Chrome团队与Firefox、MDN及Edge团队合作,正式发布了名为“Modern Web Guidance”的1.0版本开源项目 1。该项目旨在通过提供200多份精心策划的实现最佳实践指南,帮助AI编码代理优先使用CSS.highlights、field-sizing等浏览器原生的现代Web功能,从而避免依赖约2019年时期的过时JavaScript解决方案 1。
在技术实现上,“Modern Web Guidance”包含一个基于Agentic RAG的技能包和一个能够识别旧代码模式的“legacy-scan”工具 1。该项目的纯JS包大小约为7MB,搜索耗时仅约10毫秒 1。其中的“legacy-scan”工具基于3万GitHub仓库进行训练,平均精度达到92% 1。测试数据显示,引入该技能后,编码代理在特定任务上的通过率从无技能的55%显著提升至92%,这一结论基于对200个任务和约1700个断言的评估 1。用户可通过命令DISABLE_TELEMETRY=1 npx modern-web-guidance@latest install进行安装 1。
The Chrome team, in collaboration with Firefox, MDN, and Edge, has released version 1.0 of an open-source initiative called "Modern Web Guidance" 1. This project aims to direct artificial intelligence coding agents toward using native browser features such as CSS.highlights and field-sizing, rather than relying on outdated JavaScript solutions typical of the circa-2019 web era 1. The release includes a comprehensive library of over 200 curated guides totaling approximately 1.7 MB, all available as open source 1.
Technical Components and Performance Impact
To support these guidelines, the toolkit incorporates an Agentic RAG CLI packaged as pure JavaScript with a size of about 7 MB, capable of executing searches in roughly 10 milliseconds 1. Additionally, a tool named "legacy-scan," trained on data from 30,000 GitHub repositories, identifies old code patterns with an average precision of 92% 1. Testing involving 200 tasks and approximately 1,700 assertions demonstrated that integrating this skill set increased coding agent pass rates from 55% without the guidance to 92% with it 1. Users can install the package using the command DISABLE_TELEMETRY=1 npx modern-web-guidance@latest install 1.
评论
还没有评论,欢迎留下第一条。