Skip to content

Latest commit

 

History

History
184 lines (114 loc) · 8.69 KB

part2-custom-skills.md

File metadata and controls

184 lines (114 loc) · 8.69 KB

Part 2 – Custom Copilot Skills

Navigation: Part 1 – Setup & Chat | Part 2 | Part 3 – Custom Agents | Part 4 – Issues, Coding Agent & Review

Writing Custom Skills

Goal: Extend Copilot's capabilities by writing custom skills that teach it new behaviors.

What are Copilot Skills?

Skills are folders of instructions and resources that Copilot can load to improve how it responds to specific tasks. Each skill lives in .github/skills/<skill-name>/ and contains a SKILL.md file with instructions.

Key points:

  • Skills are automatically loaded by Copilot when it determines they are relevant to the user's prompt.
  • They work in Agent mode in VS Code, the Copilot Coding Agent, and GitHub Copilot CLI.
  • You can also place personal (cross-project) skills in ~/.copilot/skills/<skill-name>/SKILL.md.

Documentation: About agent skills

Exercise A: Create an explicit "Hello ASCII" skill

In this exercise, you'll create a skill that displays "Hello" in ASCII art. You'll need to explicitly ask Copilot to use it by referencing it directly.

5. Create the hello-ascii skill

Step 1 — Create the skill folder

Make sure you're on your copilot-hackathon branch before creating these files.

  • In your repo, create the directory: .github/skills/hello-ascii/

Step 2 — Write the SKILL.md

  • Create .github/skills/hello-ascii/SKILL.md with the following content:
# Hello ASCII

Display a "Hello" message using ASCII art.

## When to use

When the user explicitly asks to "show hello in ASCII art", "use the hello skill",
or "display an ASCII hello".

## Instructions

Respond with the following ASCII art:

    _   _      _ _
   | | | | ___| | | ___
   | |_| |/ _ \ | |/ _ \
   |  _  |  __/ | | (_) |
   |_| |_|\___|_|_|\___/

Always display the ASCII art exactly as shown. Do not modify or substitute it.

Step 3 — Test it

  • Open Copilot Chat in your IDE (make sure you're in Agent mode).
  • Type: "Show hello in ASCII art" or "Use the hello-ascii skill".
  • Verify Copilot responds with the ASCII art from your skill.

If Copilot doesn't pick it up, try reloading your IDE window or starting a new chat session.

Exercise B: Extend to trigger implicitly on any greeting

Now you'll create a second version of the skill with a broader description so that Copilot automatically loads it whenever you simply greet it — no explicit reference needed.

6. Create the greeting skill

Step 1 — Create a new skill folder

Make sure you're on your copilot-hackathon branch before creating these files.

  • Create the directory: .github/skills/greeting/

Step 2 — Write the SKILL.md

  • Create .github/skills/greeting/SKILL.md with the following content:
# Greeting

Respond to user greetings with a friendly ASCII art hello.

## When to use

When the user greets you. This includes any of the following:
- "hello", "hi", "hey", "howdy"
- "good morning", "good afternoon", "good evening"
- "greetings", "what's up", "yo"
- Swedish greetings: "hej", "tjena", "hallå", "god morgon", "god kväll"
- Any other conversational greeting or salutation in any language

## Instructions

When a greeting is detected, respond with this ASCII art:

    _   _      _ _       _
   | | | | ___| | | ___ | |
   | |_| |/ _ \ | |/ _ \| |
   |  _  |  __/ | | (_) |_|
   |_| |_|\___|_|_|\___/(_)

Follow the ASCII art with a brief, friendly response. Keep it short and warm.

Step 3 — Test the implicit trigger

  • Open a new Copilot Chat session (Agent mode).
  • Type just: "Hi" or "Good morning".
  • Observe whether Copilot responds with the ASCII art greeting — without you mentioning the skill at all.

Step 4 — Compare the two versions

  • Try the same greetings with only the hello-ascii skill present (rename or delete the greeting folder temporarily).
  • Notice how the broader description in the greeting skill makes Copilot load it automatically, while the narrower hello-ascii version requires explicit references.

This demonstrates how the description and "When to use" section in a skill controls when Copilot decides to load it. Broader descriptions mean more implicit activation.

Key Takeaways

Aspect hello-ascii (explicit) greeting (implicit)
Trigger User must reference it directly Any greeting triggers it
"When to use" Narrow — specific phrases only Broad — many greeting patterns
Behavior Predictable, controlled Automatic, seamless
Use case Specialized tools Ambient behaviors

Both approaches are valid — choose based on whether you want control or convenience.

Reference implementations of both skills are available in this repo under .github/skills/.

Why Skills? Real-world uses for your team

The ASCII art exercises above are intentionally simple — but skills really shine when they encode team-specific knowledge that Copilot wouldn't otherwise have. The community github/awesome-copilot repository contains hundreds of real skills, and they follow consistent patterns. Consider skills that:

  • Generate conventional commit messages — a skill like conventional-commit that guides Copilot to produce standardized commit messages following the Conventional Commits specification, with type, scope, and description, every time.
  • Plan and validate deployments — skills like devops-rollout-plan or azure-deployment-preflight encode your rollout process: preflight checks, step-by-step deployment, verification signals, and rollback procedures.
  • Automate and document GitHub Actions workflows — a skill like create-github-action-workflow-specification captures your CI/CD pipeline structure so Copilot can generate, document, or troubleshoot workflows consistently.
  • Generate tests across languages — skills like polyglot-test-agent teach Copilot to generate comprehensive unit tests for any language in your stack, following your project's testing conventions.
  • Document and onboard — skills like architecture-blueprint-generator or folder-structure-blueprint-generator produce standardized architectural documentation from your codebase, giving new developers (and Copilot) a fast path to understanding how the project is structured.

Skills turn Copilot from a general-purpose assistant into one that understands your team's way of working. They're version-controlled alongside your code, reviewed in PRs, and shared automatically with everyone on the project.

Browse the full community skill library at github/awesome-copilot for ready-to-use examples you can copy or adapt.

Documentation & Resources

Getting started:

Technical reference:

Community & examples:

  • awesome-copilot — curated collection of community skills, agents, and extensions

Next up: Part 3 – Custom Agents (continues at step 7)