microsoft/win-dev-skills

microsoft/win-dev-skills

Description: Agents and skills for building Windows apps with WinUI 3 and the Windows App SDK

Language: C#

License: MIT

Stars: 300

Forks: 13

Open issues: 25

Created: 2026-03-03T17:47:08Z

Pushed: 2026-06-15T06:32:56Z

Default branch: main

Fork: no

Archived: no

README:

WinUI agents and skills for Windows app development

A GitHub Copilot, Claude Code, and OpenAI Codex plugin for building native Windows apps with WinUI 3 and the Windows App SDK to cover the end-to-end inner loop: scaffold → design → build → run → test → package → ship.

> [!WARNING] > 🚧 Preview · v0.x — expect breaking changes. Skill names, on-disk layout, agent configuration, analyzer rule IDs, and CLI tool surfaces are all subject to change without notice. There is no SemVer commitment until v1.0. If you need a stable pin, install from a release tag instead of the rolling marketplace (see [Pinning to a release](#pinning-to-a-release)). Outputs are suggestions, not authoritative answers — review them before committing or shipping anything they produce.

Install

The plugin requires GitHub Copilot (winget install GitHub.Copilot), Claude Code, or OpenAI Codex installed.

Git (winget install Git.Git) is required for installing pluggins.

Option A — Just ask Copilot to do it

Paste this prompt into a Copilot CLI session. It installs the plugin and sets up every prerequisite in one shot:

Install the Copilot CLI plugin "winui" from microsoft/win-dev-skills, then set up my machine for WinUI 3 development. Specifically:

1. Run: copilot plugin marketplace add microsoft/win-dev-skills
2. Run: copilot plugin install winui@win-dev-skills
3. Make sure these prerequisites are present (check first; .NET only installs if missing, but WinApp CLI and templates always upgrade to latest):
- .NET SDK >= 8.0 (run `dotnet --list-sdks`; if none >= 8.0, `winget install --id Microsoft.DotNet.SDK.10 --exact --silent --accept-package-agreements --accept-source-agreements`)
- WinApp CLI: must be >= 0.3 (parse `winapp --version`); if missing or older, `winget install --id Microsoft.WinAppCLI`. THEN regardless run `winget upgrade --id Microsoft.WinAppCLI` to get the latest.
- WinUI 3 templates: always run `dotnet new install Microsoft.WindowsAppSDK.WinUI.CSharp.Templates` (after refreshing $env:Path) — it upgrades in place if already installed.
- Developer Mode (DWORD HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\AppModelUnlock\AllowDevelopmentWithoutDevLicense == 1) — ASK ME first before triggering UAC; if I decline, just print the elevated command for me to run later.
4. Print a short summary of what was installed vs already present, then tell me to start a new Copilot CLI, activate the "winui-dev" agent, and to ask it to build an app.

Option B — Install the plugin yourself, then ask the agent to set up the rest

If you'd rather run the plugin commands by hand:

GitHub Copilot CLI

The plugin is listed on awesome-copilot and can be installed directly:

copilot plugin install winui@awesome-copilot

Or add this repo as a marketplace and install from there:

copilot plugin marketplace add microsoft/win-dev-skills
copilot plugin install winui@win-dev-skills

Claude Code

claude plugin marketplace add microsoft/win-dev-skills
claude plugin install winui@win-dev-skills

OpenAI Codex

Add the microsoft/win-dev-skills marketplace, then enable the winui plugin from the plugin directory.

> Note: Codex doesn't have an "agents" concept, so the orchestrator agent isn't exposed there. The skills still work - invoke them by name (e.g. /winui-setup, /winui-design) and Codex will load them on demand.

Then start a new session and run the winui-setup skill with /winui-setup.

Once setup is done, try a real task:

> "Build me a WinUI 3 markdown editor with live preview and a custom title bar"

What gets installed

| Tool | Minimum | Recommended | Install command | |---|---|---|---| | Git | 2.54 | 2.54+ | winget install Git.Git | | .NET SDK | 8.0 | 10.0 | winget install Microsoft.DotNet.SDK.10 | | WinApp CLI | 0.3 | latest | winget install Microsoft.WinAppCLI | | WinUI 3 templates | — | latest | dotnet new install Microsoft.WindowsAppSDK.WinUI.CSharp.Templates | | Developer Mode | enabled | enabled | DWORD AllowDevelopmentWithoutDevLicense set to 1 |

Visual Studio with the WinUI workload is optional but recommended for the XAML-diagnostics workaround called out in the next section. Neither setup path will install it; if you want it, run:

winget install Microsoft.VisualStudio.Community --override "--add Microsoft.VisualStudio.Workload.Universal"

> [!IMPORTANT] > For best results, install Visual Studio with the WinUI workload. There is a known XAML-compiler issue under dotnet build where a malformed XAML file produces no useful diagnostic — the build just fails with no indication of what's wrong in which .xaml. Agents that hit this thrash through unrelated guesses. The winui-dev-workflow skill works around it via a small [BuildAndRun.ps1](plugins/winui/skills/winui-dev-workflow/BuildAndRun.ps1) helper that prefers MSBuild when it's available on the machine and falls back to dotnet build otherwise. This workaround is explicitly temporary and goes away when the next Windows App SDK release fixes the underlying compiler.

Why a Copilot CLI plugin?

Modern Windows app development covers a lot of ground — XAML and Fluent Design, MVVM, MSIX packaging, code signing, Store submission, accessibility, theming, UI automation. AI agents working from generic web context tend to mix WinUI 3 with older stacks (UWP, WPF), miss the packaged-execution model, or stop short of running and verifying what they built. These skills stay in the WinUI 3 lane end-to-end — from dotnet new through a signed MSIX — and pair the agent with tools that give it real, ground-truth answers instead of guesses.

The result: you ask copilot -p "create a WinUI 3 photo viewer with thumbnails and EXIF metadata", and the agent picks the right template, scaffolds the project, designs the XAML with theming and accessibility in mind, wires up MVVM correctly, builds, fixes errors with real diagnostics, launches through the right packaged-execution pipeline, and (if you ask) drives the running app through UI automation to validate it works.

What's in this repo...