20 upvotes
New benchmark tests AI coding agents across multiple repositories
What is CrossSWE: Benchmarking Cross-Repository Joint Development for Coding Agents?
CrossSWE is a dataset that measures how well AI coding assistants can work on tasks that involve several linked codebases. It mimics real-world development where changes in one repo affect others.
Who it is for
Researchers and tool builders can use it to gauge and improve multi-repo capabilities of their agents.
How to try it
Download the benchmark from HuggingFace: https://huggingface.co/datasets/CrossSWE