Questions about using prompts and instructions

[yorkeChen]

Select Topic Area

Question

Copilot Feature Area

General

Body

I’m working with prompt files and instructions files in VS Code while using Copilot in Agent mode, and I’m trying to understand how they interact.

Right now, I trigger a prompt by typing something like /xx.prompt.md in the chat. If I also include an instructions file, how are these two applied?

Do prompt files and instructions run together automatically?
Or does the prompt file only activate when explicitly invoked, while the instructions file serves as a persistent global rule set?

If there are recommended usage patterns or best-practice examples on how to combine prompts and instructions in an Agent workflow, I’d appreciate some guidance.

[Dinuka-Nonis]

You’re thinking about it the right way. Basically:

Prompt files only run when you explicitly call them, like when you type /xx.prompt.md. They don’t do anything on their own.

Instructions files are persistent , they act like global rules or context that the Agent keeps in mind all the time. So every prompt you run will automatically follow the instructions.

A good way to think about it is: instructions files = the Agent’s “personality” or “guidelines,” and prompt files = the specific tasks you want it to do.

A common workflow is to keep instructions files for general behaviors and context, then trigger different prompts as needed for specific tasks. That way, you don’t have to repeat context or rules every time, and your workflow stays clean and efficient.

[4syr]

Additional Context: Working with Prompts and Instructions in Copilot Agent Mode

---

Hi @yorkeChen,

Great question! @Dinuka-Nonis provided an excellent foundation. Let me expand on this with more technical details and practical examples to help you get the most out of combining prompts and instructions.

---

How They Work Together

Instructions Files (.github/copilot-instructions.md)

• Scope: Global and persistent across all Copilot interactions
• Activation: Automatically applied to every chat session
• Purpose: Define coding standards, project conventions, and behavioral guidelines
• Location: Typically placed at the repository root or workspace level

Prompt Files (.prompt.md)

• Scope: Task-specific and invoked on-demand
• Activation: Only when explicitly called via /filename.prompt.md
• Purpose: Execute specific workflows, templates, or complex tasks
• Location: Can be organized in a .github/prompts/ directory or anywhere in your workspace

---

Execution Flow

When you invoke a prompt file, here's what happens:

1. Instructions are loaded first (global context)
2. Prompt file is executed (specific task)
3. Both contexts merge for the Agent's response

Think of it as: Instructions + Prompt = Contextualized Output

---

Best Practice Examples

Example 1: Instructions File

<!-- .github/copilot-instructions.md -->

# Project Guidelines

## Code Style
- Use TypeScript strict mode
- Follow ESLint rules defined in .eslintrc
- Prefer functional components with hooks

## Naming Conventions
- Use camelCase for variables and functions
- Use PascalCase for components and classes
- Prefix interfaces with 'I'

## Testing Requirements
- Write unit tests for all business logic
- Use Jest and React Testing Library
- Aim for 80% code coverage

Example 2: Prompt File

<!-- .github/prompts/create-component.prompt.md -->

# Create React Component

Create a new React component with the following structure:
- Functional component with TypeScript
- Props interface definition
- Basic styling with CSS modules
- Unit test file with basic render test

Component name: {user will specify}

Example 3: Usage

User: /create-component.prompt.md Button

The Agent will:

• Apply the instructions (TypeScript strict, PascalCase naming, testing requirements)
• Execute the prompt (create component structure)
• Output a component that follows both the global rules and the specific template

---

Recommended Workflow Patterns

Pattern 1: Role-Based Instructions

<!-- .github/copilot-instructions.md -->

# Role
You are a senior full-stack developer specializing in React and Node.js

# Always
- Consider performance implications
- Include error handling
- Add inline documentation

Pattern 2: Prompt Library Organization

.github/prompts/
├── backend/
│   ├── create-api-endpoint.prompt.md
│   └── write-database-migration.prompt.md
├── frontend/
│   ├── create-component.prompt.md
│   └── create-hook.prompt.md
└── testing/
    ├── write-unit-test.prompt.md
    └── write-e2e-test.prompt.md

Pattern 3: Contextual Prompts
Create prompts that reference the instructions:

<!-- create-feature.prompt.md -->

# Create Feature Module

Following the project guidelines in copilot-instructions.md:
1. Create component structure
2. Add TypeScript interfaces
3. Implement business logic
4. Write comprehensive tests
5. Update documentation

---

Advanced Tips

1. Layered Instructions

   • Workspace-level: General coding standards
   • Project-level: Specific framework conventions
   • File-level: Component-specific rules

2. Dynamic Prompts

   • Use placeholders in prompts: {componentName}, {apiEndpoint}
   • Copilot will prompt you for these values

3. Chaining Prompts

   • Create workflow prompts that reference other prompts
   • Example: /full-feature.prompt.md could internally reference /create-component.prompt.md and /write-test.prompt.md

4. Version Control

   • Commit both instructions and prompts to your repository
   • Team members automatically get the same context

---

Common Pitfalls to Avoid

❌ Don't duplicate rules between instructions and prompts
✅ Keep instructions general, prompts task-specific

❌ Don't make instructions too verbose
✅ Keep them concise and actionable

❌ Don't create too many prompts
✅ Start with 5-10 core workflows, expand as needed

---

Resources

• GitHub Copilot Workspace Documentation
• Prompt Engineering Best Practices

---

Hope this provides a more complete picture of how to architect your Copilot workflow! Let me know if you need clarification on any specific pattern.

Best regards,
Asyrafi

[yorkeChen]

@Dinuka-Nonis @4syr
Thank you both for the detailed explanations. This resolved my issue. Really appreciate your support.