How to Teach an AI Assistant to Understand Azure as Well as a Solutions Architect

Recently I caught myself thinking: we've gotten used to GitHub Copilot or Claude Code knowing almost everything, but as soon as it comes to cloud specifics, hallucinations begin. The neural network might give you an Azure Functions config that's been outdated for about three years, or suggest an architecture that violates built-in subscription limits. The problem is that model knowledge is limited to the moment of their training, and Microsoft Learn documentation changes almost daily.

I stumbled upon the Agent-Skills repository from the folks at Microsoft. The idea is simple but incredibly useful: instead of making the AI "remember" documentation, you feed it structured skills. It's like a reference guide for the agent that lives right in your project and tells it: "When working with Azure, look here, use these patterns, and don't forget about these limitations."

What Is This Anyway

The project is a huge catalog of almost two hundred ready-made skills for working with Azure. These aren't just links to documentation. Each skill is a set of instructions in Markdown format that the AI assistant reads before suggesting code to you.

Developers at Microsoft went the "documentation as code" route. They took a mountain of content from Microsoft Learn — best practices, decision trees, common mistakes, and security requirements — and packaged it into LLM-friendly instructions. As a result, the agent doesn't just write code; it understands the context of the specific service.

Why Bother Adding This to Your Project

If you work with Azure, you know how many nuances there are. For example, you can configure Azure Container Apps in a dozen ways, but only a couple of them will meet Enterprise standards.

Here's what these skills give you in practice:

1. The agent stops guessing and starts using current API patterns.
2. Responses include direct links to relevant Microsoft Learn sections.
3. The AI starts warning you about limits and configuration quirks before you deploy a broken config.
4. The agent gets ready-made checklists for troubleshooting.

How It Works Under the Hood

At its core is an open Agent Skills standard. Each skill lives in its own folder inside the skills/ directory. The SKILL.md file structure looks something like this:

---
name: azure-functions
description: Инструкции по разработке и развертыванию Azure Functions
---

# Azure Functions

## Инструкции
[Тут идут конкретные шаги для агента]

## Примеры
[Примеры правильного использования сервиса]

Interestingly, the system works on a progressive disclosure principle. First, the agent sees only the name and brief description in the YAML header. If your request matches the skill's topic, the agent "unfolds" the full instruction and starts acting according to it. This saves the model's context window without cluttering it with unnecessary stuff.

How to Get This Thing Running

The easiest way is to install everything as a plugin. For VS Code, you do this through the command palette: select Chat: Install Plugin From Source and paste the repository link https://github.com/MicrosoftDocs/agent-skills.

If you prefer full control or use Cursor, you can simply clone the repository and copy the needed folders from skills/ to your project directory. The path varies for each tool:

• GitHub Copilot: .github/skills/
• Cursor: .cursor/skills/
• Claude Code: .claude/skills/

By the way, for VS Code you need to remember to enable the chat.agent.skills setting, otherwise the magic won't happen.

Practical Benefits and Categories

The repository currently has 193 skills. They're broken down by category, so you don't have to drag everything along — you can pick only what you need for your specific stack:

• Compute (13 skills): App Service, Functions, Batch.
• AI & ML (23 skills): everything about Azure AI Services and bots.
• Security (19 skills): Active Directory, Key Vault, and more.
• Infrastructure (52 skills): this is where the heavy stuff is — AKS, Arc, and network configurations.

If you don't know where to start, the docs/BUNDLES.md folder has curated collections. For example, the "Quick Start" bundle includes 7 basic services that will be useful to any Azure developer.

Is It Worth Trying

The project looks like a great way to "ground" AI assistants on real tasks. Instead of copy-pasting documentation snippets into the Copilot chat every time, you connect the repository once and get a competent assistant that knows the rules of the game in the Microsoft ecosystem.

This will be especially useful for teams that are just migrating to Azure or starting to use complex services like Durable Functions or Azure API Management. It's easy to misconfigure them, and fixing it later is time-consuming and expensive.

The only downside is that the project is still in preview, and some features require enabling experimental flags in the IDE. But even in manual mode, copying the needed .md files into your project gives a noticeable boost to code generation quality. If you're heavily invested in Azure, this repository definitely belongs in your bookmarks.