← Blog

May 23, 2026

Your AI Build Spec Works With Cursor, Codex, Windsurf, and More — Here's How

When you generate a build spec on Nicheloom, the file is called CLAUDE.md — named after Claude Code, the AI coding tool it was originally designed for.

But the name is misleading.

The file is just structured markdown: a product spec, tech stack, architecture decisions, database schema, API routes, and a prioritised feature list. There's nothing in it that's specific to Claude. You can hand it to any AI coding tool and get the same head start.

This post shows you exactly how to do that — tool by tool.

What's in the build spec

Before we get into each tool, here's what the file actually contains:

  • Product overview — what the app does, who it's for, the core value prop
  • Tech stack — recommended framework, database, auth, payments, and hosting
  • Architecture — how the pieces fit together, key data models
  • Feature list — prioritised into MVP, v1, and future scope
  • API routes — the endpoints you need to build
  • UI screens — the pages and what goes on each one

Think of it as the brief you'd give a senior developer on day one. Except your developer is an AI that can start writing code in the next five minutes.

Claude Code

This is the native case. Claude Code reads CLAUDE.md automatically when it exists in the root of your project — no setup required.

How to use it:

  1. Create a new project folder
  2. Drop the CLAUDE.md file in the root
  3. Open Claude Code in that directory
  4. Start with: "Read the CLAUDE.md file and give me a plan for what to build first"

Claude will read the spec, ask clarifying questions if needed, and start generating code that matches the architecture you've already defined. Because the context is loaded upfront, you skip the back-and-forth of explaining what you're building every session.

Best for: Developers comfortable with the terminal who want the most seamless integration.

Cursor

Cursor doesn't have a native CLAUDE.md convention, but it has two places where you can load your spec as persistent context.

Option 1 — .cursorrules file (project-wide)

  1. Rename or copy your build spec to .cursorrules in the root of your project
  2. Cursor reads this file automatically and applies it to every Composer and Chat session in that project

This is the cleanest approach — the spec is always in context without you having to paste it each time.

Option 2 — Paste into Composer

  1. Open Cursor's Composer (Cmd+I on Mac)
  2. Paste the full contents of your build spec at the start of your first message
  3. Follow it with: "This is my product spec. Start by scaffolding the project structure and the database schema."

Best for: Developers who prefer a GUI and want inline code suggestions as they type.

OpenAI Codex

Codex (OpenAI's coding agent) works through context provided at the start of a session. The build spec drops in perfectly.

How to use it:

  1. Start a new Codex session
  2. In your first message, paste the full build spec followed by your first task: "Here is my product spec: [paste spec]. Start by setting up the project with the recommended tech stack and creating the database schema."
  3. Codex will scaffold the project and you can iterate from there

Tip: Break the spec into sections if the context window feels tight. Start with the tech stack and data models, then bring in the feature list once the foundation is in place.

Best for: Developers already in the OpenAI ecosystem or using the API directly.

Windsurf (Codeium)

Windsurf uses a context system called Cascade. You can give it persistent instructions via the system prompt or by referencing a file.

How to use it:

  1. Open Windsurf and start a new Cascade session
  2. Use the @file reference to point Cascade at your build spec: "@CLAUDE.md — read this spec and help me build this product. Start with the project setup and database schema."
  3. Cascade will keep the spec in context throughout the session

Alternatively, paste the spec directly into the chat if you prefer not to use file references.

Best for: Developers who like a polished IDE experience with strong autocomplete alongside the agent.

GitHub Copilot (with Copilot Chat)

GitHub Copilot has a custom instructions file that works similarly to .cursorrules.

How to use it:

  1. In your project root, create .github/copilot-instructions.md
  2. Paste the contents of your build spec into that file
  3. Copilot Chat will use it as persistent context for all conversations in that workspace

Then open Copilot Chat and start with: "Based on the project instructions, what should I build first?"

Best for: Developers already using GitHub and VS Code who don't want to switch tools.

Gemini CLI / Google AI Studio

Google's Gemini CLI reads a GEMINI.md file from the project root — almost identical to the CLAUDE.md convention.

How to use it:

  1. Rename your build spec to GEMINI.md (or keep both files with the same content)
  2. Open Gemini CLI in your project directory
  3. Gemini will read the file automatically and use it as project context

Best for: Developers experimenting with Google's models or using the Gemini API for their project.

The tool-agnostic approach (any AI)

If you're using a tool not listed here — or just want a universal method that works everywhere — this always works:

  1. Open a new chat with any AI coding assistant
  2. Start with: "You are a senior full-stack developer. Here is the product spec for the app I want to build: [paste spec]. Do not start coding yet — first tell me your plan for what to build in what order, and flag any decisions I need to make."
  3. Review the plan, make any adjustments, then say: "Great. Now start with step 1."

The planning step is important. It forces the AI to process the full spec before writing a single line, which leads to better architecture decisions and fewer mid-build surprises.

Why the spec matters more than the tool

Here's the honest truth: the AI coding tool you use matters less than the quality of the context you give it.

A vague prompt like "build me a SaaS app for homeschoolers" will produce generic, half-finished code — in Claude, Cursor, or anywhere else.

A detailed spec with a clear tech stack, data models, and prioritised features produces code you can actually ship. Because the AI isn't guessing what you want. It's executing a plan.

That's what Nicheloom builds for you. Not just an idea — a validated idea with a ready-made spec that any AI coding tool can act on immediately.

Browse validated ideas and generate your build spec at nicheloom.com.