本文通过Terminal-Lego流水线研究交互轨迹的教学效能，发现低分代理（DeepSeek-V3.2）的轨迹比高分代理（Claude Opus 4.6）更能提升学生泛化能力，归因于环境接地监督（EGS），并展示了极佳的数据效率。
摘要：
更强的代码代理通常被认为是训练后阶段的更优教师，然而这一假设尚未与任务难度、框架设计和学生能力充分解耦。我们使用Terminal-Lego（一个可扩展的流水线，将多领域现实问题转化为环境验证的代理任务）来研究这种教学联系。令人惊讶的是，独立表现并不能决定教学效能：尽管Claude Opus 4.6在Terminal-Bench 2.0上获得更高分数，但使用来自较低分代理DeepSeek-V3.2的轨迹进行微调的学生表现出显著更强的泛化能力。我们将这种“教学悖论”归因于环境接地监督（EGS）：通过框架可见交互明确暴露“检查-行动-验证”行为的轨迹，使学生能够内化稳健的问题解决程序，而非脆弱的动作序列。扩展分析揭示了卓越的数据效率：例如，仅使用15.3k条Terminal-Lego轨迹，Qwen3-32B在Terminal-Bench 2.0上获得了24.3%的分数，与之前使用超过30倍数据量达到的最优性能相当。我们的结果表明，代理训练后的前沿不仅限于结果匹配，而是将焦点转向“框架工程”，其中环境接地交互结构的系统设计成为可复现和可泛化的代理智能的主要催化剂。
英文摘要：
Stronger code agents are commonly assumed to be superior teachers for post-training, yet this assumption remains poorly disentangled from task difficulty, harness design, and student capacity. We investigate this pedagogical link using Terminal-Lego, a scalable pipeline that transforms multi-domain real-world issues into environment-verified agentic tasks. Surprisingly, standalone performance does not dictate teaching efficacy: while Claude Opus 4.6 achieves higher scores on Terminal-Bench 2.0, students fine-tuned on trajectories from DeepSeek-V3.2, a lower-scoring agent, exhibit significantly stronger generalization. We attribute this "pedagogical paradox" to Environment-Grounded Supervision (EGS): trajectories that explicitly expose inspect-act-verify behaviors through harness-visible interactions allow students to internalize robust problem-solving routines rather than fragile action sequences. Scaling analysis reveals exceptional data efficiency: with only 15.3k Terminal-Lego trajectories, for example, Qwen3-32B achieves a 24.3% score on Terminal-Bench 2.0, rivaling previous SOTA performance established with over 30x the data volume. Our results suggest that the frontier of agent post-training lies beyond mere outcome-matching, shifting the focus toward "Harness Engineering", where the systematic design of environment-grounded interaction structures serves as the primary catalyst for reproducible and generalizable agentic intelligence.
T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks
Tencent Hy Foundation Model Frontier（腾讯Hy基础模型前沿）；National University of Singapore（新加坡国立大学）；University of Georgia（佐治亚大学）；Indiana University（印第安纳大学）；University of Maryland, College Park（马里兰大学帕克分校）
Scaling Trajectories for Complex Tasks through Recursive Self-Rewrite
Tencent HY LLM Frontier（腾讯HY大语言模型前沿）；University of Maryland, College Park（马里兰大学帕克分校）；University of Georgia（佐治亚大学）；National University of Singapore（新加坡国立大学）；Indiana University（印第安纳大学）；Washington University in St. Louis（圣路易斯华盛顿大学）；Nanyang Technological University（南洋理工大学）
Terminal-World: Scaling Terminal-Agent Environments via Agent Skills
School of Computer Science and Engineering, Beihang University, Beijing, China（北京航空航天大学计算机科学与工程学院）；Independent Researcher（独立研究者）；Beijing Institute of Technology（北京理工大学）；University of Edinburgh（爱丁堡大学）
Recursive Synthesis for Long-Horizon Terminal Tasks
Tencent HY LLM Frontier（腾讯HY大模型前沿团队）；University of Georgia（佐治亚大学）；University of Maryland, College Park（马里兰大学帕克分校）；University of Pennsylvania（宾夕法尼亚大学）；University of Minnesota, Twin Cities（明尼苏达大学双城分校）；Indiana University（印第安纳大学）；National University of Singapore（新加坡国立大学）；Hong Kong Polytechnic University（香港理工大学）
TACD: Distilling Efficient Text-to-Motion Models via Terminal Amplification Control
Sun Yat-sen University（中山大学）；Nanyang Technological University（南洋理工大学）；Wuhan University（武汉大学）；University of Science and Technology of China（中国科学技术大学）；The Hong Kong University of Science and Technology（香港科技大学）
Self-Supervised Scaling of Terminal Environments for Scientific Domains
Tencent HY LLM Frontier（腾讯HY大模型前沿团队）；University of Georgia（佐治亚大学）；University of Maryland, College Park（马里兰大学帕克分校）；National University of Singapore（新加坡国立大学）；Indiana University（印第安纳大学）；University of Illinois at Chicago（伊利诺伊大学芝加哥分校）；Hong Kong Polytechnic University（香港理工大学）
