Teach GitHub Copilot CLI Your Design Language using Skills

Teach GitHub Copilot CLI Your Design Language using Skills

Hi there! Starting a new Cloud & AI newsletter to share my learnings as I discover new exciting tech (not only) in the Microsoft world.

Github Copilot CLI: The Power of AI in Your Terminal

Instead of switching between your IDE, browser, and terminal, you now have a powerful AI agent that lives where you work—directly in your command line. Copilot CLI brings the same agentic capabilities powering GitHub's Copilot coding agent to your local environment.

What can you get our of Copilot CLI?:


  • You can build, refactor, debug
  • Access Github directly (issues / PRs, repos...) using natural language
  • Use latest standards in the industry that are built in like MCPs or Skills


What's New: Skills, MCP, and Model Selection

Recent updates have transformed Copilot CLI from a helpful assistant into a truly customizable development companion:


  • Model Selection (using /model) - Choose between the latest state of the art models incl. OpenAI Codex, Sonnet 4.5 or Opus 4.5
  • MCP Server Support (/mcp) - You can Connect Copilot to external tools and services through the Model Context Protocol. Database access, API integrations, custom tooling
  • Custom Skills (/skills) - with this latest standard from Antrhopic you can teach Copilot your specific standards, patterns, and preferences. Skills persist across sessions and are automatically applied when relevant.


Why Skills Matter

The knowledge of your company typically lives in documentation that developers must remember to reference and AI must be repeatedly prompted to follow. Skills solve this elegantly. They're markdown files that live in your repo, containing domain-specific knowledge that Copilot references automatically. Write your standards once; Copilot applies them every time.

The value is compounding:


  • Consistency - Every team member gets the same behavior
  • Efficiency - No re-explaining your preferences in every prompt
  • Quality - Codified standards reduce review cycles
  • Onboarding - New developers inherit team knowledge instantly


If you want to learn more about Skills I recommend watching this video from Anthropic:

https://www.youtube.com/watch?v=CEvIs9y1uog

My use cases: A Microsoft Inspired Branding Skill

I wanted to build a branding Skill that teaches Copilot Microsoft design colors for charts, dashboards, and other visual content. At the same time I also wanted to make sure to keep my artifacts clean with minimal aesthetic inspired by shadcn/ui.

Skills is a markdown file with frontmatter and detailed guidelines that looks like this:

Article content

The skill captures our design philosophy, typography guidelines, card styling and ready-to-use code snippets in CSS, TypeScript, and Python.

With the skill in place, the same prompt like "create a sample dashboard with chart included with Microsoft inspired design" now produces an output like the one below (using sample data). The skill is automatically invoked when Copilot detects keywords like "branding," "chart," or "Microsoft inspired design." No additional context needed.

Article content

Based on my experience, here's what makes a skill work well:

1. Clear Frontmatter - Name and description that help Copilot understand when to apply it

2. Why -Help Copilot understand the why behind your choices

3. Specific Values - Include exact values where needed

4. Multiple Output Formats - CSS, Typescript and Python in my case

5. Use AI to automate Skill creation - I used the latest Opus model with several image examples.

Getting Started with Your Own Skills

1. Create the directory: .github/skills/your-skill-name/

2. Add SKILL.md with frontmatter and detailed specifications

3. Run /skills list in Copilot CLI to verify it's loaded

4. Test with prompts that should trigger the skill

5. Iterate based on output quality

The Bigger Picture

Skills represent a shift in how we think about AI assistance. Instead of treating AI as a generic tool that requires constant guidance, we can invest time upfront to create persistent, shareable knowledge that compounds over time.

Disclaimer: All opinions expressed are my own and are not intended to represent the views, positions, or strategies of Microsoft.

Martin Bosák

AI Agent Management / Data Product Platform Architect / CFA AI Assembly Lead

4mo

Nice, Jakub, keep up! btw what about similar concepts in other LLMs?

To view or add a comment, sign in

More articles by Jakub Kudela

Others also viewed

Explore content categories