跳到正文
arXiv cs.CL· Haoran Ye, Yuxing Lu, Haonan Dong, Zhaochen Su, Guojie Song·· 10 小时前AI 评分48

Harness-Zero:通过 Agent-as-Harness 实现智能体框架蒸馏

Harness-Zero: Harness Distillation via Agent-as-Harness

AI 导读

Harness-Zero 提出“Agent-as-Harness”方法,将领域或实例优化的智能体框架作为训练时引导,把其行为蒸馏进模型权重,部署时可移除专用框架。

正文 · 原文

View PDF HTML (experimental)

Abstract:Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized ones. We therefore study agent harness distillation: using a domain- or instance-optimized harness as training-time guidance and transferring the behaviors it induces into model weights, so that its gains survive under a single fixed target harness. The challenge is that the two harnesses differ in action space and available information, so guidance from the optimized harness cannot serve directly as supervision for the target one. We introduce Harness-Zero, which enables harness distillation through agent-as-harness. Guided by the optimized harness, a harnessing agent corrects student responses before execution in the target harness's action space, turning harness guidance into training demonstrations. Fine-tuning on the resulting trajectories internalizes harness-induced behavior into the model, so the specialized harness can be removed at deployment. Our experiments spanning knowledge work, tool use, and science domains show that: (1) For frontier LLMs using the same evolved harness, agent-as-harness outperforms code-as-harness. (2) With the specialized harness removed at deployment, Harness-Zero improves the base model's macro-average task success from 23.3% to 44.3%, even exceeding the 41.7% it reaches with that harness still attached, while reducing inference token usage by 27%-86%. (3) Harness-Zero recovers harness-induced behaviors absent from the base model, with 82.3% average recovery across 28 patterns in the three domains.
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Neural and Evolutionary Computing (cs.NE)
Cite as: arXiv:2609.24974 [cs.AI]
  (or arXiv:2609.24974v2 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2609.24974

arXiv-issued DOI via DataCite

Submission history

From: Haoran Ye [view email]
[v1] Mon, 21 Sep 2026 17:55:20 UTC (413 KB)
[v2] Wed, 18 Nov 2026 19:20:46 UTC (424 KB)

来源:arXiv cs.CL · arxiv.org