Create Custom AI Agents in GitHub Copilot for QA,DEV & CI/CD 🤖

Overview

This summary describes how to create custom AI agents (GitHub Copilot chat modes) for QA, development, and CI/CD by authoring agents.md files in a repository. It covers core concepts, demonstrated features, a step-by-step demo/tutorial, benefits, and references.

Core concepts and technology

• GitHub Copilot / VS Code chat modes let you create specialized AI agents by adding simple Markdown files (for example, agents.md) into your repo’s chat modes folder. Built-in modes include: ask, edit, and agent.
• agents.md is an open, standardized format used by many projects and companies to guide coding agents. A central resource with examples and guidance is agentsmd.net.
• Agents are effectively prompt/command bundles plus tool declarations that give an LLM (e.g., custom models referenced in demos) context and tooling to perform specific tasks.

Features demonstrated

• Reproducing Playwright-style agents (planner, generator, healer) by authoring agents.md files that encapsulate:
  • Planning workflows
  • Test generation
  • Test healing
• Creating domain-specific agents for:
  • Business writing and scaffolding
  • Development (unit test generation and healing)
  • Testing / QA and UI/UX
  • CI/CD pipeline architecture
• Tool declarations and role instructions inside the agent file allow the agent to:
  • Analyze project structure
  • Generate optimized GitHub Actions workflow files tailored to language/stack (.NET, Java, Node, Python, Go, etc.)
  • Create multi-environment deployment workflows (dev/staging/production), matrix builds, and security scans
  • Generate Dependabot configurations
  • Detect Dockerfiles / docker-compose and adapt outputs accordingly
• No-code, prompt-driven agent creation: engineers can distribute specialized agents across teams so nontechnical users can execute complex tasks with single prompts.

Demo / tutorial steps

1. Read VS Code / GitHub Copilot docs on chat modes and the agents.md spec (agentsmd.net).
2. Create an agents.md (chat mode) Markdown file using one of the sample files as a template.
3. Use the provided deployer script or CLI to install chat modes into your project. Example:
   • npx copilot setup --testing (other options: --development, --all)
   • This adds files into your repo’s chat_modes (GitHub / VS Code) area.
4. Invoke the CI/CD specialized agent via Copilot chat mode in VS Code. Example prompt: “generate workflow for me” and point the agent at your target project/file. The agent analyzes the repo and produces YAML files such as:
   • ci.yml
   • deployment.staging.yml
   • deployment.production.yml
5. Inspect and use the generated GitHub Actions workflows (note: examples produce GitHub Actions YAML, not Azure pipeline YAML).

Benefits and takeaways

• Rapid, repeatable generation of infrastructure and configuration (CI/CD workflows), tests, and other developer artifacts with minimal manual coding.
• The standardized agents.md format enables interoperability across projects and teams.
• Organizations can create strict, tool-specific agents that let non-experts perform complex tasks safely and consistently.
• Demonstrated practical value: the CI/CD agent auto-detected Docker artifacts and produced tailored multi-environment workflows.

References and guides

• VS Code / GitHub Copilot documentation on using chat modes in VS Code
• agents.md specification and examples: agentsmd.net (includes ~14 sample agent Markdown files, including Playwright end-to-end testing examples)
• Playwright agent examples: planner, generator, healer
• npx copilot setup CLI for installing chat modes

Main speakers and sources

• Video presenter (creator of the demo and narrator)
• Referenced projects and documentation:
  • Playwright
  • GitHub Copilot / Visual Studio Code chat modes documentation
  • agents.md / agentsmd.net