From Copilot to Co‑Engineer: How I Supercharged GitHub Copilot for My DevOps Projects 🚀

---

🚀 How It Started — And Why I Wanted More

When GitHub Copilot first landed in my VS Code, I felt like I had a superpower. It could autocomplete Terraform blocks, Next.js API routes, even Dockerfiles. But after a few weeks, the shine dulled.

I realized Copilot didn’t really know my projects. It guessed, sometimes well, sometimes completely off. It forgot my naming conventions, ignored my AWS stack details, and occasionally proposed changes that would break my CI/CD.

That’s when I thought: If Copilot is my co‑pilot, I need to teach it how to fly my plane.

---

🧭 My Journey Through AI Coding Tools

Before arriving at Copilot Pro, I tried Cursor Pro during its free trial — and wow, it was a game changer. Using Sonnet 4, Cursor felt incredibly powerful for my cloud learning work.

Around the same time, I also explored Claude Code CLI with Anthropic’s $10 trial. I had no idea how quickly usage could add up, and my balance vanished in under 30 minutes. But what I loved about Claude Code CLI was its think → plan → execute workflow — starting with a to‑do list, then methodically checking items off as each step was completed.

After my Cursor trial expired, I tested Copilot Pro’s free trial. It was good, but I missed the structured workflow of Claude and Cursor. Copilot felt like it needed a nudge to match that level of orchestration.

That’s when I found the Awesome Copilot repo — and realized I could recreate much of the Claude/Cursor magic inside Copilot itself.

---

🔍 The Discovery — Finding "Awesome Copilot"

Late‑night search: “Make GitHub Copilot smarter.”

I found Georg.dev’s blog about copilot-instructions.md. Then Burke Holland’s Beast Mode gist. And finally, the treasure chest: Awesome Copilot.

At first, I made my rookie mistake — I used Copilot Chat in VS Code to “download the files” directly from Awesome Copilot. It grabbed them, but everything came with .md extensions that weren’t recognized. None of the modes or prompts worked.

---

🛠️ The Fix — Setting It Up Correctly

The breakthrough came when I checked the official Copilot documentation. I realized:

• Chat modes need .chatmode.md

• Prompts need .prompt.md

• Instructions need .instructions.md

I reorganized everything into a custom folder structure:

.github/
  instructions/
    project.instructions.md
  chatmodes/
    BeastMode.chatmode.md
  prompts/
    terraform.prompt.md
    ecs.prompt.md

Once renamed and restructured, the magic happened. Copilot Chat started acting like Claude Code CLI — with sub‑agent behavior and more structured problem‑solving.

---

📊 Side-by-Side Comparison

---

📈 The Impact — Copilot, but Smarter

• Onboarding: New contributors get Copilot suggestions aligned with our architecture.

• Productivity: Less time repeating context.

• Code Quality: Suggestions respect Terraform modularity and AWS best practices.

• Collaboration: Chatmodes handle multi‑step changes confidently.

Before, Copilot was a smart intern. Now, it operates like a capable co‑engineer.

---

💡 Lessons Learned

1. Extensions matter — .chatmode.md and .prompt.md unlock proper functionality.

2. Context is everything — instructions turn Copilot from guessing to knowing.

3. Agent modes amplify Copilot Chat — perfect for multi‑step DevOps workflows.

4. Keep tweaking — I’ll be experimenting with using gemini-2.5-pro in Claude CLI for hybrid workflows.

---

🔮 What’s Next — And Why You Should Try This

If you’re using Copilot in any serious project — don’t just install it, train it. Give it folders, instructions, prompts. You might just recreate a Claude/Cursor‑like experience inside VS Code.

I’m excited to continue refining my setup — and yes, I’ll also be testing Gemini‑2.5‑Pro with Claude Code CLI to see how far I can push this workflow.

Happy coding!

---

👉 Over to you: Have you tried customizing Copilot? Share your experiments or questions in the comments. Let’s make our AI teammates as sharp as our human ones.