一句话：以轨迹级分析刻画终端智能体代码评审的行为模式、隐藏成本与人类对齐情况。
基于终端的智能体代码评审可在本地开发阶段、创建 pull request 之前提供早期反馈。然而现有评估多以性能指标为中心，未能刻画扎根于代码仓库的智能体评审员的动态行为，因而难以理解其成功、失败以及在实践中产生的隐藏运行成本。本文转而以评审智能体的执行轨迹为分析对象，对其行为进行轨迹级研究。结果显示：智能体评审员虽然取得更高的评审精度，但付出了可观的探索与验证开销；成功的评审往往与更强的规划行为以及更少的下游验证相关联。作者据此指出，未来智能体代码评审系统有必要采用轨迹感知与成本敏感的评价方式。
Agentic code review in terminal-based environments enables early feedback during local development before pull request creation. However, existing evaluations remain performance-centric and fail to capture the dynamic behaviors of repository-grounded agentic reviewers. Understanding these behaviors is critical for identifying how agentic reviewers succeed, fail, and incur hidden operational costs in practice. Then, we analyze the reviewers’ behavior based on their trajectories. Our results show that agentic reviewers achieve higher review precision but incur substantial exploration and validation overhead, while successful reviews are associated with stronger planning and less downstream validation. These findings highlight the potential benefits of trajectory-aware and cost-sensitive evaluation of future agentic code review systems.
@inproceedings{abs-2607-16740,
title = {Agentic Code Review in the Terminal: A Trajectory-Level Analysis of Behavior, Cost, and Human-Alignment},
author = {Wachiraphan Charoenwet and Kla Tantithamthavorn and Patanamon Thongtanunam and Hong Yi Lin and Minwoo Jeong and Ming Wu},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834546}
}