en.Wedoany.com Reported - On May 13, Microsoft released a WinUI agent plugin that provides GitHub Copilot CLI and Claude Code with complete WinUI 3 native application development capabilities, while reducing token consumption by over 70%. Through an architectural design of "on-demand skill loading + local tool retrieval," the plugin enables AI agents to drive the full development cycle—from scaffolding to application packaging—with a significantly lower token budget.
The core of the plugin consists of an agent named winui-dev, eight skills, and a set of supporting tools. The winui-dev agent loads two skills by default: winui-dev-workflow and winui-design—the former handles project scaffolding, building, running, and error diagnosis, while the latter covers XAML layout, control selection, theming, and accessibility design. The remaining six skills are loaded on demand: code review, UI testing, MSIX packaging, WPF to WinUI migration, environment setup, and session diagnostics. Microsoft explained in its official blog that each skill is a self-contained, focused operation manual, and the agent only loads relevant skills when needed, rather than pre-loading entire reference pages. This is precisely the technical foundation for the significant reduction in token consumption.
WinUI 3 is the core UI framework for Windows 11 native applications, but its development involves multiple technical layers, including XAML, Fluent Design, MVVM architecture, MSIX packaging, code signing, Store submission, and accessibility support. General-purpose AI coding agents, lacking Windows-specific context, often generate code that mixes WinUI with older frameworks like WPF and Win32, leading to degraded application performance. Microsoft noted that general-purpose agents frequently mix legacy UWP APIs into WinUI, ignore the packaging execution model, or even stop working without running verification. The design goal of the new plugin is to advance AI agents from "generating code" to "end-to-end delivery of runnable packaged applications," covering the complete workflow from dotnet new to MSIX packaging. Installation is a single command line: /plugin install winui@awesome-copilot.
The dramatic reduction in token consumption stems from three local tools bundled with the plugin. The winui-search tool indexes the sample libraries of WinUI Gallery and Community Toolkit, allowing the agent to query real control patterns before writing XAML without consuming tokens for cloud searches. The winmd-cli tool provides offline API metadata query capabilities based on .winmd and managed .dll files, enabling the agent to verify whether an API exists before generating code, thus avoiding hallucinatory calls. winui3-analyzer is a Roslyn analyzer injected at build time to identify common WinUI 3 pitfalls such as UWP namespace leaks and missing EnsureCoreWebView2Async calls. Through a three-tier mechanism of "local documentation lookup + local API verification + build-time error correction," the agent avoids loading massive reference information in every conversation round, ultimately achieving a token reduction of over 70% on the same model. The official Microsoft blog stated: "Each skill loads only what is needed and relies on tools to do the rest, and the agent does all of this with over 70% fewer tokens consumed than when we started."
The release of this WinUI agent plugin is a key part of Microsoft's internal Project K2 initiative at the developer toolchain level. Project K2 aims to rewrite core Windows 11 components using WinUI 3 to eliminate legacy framework dependencies within the system, improving responsiveness and the native experience. Microsoft engineer Beth Pan previously stated publicly that the mission is to make WinUI 3 the best UI platform for the Windows experience and native apps, with performance at the core of everything. This plugin provides external developers with the same AI coding capabilities as Microsoft's internal teams, lowering the barrier to entry for WinUI 3. Windows development has long faced the challenge of "framework fragmentation," with multiple technology paths such as WPF, WinForms, UWP, and Electron coexisting, making cross-framework migration costly. An AI agent specifically optimized for WinUI 3 can directly bring web developers and Electron developers into the native Windows application track, without requiring them to first master the entire Windows App SDK knowledge system.
The plugin is now available to GitHub Copilot CLI through the awesome-copilot registry, and Claude Code can also invoke it. Microsoft is also openly soliciting developer feedback on its GitHub repository to continuously optimize agent performance. The eight skills of the winui-dev agent cover the complete development loop from environment setup, project scaffolding, build and run, UI design, code review, and UI testing to packaging. Developers simply need to issue natural language instructions like "create a WinUI 3 photo viewer with thumbnails and EXIF metadata," and the agent can automatically complete the entire workflow of template selection, project generation, build and run, and feature verification.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
