Research

Paper

TESTING March 02, 2026

CoVe: Training Interactive Tool-Use Agents via Constraint-Guided Verification

Authors

Jinpeng Chen, Cheng Gong, Hanbo Li, Ziru Liu, Zichen Tian, Xinyu Fu, Shi Wu, Chenyang Zhang, Wu Zhang, Suiyun Zhang, Dandan Tu, Rui Liu

Abstract

Developing multi-turn interactive tool-use agents is challenging because real-world user needs are often complex and ambiguous, yet agents must execute deterministic actions to satisfy them. To address this gap, we introduce \textbf{CoVe} (\textbf{Co}nstraint-\textbf{Ve}rification), a post-training data synthesis framework designed for training interactive tool-use agents while ensuring both data complexity and correctness. CoVe begins by defining explicit task constraints, which serve a dual role: they guide the generation of complex trajectories and act as deterministic verifiers for assessing trajectory quality. This enables the creation of high-quality training trajectories for supervised fine-tuning (SFT) and the derivation of accurate reward signals for reinforcement learning (RL). Our evaluation on the challenging $τ^2$-bench benchmark demonstrates the effectiveness of the framework. Notably, our compact \textbf{CoVe-4B} model achieves success rates of 43.0\% and 59.4\% in the Airline and Retail domains, respectively; its overall performance significantly outperforms strong baselines of similar scale and remains competitive with models up to $17\times$ its size. These results indicate that CoVe provides an effective and efficient pathway for synthesizing training data for state-of-the-art interactive tool-use agents. To support future research, we open-source our code, trained model, and the full set of 12K high-quality trajectories used for training.

Metadata

arXiv ID: 2603.01940
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-02
Fetched: 2026-03-03 04:34

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.01940v1</id>\n    <title>CoVe: Training Interactive Tool-Use Agents via Constraint-Guided Verification</title>\n    <updated>2026-03-02T14:56:35Z</updated>\n    <link href='https://arxiv.org/abs/2603.01940v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.01940v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Developing multi-turn interactive tool-use agents is challenging because real-world user needs are often complex and ambiguous, yet agents must execute deterministic actions to satisfy them. To address this gap, we introduce \\textbf{CoVe} (\\textbf{Co}nstraint-\\textbf{Ve}rification), a post-training data synthesis framework designed for training interactive tool-use agents while ensuring both data complexity and correctness. CoVe begins by defining explicit task constraints, which serve a dual role: they guide the generation of complex trajectories and act as deterministic verifiers for assessing trajectory quality. This enables the creation of high-quality training trajectories for supervised fine-tuning (SFT) and the derivation of accurate reward signals for reinforcement learning (RL). Our evaluation on the challenging $τ^2$-bench benchmark demonstrates the effectiveness of the framework. Notably, our compact \\textbf{CoVe-4B} model achieves success rates of 43.0\\% and 59.4\\% in the Airline and Retail domains, respectively; its overall performance significantly outperforms strong baselines of similar scale and remains competitive with models up to $17\\times$ its size. These results indicate that CoVe provides an effective and efficient pathway for synthesizing training data for state-of-the-art interactive tool-use agents. To support future research, we open-source our code, trained model, and the full set of 12K high-quality trajectories used for training.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-02T14:56:35Z</published>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Jinpeng Chen</name>\n    </author>\n    <author>\n      <name>Cheng Gong</name>\n    </author>\n    <author>\n      <name>Hanbo Li</name>\n    </author>\n    <author>\n      <name>Ziru Liu</name>\n    </author>\n    <author>\n      <name>Zichen Tian</name>\n    </author>\n    <author>\n      <name>Xinyu Fu</name>\n    </author>\n    <author>\n      <name>Shi Wu</name>\n    </author>\n    <author>\n      <name>Chenyang Zhang</name>\n    </author>\n    <author>\n      <name>Wu Zhang</name>\n    </author>\n    <author>\n      <name>Suiyun Zhang</name>\n    </author>\n    <author>\n      <name>Dandan Tu</name>\n    </author>\n    <author>\n      <name>Rui Liu</name>\n    </author>\n  </entry>"
}