Skip to content
Prompt Coach

How to write Claude Code prompts when you can't code

Updated:

Claude Code is Anthropic’s coding assistant that works directly on your computer. It reads and writes files, runs commands and can build an entire project. That makes it powerful, and it changes how you should ask for things. It isn’t a chat window handing you text; it’s a collaborator with its hands on your files.

This guide is for people who don’t code (or code a little) and want good results without getting lost.

What Claude Code needs to know

Where the project stands

New or existing? What’s in the folder? If you’re not sure, have it find out: “Explore this folder and summarize what’s here before changing anything.”

The outcome, not the technology

You don’t need to pick a framework. Describe what should happen: “When someone taps the button, open a text message to my number with ‘I’d like to book’ pre-filled.” If you have technical preferences (free hosting, no frameworks), list them as constraints.

Your skill level

“I can’t code” is incredibly useful information. It changes how Claude Code explains things and what it asks you to do by hand.

Ask for a plan before code

For anything beyond a tiny change, this sentence saves hours:

Propose a plan first, and wait for my approval before editing files.

You get to correct course before a dozen files exist. Claude Code also has a plan mode built for exactly this: it researches and proposes without modifying anything.

Define how to verify

A task is done when it’s been checked, not when the AI says it’s done. Make verification concrete:

  • “Run the project and confirm it starts without errors.”
  • “Run the tests and show me the output.”
  • “Tell me exactly what I need to do to see it working in my browser.”

Mark what’s off-limits

If some parts already work, say so: “Don’t touch the contact form, it works.” And require confirmation before anything hard to undo: deleting files, deploying, pushing to a shared repository.

When something breaks

“It’s broken, fix it” forces guesswork. Give three facts:

Expected: after clicking "Send," see the thank-you message.
Actual: the page reloads and nothing appears.
Error (from the browser console): Uncaught TypeError: form is null

Rules you repeat every time: CLAUDE.md

If you keep saying the same things each session (“explain in plain English,” “don’t install anything without asking”), put them in a CLAUDE.md file in the project folder. Claude Code reads it at the start, so you don’t have to repeat yourself.

A starter template

<context>
[What the project is, new or existing, your skill level.]
</context>

<task>
[What should exist or work at the end.]
</task>

<constraints>
[What not to touch, technical preferences, budget.]
</constraints>

<how_to_work>
1. Explore the project and summarize what you find.
2. Propose a plan and wait for my approval.
3. Make small changes and verify each step.
4. Ask before deleting files, deploying or pushing.
</how_to_work>

<success_criteria>
[How you'll check it's right.]
</success_criteria>

Pick “Claude Code” as the target in the coach and the prompt comes out with this way of working built in. For a full example, see the dog-grooming site in before and after.

Keep learning