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
有人把 2756 个智能体技能环境做成了训练集，微调后的 Qwen 35B 跑赢了好几个闭源旗舰模型，做 Agent 训练的可以看看。
SkillGym 框架把人类编写的智能体技能转化为可执行、可验证的训练环境，用代码检查器验证结果，并通过对比执行评估技能依赖度。团队构建并发布了 2,756 个覆盖 12 个类别的环境，收集了 8,364 条成功轨迹，平均每条包含 49 次工具调用和超过 6 万 token 的日志。在 Claude Code 下，监督微调让 Qwen3.5-35B-A3B 在 GDPval-AA v2 上提升 199 Elo，在 Terminal-Bench 2.1 上提升 19.10 个百分点。其 35B 的 SkillGym-Agent 在带技能的 SkillsBench 上达到 51.47%，超过已报告的 Claude Sonnet 4.6、GPT-5.4 Mini 和 DeepSeek V4 Pro 的成绩。
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.
