (ML) research end to end. These agents combine a model backbone with a harness for planning, execution, memory, and verification, but this architecture still leaves domain-specific know-how outside the agent. We call this missing layer operational knowledge, the know-how that separates knowing a method from making it work. That knowledge is not absent from the field. It appears in repositories and papers, but in forms written for human readers and too large to load during a task. Once distilled into compact, verified skills, this knowledge can be reused across tasks rather than rediscovered during each run. We present DisCo, a skill-powered research agent that creates skills and uses them during research. Its distillation runs in two complementary forms: task-agnostic, condensing the field's widely used repositories into reusable skills, and task-oriented, producing the skills a concrete task calls for. The former, applied across the open ecosystem, yields the AREX-Skill Library, with 5,000+ verified skills distilled from 1,000 widely used ML repositories and organized into 20 areas and 178 capability families. With the GPT-5.5 backbone, research harness, and downstream execution budget held fixed, the skill-equipped research agent scores 134.3% higher on MLE-bench, 34.4% higher on PaperBench, 9.2% higher on FrontierCS, and 14.0% higher on PassNet than the same agent without skills. These gains come from adding distilled operating context under that fixed setup.