Create your first Agent using Copilot and Repomix

Author: Aridane Martín
Tech Lead – Pandora Project | Software Developer
Conference: OpenCommit Fest

---

Creating AI agents that truly understand your codebase is easier than you think. This guide will show you how to use Repomix and GitHub Copilot to build intelligent assistants for your development team.

What is Repomix?

Repomix is a powerful tool that creates a single, comprehensive file containing your entire codebase. It intelligently combines all your source code, configuration files, and documentation into one cohesive document. This will help us feed our agent with the full context of our project.

Key Benefits:

• Complete Context: AI sees your entire project structure
• Intelligent Filtering: Excludes unnecessary files (node_modules, logs, etc.)
• Smart Organization: Maintains file relationships and dependencies
• Size Optimization: Configurable limits to stay within AI context windows

Step 1: Compress your codebase into a single file

First, you’ll need to install Repomix and generate a comprehensive overview of your codebase:

# Install Repomix globally
npm install -g repomix

# Generate a comprehensive codebase file
repomix --output codebase-overview.md

          
          Copy
        

This command will create a single markdown file containing your entire codebase, intelligently organized and filtered to include only relevant files.

Step 2: Create your agent with full context

Now that you have a comprehensive overview of your codebase, you can create an AI agent that truly understands your project. Here’s how to set it up:

Example: Onboarding Mentor Agent

// Example Prompt for GitHub Copilot:
Create a new agent that will be mentoring new onboarding developers in our company. The agent should be able to answer questions about the codebase and the project and when you ask him to generate code, he should generate code that is compliant with our codebase and our project. Every time you ask him to generate code he should explain a bit about the conventions we used in the project defined in our copilot-instructions.md file.

Here's our complete codebase context:
[Paste the content from your codebase-overview.md file here]

          
          Copy
        

Example: Code Review Agent

// Example Prompt for Code Review Agent:
Create an agent that specializes in code reviews for our project. The agent should:
- Understand our coding conventions and patterns
- Identify potential bugs and improvements
- Suggest optimizations based on our existing codebase
- Explain architectural decisions made in our project

Codebase context:
[Include your codebase overview here]

          
          Copy
        

Step 3: Customize your agent’s behavior

Once you have your agent set up, you can customize its behavior by:

1. Defining specific roles: Make it clear what the agent should and shouldn’t do
2. Setting tone and style: Define how the agent should communicate
3. Adding project-specific knowledge: Include your team’s conventions and best practices
4. Creating specialized agents: Build different agents for different purposes (mentoring, code review, documentation, etc.)

Advanced Tips

Context Management

• Use Repomix’s filtering options to focus on relevant files
• Set size limits to stay within AI context windows
• Update your codebase overview regularly as your project evolves

Agent Specialization

• Create multiple agents for different purposes
• Use specific prompts to define each agent’s expertise
• Test your agents with real scenarios from your development workflow

Team Integration

• Share agent configurations with your team
• Create documentation for how to use each agent
• Gather feedback and iterate on agent behavior

Wrap Up

Creating AI agents that understand your codebase is a game-changer for development teams. By using Repomix to provide complete context and GitHub Copilot to create specialized agents, you can build intelligent assistants that:

• Understand your project: Full codebase awareness
• Follow your conventions: Consistent with your team’s style
• Scale knowledge: Help onboard new developers quickly
• Improve code quality: Provide intelligent code reviews and suggestions

🚀 The goal: Make AI work for your entire team by creating agents that truly understand your codebase and development practices.

---

© Aridane Martín – 2025