SkillGym：将人类技能内化为 LLM 可复用能力的训练框架
SkillGym: Internalizing Human Skills into LLMs for Real-World Problem Solving
把人类技能攻略变成训练环境喂给模型，微调后 35B 小模型在 SkillsBench 上跑赢了 Claude Sonnet 4.6，方法思路挺有意思。
SkillGym 是一个把人类编写的智能体技能转化为可执行、可验证训练环境的框架，通过技能转任务流水线、代码检查器验证结果，并发布 2,756 个覆盖 12 个类别的环境。研究团队收集了 8,364 条成功轨迹，平均每个任务包含 49 次工具调用和超过 6 万 token 的日志。在 Claude Code 环境下，对 Qwen3.5-35B-A3B 做监督微调后，GDPval-AA v2 上提升 199 Elo，Terminal-Bench 2.1 提升 19.10 个百分点，SkillsBench v1.1 在有无技能两种设置下分别提升 28.13 和 12.38 分。其 35B 的 SkillGym-Agent 在技能辅助 SkillsBench 上达到 51.47%，超过 Claude Sonnet 4.6、GPT-5.4 Mini 和 DeepSeek V4 Pro 的已报告成绩。
SkillGym: Internalizing Human Skills into LLMs for Real-World Problem Solving
Human-written agent skills encode rich workflows for real-world problem solving, but are typically used as external inference-time instructions rather than internalized as reusable model capabilities. We introduce \texttt{SkillGym}, a framework that transforms these skills into executable, verifiable training environments for large language model agents. Its skill-to-task pipeline instantiates concrete tasks, verifies outcomes with code-based checkers, and assesses empirical skill dependence through contrastive executions. We construct and release 2,756 environments across 12 categories and collect 8,364 successful trajectories from multiple models and harnesses, averaging 49 tool calls and over 60k logged text tokens. These resources support supervised fine-tuning on verified workflows and reinforcement learning with outcome-based rewards. Under Claude Code, supervised fine-tuning improves Qwen3.5-35B-A3B by 199 Elo on GDPval-AA v2, 19.10 percentage points on Terminal-Bench 2.1, and 28.13 and 12.38 points on SkillsBench v1.1 with and without skills, respectively. Our 35B \texttt{SkillGym-Agent} reaches 51.47\% on skill-assisted SkillsBench, exceeding reported scores for Claude Sonnet 4.6, GPT-5.4 Mini, and DeepSeek V4 Pro. Without skills, it also surpasses skill-assisted bases under Codex and Claude Code, suggesting reusable procedural competence.