Repo-To-Skill：将 GitHub 代码仓库蒸馏为 AI4AI 技能

Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills

September 2, 2026

作者: Jianlyu Chen, Yuyang Hu, Hongjin Qian, Jiawei Liu, Wenqing Wei, Xiaolong Chen, Defu Lian, Zhicheng Dou, Chaozhuo Li, Qiwei Ye, Zheng Liu

cs.AI

摘要

自主智能体正开始端到端地开展机器学习（ML）研究。这类智能体将模型骨干与用于规划、执行、记忆和验证的框架相结合，但这一架构仍未将领域特有的专门知识纳入智能体内部。我们将这一缺失层称为操作知识——即区分“了解一种方法”与“让方法真正奏效”的专门诀窍。这类知识在领域中并不匮乏，它存在于代码库和论文中，但其形式是为人类读者编写的，且体量过大，无法在任务执行期间载入。一旦被蒸馏为紧凑且经过验证的技能，这类知识便可在不同任务间复用，而不必在每次运行中重新摸索。 我们提出 DisCo——一个技能驱动的研究智能体，它创建技能并在研究过程中加以使用。其蒸馏过程以两种互补的形式进行：任务无关蒸馏将该领域广泛使用的代码库凝练为可复用技能，任务导向蒸馏则产出具体任务所需的技能。前者应用于整个开放生态，产出了 AREX 技能库（AREX-Skill Library），其中包含从 1,000 个广泛使用的 ML 代码库中蒸馏出的 5,000 余项经过验证的技能，划分为 20 个领域和 178 个能力族。在 GPT-5.5 骨干、研究框架与下游执行预算均保持固定的条件下，配备技能的研究智能体在 MLE-bench 上的得分高出 134.3%，在 PaperBench 上高出 34.4%，在 FrontierCS 上高出 9.2%，在 PassNet 上高出 14.0%，均优于不配备技能的同一智能体。这些提升源于在上述固定设定下额外加入的经蒸馏的操作上下文。

English

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.