Dev & Engineering ✓ Microsoft · Official azuremicrosoft-foundryazure-sdkmcp-serverscopilot-sdkcustom-agentscode-testing

Microsoft Agent Skills

Reusable Azure SDK and Microsoft Foundry guidance for AI coding agents.

FollowSkills review · FSRS-1.0
Use with care
74/ 100
Utility16 / 20
Reliability14 / 20
Safety20 / 25
Evidence8 / 15
Usability8 / 10
Maintenance8 / 10

Officially published by Microsoft (verifiable official source), for Azure SDKs and Microsoft AI Foundry, with CI badges for evals/tests; README flags active development, with a companion technical blog post for cross-verification.

Evidence confidence:Medium Reviewed Jul 18, 2026 Reviewed revision None
Review evidence [1]
See the full review method →

What it does & when to use it

Microsoft Agent Skills is an open-source repository of skills, custom agents, AGENTS.md templates, and MCP configurations for coding agents working with Azure SDKs and Microsoft AI Foundry. Its catalog spans Python, .NET, TypeScript, Java, and Rust, covering AI services, storage, messaging, identity, monitoring, infrastructure, and frontend patterns. The repository also includes installable plugins, prompt templates, role-specific agents, and a GitHub Copilot SDK-based test harness. It is explicitly a work in progress, so coverage and SDK guidance may change.

It provides an npx installation wizard for selecting skills and placing them in an agent directory such as .github/skills/, with symlink-based sharing across agents and projects. Its SKILL.md resources provide guidance for Azure SDKs, Microsoft Foundry, Copilot SDK, MCP servers, frontend systems, and cloud architecture. The repository also supplies deep-wiki and azure-skills plugins, AGENTS.md and role-agent templates, MCP server configurations, reusable prompts, and acceptance tests with mock execution and iterative evaluation patterns.

  1. An Azure developer using Python, .NET, TypeScript, Java, or Rust needs service-specific SDK patterns for authentication, imports, storage, messaging, or AI workloads.
  2. A team is building Microsoft Foundry projects, model deployments, hosted agents, workflows, knowledge bases, memory, observability, or governance features.
  3. A coding-agent user wants backend, frontend, infrastructure, planning, or presentation context packaged as reusable agent resources.
  4. A developer needs to build MCP servers or configure documentation, GitHub, browser automation, Terraform, or utility MCP servers.
  5. A team wants acceptance criteria, mock scenarios, and iterative evaluation for code generated from agent skills.

Pros & cons

Pros
  • Broad coverage of Azure SDKs and Microsoft Foundry across five programming languages.
  • Combines skills with plugins, custom agents, AGENTS.md templates, MCP configurations, prompts, and tests.
  • Provides a copyable installation command and multi-agent sharing patterns.
  • Includes acceptance criteria, test scenarios, mock mode, and iterative quality evaluation.
Limitations
  • The repository is explicitly under active development, so guidance and coverage may change.
  • The README contains inconsistent skill totals and category counts, so no single exact total can be confirmed from the supplied material.
  • SKILL.md content is not cached, so individual skill prerequisites, triggers, and implementation details cannot be verified here.
  • The test-coverage summary covers only a stated subset of languages and skills and does not establish equal coverage for every resource.

How to install

Recommended command: npx skills add microsoft/skills. Use the wizard to select the required skills; installation places them in the chosen agent directory, such as .github/skills/ for GitHub Copilot. Alternatively, run git clone https://github.com/microsoft/skills.git and copy selected skill directories, or use the documented symlink approach.

How to use

Install only the skills relevant to the current project. Then give a compatible coding agent a concrete request, for example: "Use azure-ai-projects-py to create a Microsoft Foundry project client and configure evaluations." The README does not document one universal trigger syntax or one single skill path; the exact location depends on the selected skill and agent configuration.

FAQ

Who should adopt this repository?
It is aimed at developers and teams using Azure SDKs, Microsoft Foundry, or AI coding agents such as GitHub Copilot.
Should all skills be installed?
No. The README recommends installing only skills essential to the current project to avoid diluted context and wasted tokens.
Does installation require network access and local tooling?
The documented installation methods use npx and git clone, and the repository references MCP, documentation, and GitHub resources, so installation or some capabilities require network and command-line access.
Is it intended for non-Microsoft cloud or general office workflows?
The supplied material focuses on Azure SDKs, Microsoft Foundry, and coding-agent workflows; it does not establish broader support for other clouds or general office use.

Related skills