biningbin/active-swe repository overview

⁠Active-SWE Docker Images

Official task images for Active-SWE: Benchmarking Coding Agents for Proactive Bug Fixing without Issue Reports.

Project Page⁠ | Paper⁠ | Dataset⁠ | Code⁠

⁠About

Active-SWE evaluates whether coding agents can proactively discover and fix bugs without instance-specific issue reports. The extended benchmark contains 1,663 tasks across six bug categories and eight programming languages. The curated main split contains 400 tasks.

This repository stores the public Docker images referenced by the dataset's image column. Images contain the repository snapshot, build tools, and baked dependencies needed by the evaluation harness.

⁠Pulling an image

Always use the exact image reference from the dataset row. There is intentionally no single latest image for the benchmark.

docker pull docker.io/biningbin/active-swe:simple-pylint-dev__pylint-5417

Tag format:

{set}-{instance_id}

where set is simple or hard.

To download all images referenced by an input JSONL with the official codebase:

bash scripts/download_images.sh runs/input.jsonl 4

⁠Data and evaluation

• Dataset: https://huggingface.co/datasets/XLearning-SCU/Active-SWE⁠
• Evaluation code: https://github.com/XLearning-SCU/Active-SWE⁠
• Image count in the public release: 1,663 task tags
• Architectures: use the exact platform provided by each published image

The dataset includes oracle and evaluation fields. Do not expose patch, eval_script, FAIL_TO_PASS, PASS_TO_PASS, or source-level oracle mappings to the model during generation.

⁠Citation

@article{li2026activeswe,
  title={Active-SWE: Benchmarking Coding Agents for Proactive Bug Fixing without Issue Reports},
  author={Li, Haobin and Deng, Ping and Qian, Weizhong and Jiang, Liang and Huang, Zhenyu and Yang, Mouxing and Peng, Xi},
  journal={arXiv preprint arXiv:2608.04682},
  year={2026}
}