Copilot Library

Augment GitHub Copilot with custom domain knowledge. Improve generic code model alignment and context grounding. Reduce token use by providing more context to coding agents.

Features

• Collect code examples/files with prompt, incorrect/correct responses, and explanations
• Configurable metadata fields (languages, frameworks, topics, or add custom via settings)
• View, sort, search by folder, language or metadata fields
• Export to JSONL format optimized for model training

Installation

1. Go to the Releases page
2. Download the latest .vsix file from the release assets
3. In VS Code, open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P)
4. Run the command Extensions: Install from VSIX...
5. Select the downloaded .vsix file
6. Reload VS Code when prompted

Configuration

• Configure storage location - Store your data in cloud-synced folders (OneDrive, Google Drive, etc.) Default location is extension storage folder
• Configure GitHub Copilot chat integrations (Check out this readme for complete Copilot integration guide)

Usage

Collect your data

1. Right-click on a file, file tab or a code selection inside a file

2. Choose Copilot Library > Add new (in)valid response snippet to create new record or Add to existing record to update existing record.

3. Fill out the form:

   • Name & Description: What this training example demonstrates
   • Prompt: The instruction that should generate the code
   • Incorrect Response: Multi-file example of what NOT to generate
   • Explanation: Why the incorrect approach is wrong (optional)
   • Correct Response: Multi-file example of what SHOULD be generated
   • Metadata: Languages, frameworks, topics (all optional and customizable)

   and hit Create/Update

4. View and manage records in the Training Records tree view

Export training dataset for fine-tuning

• Export complete training dataset to JSONL format by clicking "Export Training Dataset"

Record Structure

{
  "prompt": "Create a Blazor Server feature to update a user profile.",
  "incorrect_response": {
    "User.razor": "// Bad example without proper patterns",
    "UserService.cs": "// Bad example without DI/async"
  },
  "explanation": "The incorrect version lacks dependency injection, async methods, and proper error handling.",
  "correct_response": {
    "User.razor": "// Good example with proper Blazor patterns",
    "UserService.cs": "// Good example with DI and async"
  },
  "metadata": {
    "languages": ["csharp", "html"],
    "frameworks": ["blazor", "aspnet-core"],
    "topics": ["user-interface", "dependency-injection"]
  }
}

Customize metadata

You can customize the metadata fields by clicking the settings button (⚙️) in the View Options panel to open the extension-config.json file. This allows you to add custom fields, modify existing ones, or create specialized field types for your training data.

Available Field Types

• text: Single-line text input (e.g., project name, author)
• select: Single choice from predefined options (e.g., difficulty level)
• multiselect: Multiple choices from predefined options (e.g., programming languages)
• number: Numeric input with optional min/max validation (e.g., complexity score)
• boolean: True/false checkbox (e.g., includes tests, production ready)
• tags: Free-form tags that users can type (e.g., frameworks, topics)

Field Configuration Properties

{
  "id": "unique_field_identifier",
  "label": "Display Name",
  "type": "text|select|multiselect|number|boolean|tags",
  "required": false,
  "options": ["option1", "option2"],      /* For select/multiselect only */
  "defaultValue": "default",              /* Optional default value */
  "description": "Help text for users",   /* Optional tooltip */ 
  "validation": {                         /* Optional validation rules */
    "min": 0,                            /* For number fields */
    "max": 10,                           /* For number fields */
    "pattern": "regex_pattern"           /* For text fields */
  }
}

Examples

Add a difficulty rating:

{
  "id": "difficulty",
  "label": "Difficulty Level", 
  "type": "select",
  "required": false,
  "defaultValue": "Beginner",
  "options": ["Beginner", "Intermediate", "Advanced", "Expert"],
  "description": "Code complexity level"
}

Add a production-ready flag:

{
  "id": "production_ready",
  "label": "Production Ready",
  "type": "boolean",
  "required": false,
  "defaultValue": false,
  "description": "Is this code suitable for production use?"
}

After modifying extension-config.json, restart VS Code to apply the changes.

Configuration

• copilotLibrary.datasetPath: Training dataset export location