Datasets:
The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Error code: FileFormatMismatchBetweenSplitsError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
WideSWE: Can Coding Agents Coordinate Changes Across Repositories?
A benchmark for implementing one shared feature or bug fix across multiple repositories.
Overview
WideSWE contains 120 real-world software-engineering tasks across 41 software ecosystems, balanced between 60 bug fixes and 60 features. Each task requires coordinated changes across multiple repositories in the same ecosystem.
Agents receive one shared request and a historical workspace containing target repositories and, where available, additional context repositories. Hidden tests evaluate the resulting changes; a task succeeds only when every target repository passes all required checks.
This dataset provides the task definitions, pinned repository revisions, environment configurations, reference patches, and hidden-test definitions. The code repository provides the evaluation harness and instructions for preparing workspaces and running agents.
Browse all tasks · View results · Run evaluation
Dataset Contents
release120.tsv is the canonical task index. Each task is stored
under cases/<ecosystem>/<case_id>/.
| File or directory | Contents |
|---|---|
prompt.md |
Shared task prompt shown to the agent. |
repos.yaml |
Participating repositories and pinned revisions. |
task.yaml, harness.yaml, validation.yaml |
Task and evaluation configuration. |
environment/ |
Dockerfiles, dependency locks, test profiles, and runtime helpers. |
gold/metadata.yaml, gold/patches/ |
Reference implementation metadata and per-repository patches. |
hidden/patches/ |
Evaluator tests applied during scoring. |
oracles/ |
Hidden-test expectations and scoring manifests. |
Repository source trees can be reconstructed from the remotes and revisions in
each repos.yaml file. Reference patches and hidden tests are evaluator-side
artifacts and must not be exposed to the agent during a benchmark run.
Download
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="wangbaoyi/WideSWE",
repo_type="dataset",
local_dir="wideswe-data",
)
For environment setup and evaluation commands, see the code README.
Citation
@misc{wang2026wideswecodingagentscoordinate,
title={WideSWE: Can Coding Agents Coordinate Changes Across Repositories?},
author={Baoyi Wang and Xingliang Wang and Jinyang Wu and Keming Wu and Chen Zhi and Jianwei Yin},
year={2026},
eprint={2609.33382},
archivePrefix={arXiv},
primaryClass={cs.SE},
url={https://arxiv.org/abs/2609.33382},
}
Licensing
Repository-derived patches and tests remain subject to the licenses of their respective upstream repositories. Benchmark metadata and harness code may have separate licensing terms in the accompanying code release. Users are responsible for complying with all applicable upstream licenses.
- Downloads last month
- 43