DeNovoSWE: Scaling Long-Horizon Environments for Generating Entire Repositories from Scratch
Researchers have developed new frameworks to automate the creation and management of software repositories, addressing a key bottleneck in automated software engineering. One system, RepoLaunch, successfully builds and tests code across various languages and platforms with a 78% success rate. Another effort introduces DeNovoSWE, a large dataset of 4,818 instances for training code agents to generate entire repositories from documentation, significantly improving performance on complex tasks. AI
IMPACT These advancements in automated repository generation and large-scale datasets are crucial for training more capable AI agents in software development.