Claude Code with Copilot Studio: The Playbook
Howdang Rashid
Saturday, 22 August 2026 · 3 min read
Building agents in Copilot Studio has always been a UI-first experience - clicking through menus, configuring what Microsoft gives you. It works, but it can feel slow, especially when you already know what you want to build. Then Microsoft's CAT team shipped something that changes the equation: a plugin that lets Claude Code operate Copilot Studio directly from your terminal.
The full playbook carousel is free in the Powercademy Success Kit.
What is the Skills for Copilot Studio plugin?
An open-source plugin from Microsoft's CAT team (Copilot Acceleration Team) that plugs into Claude Code and lets you author, test, and troubleshoot agents from your terminal using natural language. The magic is what it adds: the CAT team works directly with enterprise customers on production agent architecture, and they've encoded their internal patterns, full YAML schema knowledge, and best practices straight into the plugin. The result: a general coding agent plus platform knowledge equals a Copilot Studio developer - and Microsoft's own framing is "up to 20x faster" than UI development.
What are the three commands?
/copilot-studio:author - your creator. Describe what you need in natural language and it generates the full YAML: topics, actions, knowledge sources, triggers, variables. "Create a topic that handles IT service requests" or "refactor this agent to scale to 30+ use cases".
/copilot-studio:test - send test utterances to your published agent from the terminal: point-tests, batch suites across ten scenarios at once, and evaluation analysis, all without leaving your shell.
/copilot-studio:troubleshoot - describe the problem ("agent is hallucinating, not using KB data", "wrong topic triggering") and it diagnoses the cause, suggests fixes, and can make the changes itself.
What's the workflow?
Create the baseline agent through the Copilot Studio portal.
Clone it locally using the VS Code Copilot Studio extension.
Author, test, and debug in your terminal with the three plugin commands.
Push via the VS Code extension, then publish in the portal.
Prerequisites: Claude Code or GitHub Copilot CLI, VS Code with the Copilot Studio extension, and a Copilot Studio agent (a blank one is fine).
Is it production-ready?
My honest take: not fully, yet. It's experimental, though the team is genuinely responsive to suggestions and improvements. What's significant is the direction: Microsoft is encoding its best practices directly into tools like Claude Code, and the gap between a new maker and someone who can develop this stuff under the hood just got smaller.
FAQ
Does this work with GitHub Copilot too, or only Claude Code?
Both - the plugin targets Claude Code and the GitHub Copilot CLI. Same commands, same CAT-team knowledge, whichever assistant your organisation runs.
Do I still need the Copilot Studio portal?
Yes - the workflow starts there (create the baseline agent) and ends there (publish). The terminal takes over the middle: authoring, testing, and debugging, which is where most of the time went anyway.
Is terminal-based agent building safe for production agents?
With discipline, yes: everything flows through YAML you can review and source-control, which is arguably safer than untracked portal clicks. But the plugin is experimental - the schema can change, and AI output still needs human review before publishing.
Where do I get the plugin?
It ships in Microsoft's open-source power-platform-skills marketplace on GitHub - add the marketplace to Claude Code or Copilot CLI and install the copilot-studio plugin. The full walkthrough video is linked in the playbook.
The full playbook is in the Powercademy Success Kit - free, along with hundreds of other cheat sheets, roadmaps, and guides for Microsoft professionals.