The Prompt Add-on That Saves My Code Reviews

If you and your team are using tools like Claude-Code, Antigravity, or GitHub Copilot, you’ve likely encountered that awkward moment during a code review. You submit a PR, and a teammate comments: "Did the AI write this? It’s super long/unreadable/bloated."

AI is incredibly powerful, but its default "personality" often leans toward verbosity and generic patterns that don't always align with your specific team standards.

To solve this, I started using a specific "Prompt Add-on" that forces the AI to look at local project context before it writes a single line of code. This simple habit has been a lifesaver for keeping our codebase maintainable, readable, and scalable—regardless of whether we are working in JavaScript, TypeScript, Python, or Go.

The "Secret Sauce" Prompt

Whenever I start a new task or a complex refactor, I append this instruction to my request:

"I want you to read the [Gemini.md / Claude.md / Copilot.md] file that's in the project root, and I want you to follow the prompt thoroughly, making sure you follow the prompt step-by-step to build out the software that I want you to build. Also, when you do build the software, I also want you to look at the AI-Rules.md and brand-guidelines.md files that are also in the project to make sure that you get the guidelines right for the software that you're building. Remember that the guidelines are designed to keep the codebase maintainable, readable, and scalable. The core philosophy is: Write code for humans first, machines second."

Why This Works

1. The "Persona" File (Claude.md, Gemini.md, etc.)

By pointing the tool to a file named after itself in the root directory, you can define exactly how you want that specific model to behave. For example:

• TypeScript: "Never use any; always prefer interfaces over types."
• Python: "Follow PEP 8 strictly and use type hints."
• Logic: "If a function is longer than 20 lines, break it down."

2. The AI-Rules.md File

This is where you store your "Anti-AI-Bloat" rules. While the persona file is about style, the rules file is about architecture. It reminds the AI of your team's specific choices (e.g., "We use Tailwind, not CSS-in-JS" or "We use FastAPI, not Flask").

3. The Human-First Philosophy

The most important part of the prompt is the final sentence: "Write code for humans first, machines second." When an AI knows it is being judged on readability and maintainability rather than just "making it work," the output changes instantly. It stops over-engineering and starts writing the clean, concise code that your senior devs actually want to see.

How to Implement This Today

1. Create the files: Add a .md file for your AI tool of choice and an AI-Rules.md to your repository root.
2. Define your "Banned" patterns: List the things that usually trigger "AI-code" red flags in your reviews.
3. Always link them: Use the prompt add-on above as a snippet or a shortcut.

By forcing the AI to acknowledge these local "instruction manuals," you turn a generic code generator into a specialized member of your team who knows your standards inside and out.

Happy coding—for humans!