Skip to content

Files

Latest commit

 

History

History
 
 

README.md

Note

Copilot Extensions are in public preview and may be subject to change. Pre-release terms apply. You must have a GitHub Copilot license of any type to use Copilot Extensions.

Skillset Example

Learn to build a Copilot Extension as a skillset. Just define your API endpoints and let Copilot handle the complex AI logic. This sample demonstrates the lighter, faster alternative to building a traditional agent.

Architecture

  • Skillsets: Define up to 5 API endpoints that Copilot can call directly. Copilot handles all AI interactions, prompt engineering, and response formatting.
  • Agents: Provide full control over the interaction flow, including custom prompt crafting and specific LLM model selection.

Architectural comparison between Skillsets and Agents

Key Differences

Skillsets Agents
Development Time Minutes to hours Several days to weeks
Implementation Configure API endpoints only. Quicker if you have existing APIs. Custom AI logic, prompt engineering, interaction handling.
Best For • Quick API integrations
• Basic service functionality
• Standard Copilot interactions
• Minimal infrastructure needs		 • Complex interaction flows
• Custom LLM model control
• Custom prompt crafting
• Advanced state management
Control Level Copilot platform handles AI interactions Full control over entire workflow
Data Volume Good for any volume, if the API handles efficient search/retrieval Good for any volume, but requires robust search/retrieval at larger sizes

Example Prompts

This extension turns three public APIs into functions that can be used in Copilot Chat":

@extension generate a commit message
→ "Whatever will be, will be 8{)"

@extension generate 2 short lorem ipsum paragraphs
→ [HTML-formatted Lorem Ipsum text]

@extension generate random user data
→ {
  "name": "Miss Mona",
  "email": "mona.lisa@github.com",
  ...
}

Quick Start

1. Set Up Repository

git clone git@github.com:copilot-extensions/skillset-example.git
cd skillset-example
go mod tidy

2. Make It Accessible

  1. Run your app:
go run .
  1. Then choose one method to expose it and copy the URL for your GitHub App config:
  • VS Code: Set port 8080 public, copy Local Address
  • Codespaces: Set port 8080 public, copy URL
  • Local with ngrok:
    1. Install ngrok from ngrok.com
    2. Start tunnel: ngrok http 8080
    3. Copy the Forwarding URL

Note: This step is for development and testing only. For production use, deploy your extension to a public server with proper infrastructure and security.

3. Configure GitHub App

  1. Create a new GitHub App at github.com/settings/apps/new or go to an existing one
  2. In the Copilot tab:
    • Set type to "Skillset"
    • Define the three skills below, using your URL from step 2:
Name: random_commit_message
Inference description: Generates a random commit message
URL: https://<your ngrok domain>/random-commit-message
Parameters: { "type": "object" }
Return type: String
---
Name: random_lorem_ipsum 
Inference description: Generates a random Lorem Ipsum text.  Responses should have html tags present.
URL: https://<your ngrok domain>/random-lorem-ipsum
Parameters: 
{
   "type": "object",
   "properties": {
      "number_of_paragraphs": {
         "type": "number",
         "description": "The number of paragraphs to be generated.  Must be between 1 and 10 inclusive"
      },
      "paragraph_length": {
         "type": "string",
         "description": "The length of each paragraph.  Must be one of \"short\", \"medium\", \"long\", or \"verylong\""
      }
   }
}
Return type: String
---
Name: random_user
Inference description: Generates data for a random user
URL: https://<your ngrok domain>/random-user
Parameters: { "type": "object" }
Return type: String
  1. In the General tab of your GitHub App settings (https://github.com/settings/apps/<app_name>)
    • Homepage URL: You can set to any URL for quick testing (like https://github.com).
    • Callback URL: Only needed if your extension uses OAuth for authentication. Since this example doesn't use OAuth, set to any URL (like https://github.com)
  2. Ensure your permissions are enabled in Permissions & events > Account Permissions > Copilot Chat > Access: Read Only
  3. Install your app at https://github.com/apps/<app_name>

4. Try It Out

  1. Open Copilot Chat on github.com or in any supported IDE
  2. Send a test prompt like @test-data-bot generate a commit message

Note: It may take a few seconds for your extension to appear. Try refreshing if needed.

Documentation