Cognition公司推出了Code Scans功能，这是一项由Devin AI驱动的工具，可将工程目标转化为代码库级别的具体改进1。该功能通过Devin webapp中的/scan命令启动1。
Code Scans采用Agentic MapReduce架构设计，包含Plan、Shard、Map、Reduce四个阶段1，通过分布式AI agents并行调查代码库并生成拉取请求1。在实测中，该功能展现出显著效果：Dioxus仓库的Rust编译时间从58.6秒降至21.0秒，降幅达64%1；devin.ai网站的Ahrefs健康评分从87提升至921，慢页面减少73%1；SEO扫描在两个仓库中发现了44个问题1。
Cognition has introduced Code Scans, a new capability powered by its Devin AI platform that transforms engineering objectives into concrete, codebase-level improvements 1. The feature operates through an Agentic MapReduce architecture, which orchestrates distributed AI agents to investigate codebases in parallel and generate pull requests 1.
The system employs a four-stage workflow comprising Plan, Shard, Map, and Reduce phases 1. Users can initiate Code Scans via the /scan command within the Devin webapp 1. Testing demonstrates significant performance gains: the Dioxus repository experienced a 64% reduction in Rust compilation time, dropping from 58.6 seconds to 21.0 seconds 1. Beyond performance metrics, the feature has delivered measurable SEO improvements—devin.ai's Ahrefs health score increased from 87 to 92, while slow pages on the site decreased by 73% 1. In aggregate SEO scanning across two repositories, the tool identified 44 distinct issues 1.
评论
还没有评论，欢迎留下第一条。