Tool agents improve request fulfillment with faster repair searches
One Readout, Many Repairs: Diffusion-Guided Hierarchical Search for Tool-Agent Repair
Artificial IntelligenceComputation and Language
Summary
Sometimes, computer programs that use other tools to help perform tasks don’t complete what the user wants perfectly. The authors studied how to fix these programs without starting over each time something goes wrong. They developed a method called ReCommit that smartly guides the repair process by focusing on sets of possible actions rather than redoing everything from scratch. This approach lets the program find fixes more quickly and successfully, saving time and effort when fixing errors.
What this means in practice
- •For enterprise software developers: Improve automated tool agents that integrate multiple services by reducing repair time and increasing recovery success for failed user requests.
- •For automation platform engineers: Enhance reliability of multi-tool workflows by guiding error recovery with reusable operation groups to explore alternative fixes efficiently.
Authors
Xiang Xia, Cheng Yan, Fan Xu, Zhijun Fan, Shuyuan Zhang, Wuyang Zhang
Abstract
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.