DOCKSMITH: Scaling Reliable Coding Environments via an Agentic Docker Builder

Jiaran Zhang, Lu Ma, Yanhao Li, Fanqi Wan, Di Qi, Xin Wu, Zhewei Huang, Liangyu Chen, Yingwei Ma, Qi Han, Xiangyu Zhang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:157646-157665, 2026.

Abstract

Reliable Docker-based environment construction is a dominant bottleneck for scaling execution-grounded training and evaluation of software engineering agents. We introduce DockSmith, a specialized agentic Docker builder designed to address this challenge. DockSmith treats environment construction not merely as a preprocessing step, but as a core agentic capability that exercises long-horizon tool use, dependency reasoning, and failure recovery, yielding supervision that transfers beyond Docker building itself. DockSmith is trained on large-scale, execution-grounded Docker-building trajectories produced by a SWE-Factory–style pipeline augmented with a loop-detection controller and a cross-task success memory. Training a 30B-A3B model on these trajectories achieves open-source state-of-the-art performance on Multi-Docker-Eval, with 39.72% Fail-to-Pass and 58.28% Commit Rate. Moreover, DockSmith improves out-of-distribution performance on SWE-bench Verified, SWE-bench Multilingual, and Terminal-Bench 2.0, demonstrating the broader agentic benefits of environment construction. Our model and Docker-building trajectories are publicly available at https://huggingface.co/JiaranZhang/DockSmith.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-zhang26ef, title = {{DOCKSMITH}: Scaling Reliable Coding Environments via an Agentic Docker Builder}, author = {Zhang, Jiaran and Ma, Lu and Li, Yanhao and Wan, Fanqi and Qi, Di and Wu, Xin and Huang, Zhewei and Chen, Liangyu and Ma, Yingwei and Han, Qi and Zhang, Xiangyu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {157646--157665}, 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/zhang26ef/zhang26ef.pdf}, url = {https://proceedings.mlr.press/v306/zhang26ef.html}, abstract = {Reliable Docker-based environment construction is a dominant bottleneck for scaling execution-grounded training and evaluation of software engineering agents. We introduce DockSmith, a specialized agentic Docker builder designed to address this challenge. DockSmith treats environment construction not merely as a preprocessing step, but as a core agentic capability that exercises long-horizon tool use, dependency reasoning, and failure recovery, yielding supervision that transfers beyond Docker building itself. DockSmith is trained on large-scale, execution-grounded Docker-building trajectories produced by a SWE-Factory–style pipeline augmented with a loop-detection controller and a cross-task success memory. Training a 30B-A3B model on these trajectories achieves open-source state-of-the-art performance on Multi-Docker-Eval, with 39.72% Fail-to-Pass and 58.28% Commit Rate. Moreover, DockSmith improves out-of-distribution performance on SWE-bench Verified, SWE-bench Multilingual, and Terminal-Bench 2.0, demonstrating the broader agentic benefits of environment construction. Our model and Docker-building trajectories are publicly available at https://huggingface.co/JiaranZhang/DockSmith.} }
Endnote
%0 Conference Paper %T DOCKSMITH: Scaling Reliable Coding Environments via an Agentic Docker Builder %A Jiaran Zhang %A Lu Ma %A Yanhao Li %A Fanqi Wan %A Di Qi %A Xin Wu %A Zhewei Huang %A Liangyu Chen %A Yingwei Ma %A Qi Han %A Xiangyu Zhang %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-zhang26ef %I PMLR %P 157646--157665 %U https://proceedings.mlr.press/v306/zhang26ef.html %V 306 %X Reliable Docker-based environment construction is a dominant bottleneck for scaling execution-grounded training and evaluation of software engineering agents. We introduce DockSmith, a specialized agentic Docker builder designed to address this challenge. DockSmith treats environment construction not merely as a preprocessing step, but as a core agentic capability that exercises long-horizon tool use, dependency reasoning, and failure recovery, yielding supervision that transfers beyond Docker building itself. DockSmith is trained on large-scale, execution-grounded Docker-building trajectories produced by a SWE-Factory–style pipeline augmented with a loop-detection controller and a cross-task success memory. Training a 30B-A3B model on these trajectories achieves open-source state-of-the-art performance on Multi-Docker-Eval, with 39.72% Fail-to-Pass and 58.28% Commit Rate. Moreover, DockSmith improves out-of-distribution performance on SWE-bench Verified, SWE-bench Multilingual, and Terminal-Bench 2.0, demonstrating the broader agentic benefits of environment construction. Our model and Docker-building trajectories are publicly available at https://huggingface.co/JiaranZhang/DockSmith.
APA
Zhang, J., Ma, L., Li, Y., Wan, F., Qi, D., Wu, X., Huang, Z., Chen, L., Ma, Y., Han, Q. & Zhang, X.. (2026). DOCKSMITH: Scaling Reliable Coding Environments via an Agentic Docker Builder. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:157646-157665 Available from https://proceedings.mlr.press/v306/zhang26ef.html.

Related Material