One Readout, Many Repairs: Diffusion-Guided Hierarchical Search for Tool-Agent Repair

📄 arXiv: 2609.34879v1 📥 PDF

作者: Xiang Xia, Cheng Yan, Fan Xu, Zhijun Fan, Shuyuan Zhang, Wuyang Zhang

分类: cs.AI, cs.CL

发布日期: 2026-09-28


💡 一句话要点

提出ReCommit框架以解决工具代理修复问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 工具代理 修复算法 扩散模型 分层搜索 人工智能 故障恢复 操作支持 企业服务

📋 核心要点

  1. 现有工具代理在执行调用时,尽管能够成功,但仍可能无法满足用户的请求,导致修复过程复杂且计算成本高。
  2. 本文提出ReCommit框架,通过分层搜索操作支持,重用操作类型评分,降低修复计算成本,同时提高工具代理的失败恢复能力。
  3. 在Agent-Diff基准上,ReCommit在修复预算为3和13时,分别实现了75.9%和63.2%的恢复增益,并显著减少了修复时间。

📝 摘要(中文)

工具代理通过大型语言模型与外部工具交互,但成功的调用仍可能无法满足用户请求。工具代理修复旨在寻找替代调用序列以成功执行并满足原始请求。现有方法在操作选择和具体实现上面临高昂的计算成本。为此,本文提出了一种分层搜索方法,将修复过程视为对操作支持的搜索,并引入ReCommit框架,通过单次并行读取的掩蔽扩散语言模型重用操作类型评分,从而降低修复计算成本。实验结果表明,在Agent-Diff基准上,ReCommit在修复预算为3和13时分别实现了75.9%和63.2%的相对恢复增益,同时均衡了恢复与成本的关系。

🔬 方法详解

问题定义:本文旨在解决工具代理在执行调用时无法满足用户请求的问题。现有方法在操作选择和具体实现上重复计算,导致修复过程成本高昂。

核心思路:论文通过将修复过程视为对操作支持的分层搜索,提出了ReCommit框架,旨在重用操作类型评分以降低计算成本,同时探索替代的操作实现。

技术框架:ReCommit框架包括两个主要模块:操作支持搜索和实现搜索。操作支持搜索通过掩蔽扩散语言模型的并行读取获取操作类型评分,指导后续的实现搜索。实现搜索则在每个支持内探索不同的实体绑定、参数和动作组合。

关键创新:ReCommit的创新在于其训练无关的设计,利用单次并行读取的方式重用操作评分,显著减少了修复计算的重复性,与现有方法相比,提升了效率和效果。

关键设计:在ReCommit中,操作类型评分的计算是通过掩蔽扩散语言模型实现的,具体的参数设置和损失函数设计未在摘要中详细说明,需参考论文原文获取更多技术细节。

🖼️ 关键图片

fig_0
fig_1

📊 实验亮点

实验结果显示,ReCommit在Agent-Diff基准上实现了75.9%和63.2%的相对恢复增益,同时在修复预算为3和13时,分别减少了61.3%和51.3%的平均修复时间,展现了优越的恢复与成本平衡。

🎯 应用场景

该研究的潜在应用领域包括企业服务、智能助手和自动化工具等,能够显著提升工具代理的响应能力和用户体验。未来,ReCommit框架可能在更广泛的人工智能应用中发挥重要作用,尤其是在需要高效故障恢复的场景中。

📄 摘要(原文)

Tool agents use large language models to act through external tools, yet successfully executed calls can still leave user requests unfulfilled. Tool-agent repair seeks alternative call sequences that execute successfully and fulfill the original requests. However, repair requires exploring both operation choices and their concrete realizations, making complete-sequence regeneration costly. Moreover, regeneration repeats operation selection even when failure arises from how those operations are realized. The resulting challenge is to reduce this repetition while preserving exploration of alternative operations and realizations. Therefore, we formulate repair as hierarchical search over operation supports, which we introduce as sets of permitted operation types that define reusable search regions for concrete tool-call sequences. We propose ReCommit, a training-free, diffusion-guided framework for improving tool-agent failure recovery while reducing repair computation. ReCommit amortizes operation-level proposal computation across repair trials by reusing operation-type scores from a single parallel readout of a masked diffusion language model. These scores guide search across supports, while realization search explores alternative entity bindings, arguments, and action composition within each support. Experiments on real failures across four enterprise services in the Agent-Diff benchmark show 75.9\% and 63.2\% relative recovery gains with 61.3\% and 51.3\% reductions in mean full-budget repair time at repair budgets $B=3$ and $B=13$, respectively, over the strongest evaluated 8B comparison method. ReCommit achieves a favorable recovery--cost trade-off, including in comparisons with the evaluated 32B models.