microsoft/skills
Skills, MCP servers, Custom Agents, Agents.md for SDKs to ground Coding Agents
Microsoft / skills – Agent Skills for Azure‑powered AI coding agents
What it is – A curated collection of “skills” – ready‑made code snippets, configuration templates, and small plug‑ins that teach AI coding agents (e.g., GitHub Copilot CLI, Copilot in VS Code) how to work fluently with Microsoft Azure services, the Microsoft Foundry platform, and a host of other Microsoft SDKs. Each skill lives in its own folder (e.g., azure-cosmos-db-py/) and can be installed into an agent’s .github/skills/ directory so the agent can call the appropriate SDKs, generate boiler‑plate, or perform common operations without you having to write the low‑level code yourself.
Why it matters
- Domain knowledge for agents – Large‑language‑model agents already know the shape of Azure SDKs, but they need the right activation context to surface the right patterns. Skills provide that context, turning a generic LLM into a specialist Azure developer.
- Plug‑and‑play – Over 170 skills cover Azure core services (Cosmos DB, Storage, Event Grid, etc.), AI services (OpenAI, Vision, Speech), Microsoft 365, and the new Foundry agent platform. Install only the ones you need, keeping the agent’s prompt size small and the token budget efficient.
- Consistent tooling – Skills also ship with MCP (Model‑Control‑Plane) server templates, custom agents, and
AGENTS.mdconfiguration files, giving you a repeatable way to spin up agents for backend, frontend, infrastructure, or planning roles.
Quick start (one‑liner)
npx skills add microsoft/skills # launches a wizard to pick the skills you want
The wizard copies the selected skill folders into your project (e.g., .github/skills/azure-cosmos-db-py/) and creates symlinks if you share the same skills across multiple agents.
Alternative: clone the repo and copy the needed folders manually, or symlink them for a multi‑project setup.
Core pieces of the repository
| Piece | What it gives you |
|---|---|
| Skill catalog (≈175 skills) | Ready‑made, language‑specific wrappers for Azure SDKs, Foundry APIs, and other Microsoft services. Each skill includes sample code, a skill.md description, and optional MCP server scaffolding. |
| Plugins | Small npm/py/nuget packages (e.g., deep-wiki, azure‑skills) that can be added to an agent’s runtime to expose extra tooling. |
| Custom agents | Role‑oriented agent templates (backend, frontend, infra, planner) with pre‑filled AGENTS.md files that tell the agent how to behave. |
| MCP configs | Pre‑configured Model‑Control‑Plane servers for things like docs lookup, web‑search, code‑interpreter, and A2A tool orchestration. |
| Documentation & Skill Explorer | A web UI (https://microsoft.github.io/skills/) that lists every skill, shows a one‑click install button, and links to detailed usage docs. |
Typical use cases
- Accelerate Azure SDK development – Add
azure-storage-blob-pyand let Copilot generate correct Blob‑storage CRUD code instantly. - Build a Foundry‑hosted agent – Use the
foundry-hosted-agentsskill to scaffold a containerized agent that registers with the Foundry MCP, gets its own Entra identity, and can be called via the Foundry API. - Create long‑term memory for agents – Install
foundry-memoryto give an agent a persistent knowledge store that survives across sessions. - Enforce governance – Deploy
foundry-governanceto set RBAC, routing policies, and responsible‑AI checks for a fleet of agents. - Generate production‑ready scaffolding – Combine
fastapi-router-pywith a storage skill to spin up a fully typed FastAPI service with authentication and data persistence.
Who should use it
- Developers building AI‑assisted tools on Azure who want their Copilot/Foundry agents to understand Azure‑specific patterns out of the box.
- DevOps / Platform engineers who need repeatable, version‑controlled skill sets across many repositories or teams.
- Machine‑learning engineers leveraging Azure AI services (OpenAI, Vision, Speech) who want agents to generate correct SDK calls.
- Product teams that want to enforce consistent architecture (e.g., well‑architected cloud solutions) via the
cloud-solution-architectskill.
Getting help / contributing
- Tests & CI – Badges at the top of the README show that every skill is exercised by a test harness and Copilot‑SDK evaluation workflow.
- Contributions – The repo is marked Work in Progress; you can open PRs to add new skills, improve documentation, or expand test coverage.
- Docs – Full documentation lives at https://microsoft.github.io/skills/ (installation guide, skill catalog, API reference).
TL;DR
microsoft/skills is a real, actively maintained toolbox that turns generic AI coding agents into Azure‑savvy assistants by providing hundreds of plug‑and‑play “skills” (code templates, MCP servers, and configuration files). Install only the skills you need, keep the agent’s context small, and let the agent generate correct Azure SDK code instantly.
Links
- Repository: https://github.com/microsoft/skills
- Skill Explorer UI: https://microsoft.github.io/skills/
- Blog post (context‑driven development): https://devblogs.microsoft.com/all-things-azure/context-driven-development-agent-skills-for-microsoft-foundry-and-azure/
Related
- Project
- Project
- Project
- Project
- Project