Microsoft’s Official .NET Skills Library Empowers AI Coding Agents

Background and Motivation

AI coding agents (Claude Code, Cursor, Codex) can generate code in many languages, but they lack deep .NET knowledge. They often miss .NET‑specific details such as MSBuild error codes, EF Core migration commands, xUnit version differences, and .NET 11 API changes, leading to outdated or inaccurate suggestions.

Agent Skills Standard

Agent Skills is a 2025 plugin standard that packages domain‑specific knowledge as a directory plus a descriptor file. Skills are read by agents as a combination of prompt, tool description, and context, enabling the agent to perform tasks it could not handle before (e.g., “use MSBuild to diagnose errors”).

dotnet/skills Overview

dotnet/skills (4.9k ★, official .NET team) implements the Agent Skills standard for .NET. It provides 15 independent plugins covering LSP, MSBuild, testing, migration, AI integration, Blazor, MAUI and more, and distributes them through the plugin marketplaces of four AI agents: Copilot CLI, Claude Code, Cursor and Codex.

Problems Addressed

Relying on a generic agent alone works for simple C# but fails on deep .NET issues.

Pasting a massive .NET context into the system prompt pollutes the prompt and wastes tokens.

Writing custom .NET tooling scripts causes each team to recreate the same wheels.

dotnet/skills replaces these approaches with a standard plugin model, allowing the .NET team to maintain authoritative knowledge that can keep pace with .NET releases.

Plugin Catalog

Core development (5) : dotnet (C# LSP + general skills), dotnet‑msbuild, dotnet‑test, dotnet‑nuget, dotnet‑advanced.

Data + diagnostics + AI (5) : dotnet‑data (EF Core), dotnet‑diag, dotnet‑ai (LLM/RAG/MCP), dotnet‑template‑engine, dotnet‑maui.

Web + upgrade + migration (5) : dotnet‑aspnetcore, dotnet‑blazor, dotnet‑upgrade, dotnet‑test‑migration, dotnet11 (new .NET 11 API).

Installation is a single command per plugin; each plugin can be added or removed independently.

Real‑World Scenarios

Scenario 1 – MSBuild error diagnosis: Running dotnet build produces cryptic errors. Installing dotnet‑msbuild lets the agent follow the official MSBuild error‑code workflow, raising accuracy from ~50 % to over 90 %.

Scenario 2 – .NET 6 → .NET 11 upgrade: Installing dotnet‑upgrade and dotnet11 gives the agent knowledge of breaking changes, API renames and removals, producing an upgrade path verified by Microsoft.

Scenario 3 – Test framework migration (xUnit → MSTest v4): The dotnet‑test‑migration plugin encodes syntax differences and migration steps, and an orchestrator agent can perform the migration more reliably than a simple API‑by‑API comparison.

Architecture and Design Choices

The plugins adopt a marketplace model because all four target agents already support a marketplace protocol. This avoids requiring users to learn a new tool; they simply add the plugin to the existing agent.

The .NET team maintains the plugins to ensure deep expertise and timely updates, addressing the two shortcomings of community‑maintained skills: limited domain depth and lagging updates.

A public dashboard (https://dotnet.github.io/skills) continuously runs automated tests, tracking each plugin’s accuracy, efficiency (steps taken), and improvement trend – effectively a CI/CD pipeline for AI tools.

The project is released under the MIT license, allowing unrestricted commercial use.

Installation

/plugin marketplace add dotnet/skills /plugin install <plugin>@dotnet-agent-skills  # restart
/skills

Only the needed plugins need to be installed; the README recommends starting with dotnet (LSP + general skills) and adding others such as dotnet‑msbuild, dotnet‑ai, or dotnet‑aspnetcore as required.

Advantages

Official Microsoft backing provides authoritative knowledge.

15 plugins cover virtually every .NET scenario.

Cross‑platform support for four AI agents prevents lock‑in.

Continuous evaluation via the dashboard offers transparent quality metrics.

MIT license is business‑friendly.

Follows the open Agent Skills standard, not tied to a single ecosystem.

Automated validator ensures format correctness of each release.

Limitations and Risks

Large plugin set can be overwhelming for newcomers.

Knowledge may lag behind the latest .NET releases because updates require manual PRs.

AI agents can still hallucinate despite better context.

Only useful for .NET projects; non‑.NET codebases gain no benefit.

Dashboard metrics lack industry benchmarks, making absolute scores hard to interpret.

Validator checks format only; content usefulness still relies on human review.

Design Trade‑offs

Official maintenance vs community speed – higher authority but potentially slower iteration.

Multiple small plugins vs a monolithic plugin – flexibility at the cost of discovery effort.

Supporting four marketplaces vs deep integration with one – higher development effort but avoids vendor lock‑in.

Public dashboard vs hidden metrics – transparency versus potential competitive disadvantage.

MIT license vs commercial SDK – openness versus possible revenue streams.

Conclusion

dotnet/skills enables AI coding agents to “understand” the .NET ecosystem through 15 officially maintained plugins, distributed across four agents and continuously evaluated. The core value lies in domain knowledge owned by the domain experts, giving developers confidence that AI can reliably write and maintain .NET code.