Terminal agents struggle to detect and fix their own errors reliably
Can Terminal Agents Trust Their Own Verification? Diagnosing and Improving Self-Verification
Computation and LanguageMachine Learning
Summary
Some AI tools solve problems by trying commands and checking their own work through a computer terminal. The paper shows that these tools often fail to notice when they make mistakes or to fix them successfully. The researchers studied this issue closely using a special test setup and found that while these tools usually try to check their answers, they miss many errors and don't fix half of the errors they find. To improve this, the authors developed a new method that teaches the tools to verify and correct better, leading to more successful problem solving.
What this means in practice
- •For software engineering teams: Improve automated scripting tools to better detect and correct their own errors during task execution in command-line environments.
- •For devops teams: Enhance reliability of terminal-based automation systems by incorporating stronger verification and error recovery methods.
Authors
Yingfeng Luo, Shaowei Wei, Daixin Wang, Dingyang Lin, Kaiyan Chang, Weiqiao Shan, Tong Zheng, Zhiqiang Zhang, Jingbo Zhu, Tong Xiao
Abstract
Terminal agents rely on self-verification to assess and correct their solutions as they solve tasks through interaction with command-line environments. Yet how trustworthy such self-verification is remains poorly understood. To investigate this question systematically, we introduce a diagnostic framework that identifies the first complete solution in each trajectory, determines whether it is objectively correct, and uses this ground truth to quantify the agent's subsequent verification and recovery behavior. Applying it to ten terminal agents on TerminalBench2.1, we find that verification is nearly universal after a complete candidate is formed, yet only 61.43\% of incorrect candidates are detected and only 49.36\% of detected errors are successfully repaired. These results show that the main weakness in self-verification lies not in initiating verification, but in detecting and repairing errors. Motivated by these findings, we propose Student-Conditioned Verification Distillation (SCVD), which lets the student first produce a candidate solution and distills a stronger teacher's subsequent verification and recovery from the same interaction context. Across three Qwen3.5 backbones, SCVD improves \textsc{Pass@1} on TerminalBench2.1 by 9.74--16.85 percentage points over the corresponding base models and by 4.49--8.61 points over the standard full-trajectory distillation, while avoiding the pronounced out-of-distribution degradation of full-trajectory distillation on SWE-bench Verified.