SWE-Review turns one-shot pull-request generation into a generate-review-revise loop. Given an issue and an AI-generated PR, a reviewer agent explores the repository, decides whether the change should be accepted, and produces structured revision feedback. The authors introduce SWE-Review-Bench for evaluating review correctness and downstream resolution, plus SWE-Review-Traj for studying agentic review and reviewer training. According to the abstract, the approach improves PRs iteratively, beats single-turn fixed-context review in decision accuracy and post-revision resolve rate, transfers to issue-resolution models, and supports effective test-time scaling.
No heat snapshots are available in the last 24 hours.