Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments

Siwei Wu, Yizhi Li, Yuyang Song, Wei Zhang, Yang Wang, Riza Batista-Navarro, Xian Yang, Mingjie Tang, Bryan Dai, Jian Yang, Chenghua Lin
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:136233-136282, 2026.

Abstract

Training agentic models for terminal-based tasks critically depends on high-quality terminal trajectories that capture realistic long-horizon interactions across diverse domains. However, constructing such data at scale remains challenging due to two key requirements: Executability, since each instance requires a suitable and often distinct Docker environment; and Verifiability, because heterogeneous task outputs preclude unified, standardized verification. To address these challenges, we propose TerminalTraj, a scalable pipeline that (i) filters high-quality repositories to construct Dockerized execution environments, (ii) generates Docker-aligned task instances, and (iii) synthesizes agent trajectories with executable validation code. Using TerminalTraj, we curate 32K Docker images and generate 50,733 verified terminal trajectories across eight domains. Models trained on this data with the Qwen2.5-Coder backbone achieve consistent performance improvements on TerminalBench (TB), with gains of up to 20% on TB 1.0 and 10% on TB 2.0 over their respective backbones. Notably, TerminalTraj-32B achieves strong performance among models with fewer than 100B parameters, reaching 35.30% on TB 1.0 and 22.00% on TB 2.0, and demonstrates improved test-time scaling behavior.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wu26ad, title = {Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments}, author = {Wu, Siwei and Li, Yizhi and Song, Yuyang and Zhang, Wei and Wang, Yang and Batista-Navarro, Riza and Yang, Xian and Tang, Mingjie and Dai, Bryan and Yang, Jian and Lin, Chenghua}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {136233--136282}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wu26ad/wu26ad.pdf}, url = {https://proceedings.mlr.press/v306/wu26ad.html}, abstract = {Training agentic models for terminal-based tasks critically depends on high-quality terminal trajectories that capture realistic long-horizon interactions across diverse domains. However, constructing such data at scale remains challenging due to two key requirements: Executability, since each instance requires a suitable and often distinct Docker environment; and Verifiability, because heterogeneous task outputs preclude unified, standardized verification. To address these challenges, we propose TerminalTraj, a scalable pipeline that (i) filters high-quality repositories to construct Dockerized execution environments, (ii) generates Docker-aligned task instances, and (iii) synthesizes agent trajectories with executable validation code. Using TerminalTraj, we curate 32K Docker images and generate 50,733 verified terminal trajectories across eight domains. Models trained on this data with the Qwen2.5-Coder backbone achieve consistent performance improvements on TerminalBench (TB), with gains of up to 20% on TB 1.0 and 10% on TB 2.0 over their respective backbones. Notably, TerminalTraj-32B achieves strong performance among models with fewer than 100B parameters, reaching 35.30% on TB 1.0 and 22.00% on TB 2.0, and demonstrates improved test-time scaling behavior.} }
Endnote
%0 Conference Paper %T Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments %A Siwei Wu %A Yizhi Li %A Yuyang Song %A Wei Zhang %A Yang Wang %A Riza Batista-Navarro %A Xian Yang %A Mingjie Tang %A Bryan Dai %A Jian Yang %A Chenghua Lin %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wu26ad %I PMLR %P 136233--136282 %U https://proceedings.mlr.press/v306/wu26ad.html %V 306 %X Training agentic models for terminal-based tasks critically depends on high-quality terminal trajectories that capture realistic long-horizon interactions across diverse domains. However, constructing such data at scale remains challenging due to two key requirements: Executability, since each instance requires a suitable and often distinct Docker environment; and Verifiability, because heterogeneous task outputs preclude unified, standardized verification. To address these challenges, we propose TerminalTraj, a scalable pipeline that (i) filters high-quality repositories to construct Dockerized execution environments, (ii) generates Docker-aligned task instances, and (iii) synthesizes agent trajectories with executable validation code. Using TerminalTraj, we curate 32K Docker images and generate 50,733 verified terminal trajectories across eight domains. Models trained on this data with the Qwen2.5-Coder backbone achieve consistent performance improvements on TerminalBench (TB), with gains of up to 20% on TB 1.0 and 10% on TB 2.0 over their respective backbones. Notably, TerminalTraj-32B achieves strong performance among models with fewer than 100B parameters, reaching 35.30% on TB 1.0 and 22.00% on TB 2.0, and demonstrates improved test-time scaling behavior.
APA
Wu, S., Li, Y., Song, Y., Zhang, W., Wang, Y., Batista-Navarro, R., Yang, X., Tang, M., Dai, B., Yang, J. & Lin, C.. (2026). Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:136233-136282 Available from https://proceedings.mlr.press/v306/wu26ad.html.

Related Material