Prompt library
Vibe coding prompts: a template library from first idea to launch
Updated November 17, 20268 min read
The best vibe coding prompts say who you are, what you want, what "done" looks like, and ask the AI to work in small steps. This page is a library of copy-paste prompt templates sorted by stage: planning, building, fixing bugs, design changes, security and launch. Each one comes with a line on when to use it, and they work in Claude Code, Codex, Gemini CLI, Lovable, Bolt, Replit and other tools for building apps with AI.
How to use this prompt library
Anything in [square brackets] is a gap for you to fill in. Delete the brackets when you do. Keep the rest of the wording, because each line is there for a reason: it stops the AI from guessing, from doing too much at once, or from hiding what it changed.
Use one template per message. If a reply is long or confusing, answer with "explain that in three short sentences" before you move on. And when a prompt works especially well for your project, save your filled-in version, since you will need it again next week.
Planning prompts: before any code exists
Turn an idea into a plan
When to use it: the very first message of a new project.
I'm not a developer. I want to build: [one-sentence idea].
It's for: [who will use it].
The main thing they do is: [main action].
Not in version one: [features to leave out].
Don't write code yet. Suggest the simplest way to build this,
list the steps in order (smallest first) and ask me any
questions you need answered. Use plain English.Compare options before choosing
When to use it: you face a choice you don't understand, like which database or login service to use.
I need to [goal]. Give me up to 3 ways to do it.
For each one: how hard it is, what it might cost,
and its biggest downside. Recommend one and say why.
Don't build anything yet.Write the project notes file
When to use it: once the plan is agreed, so every future chat starts with the same facts.
Create a short notes file the AI reads at the start of
every chat (CLAUDE.md for Claude Code, AGENTS.md for Codex).
Include: what the app does, who it's for, how to start it,
and these rules: explain things simply, work in small steps,
save a checkpoint after each working step, never put
secret keys in the code.Building prompts: one feature at a time
Build a single step
When to use it: working through the plan, one step per message.
Build step [number] only: [describe it].
When you finish, tell me how to open it, exactly what
to click to check it works, and anything you weren't sure
about. Then stop and wait for me.Add a feature to a working app
When to use it: the app already works and you want something new.
Add this feature: [describe it from the user's side].
Done means: [what I should see when it works].
Change as little as possible and leave unrelated parts
alone. List every file you changed when you finish.Save a checkpoint
When to use it: every time something works, before you ask for anything else.
That works. Save a checkpoint now with git and write
a short note saying what is working at this point.Explain what you just built
When to use it: you want to understand your own app a little better.
Explain how the [feature] works as if I'm 12.
Where is the data stored, what happens when I click
[button], and what could go wrong with it?Prompts for fixing bugs without making it worse
Report a bug the AI can act on
When to use it: something doesn't work the way you expected.
Bug:
What I did: [the steps]
What I expected: [expected result]
What happened: [actual result]
Error message: [paste all of it, or say none]
Find the cause before changing anything. Explain the
cause in plain English, then fix only that.Break out of a fix loop
When to use it: the AI has tried to fix the same thing two or three times and it still fails.
Stop making changes. We've tried this several times.
Go back to the last working checkpoint. Then tell me what
you think is really wrong and suggest a different approach
before you try again.Translate an error message
When to use it: you see red text and have no idea what it means.
Explain this error like I'm not a programmer:
[paste error]
What does it mean, what likely caused it, and is it
serious or safe to ignore?Design change prompts for how the app looks
Change one thing on one page
When to use it: a button, colour or layout bothers you.
On the [page name] page, [describe the change].
For example: make the Sign up button bigger and green,
and keep it visible on small phones.
Check it at phone width and at desktop width.Match a style you like
When to use it: the app works but looks plain, and you know the feel you want.
Restyle the app to feel [calm and simple / bold and fun].
Use at most 2 main colours and one font. Don't change
how anything works, only how it looks. Show me the
home page first and wait before doing the others.Fix a phone layout
When to use it: it looks fine on a laptop but broken on a phone.
On phones (about 375 pixels wide) the [page] has these
problems: [list them]. Fix the layout for small screens
without changing the desktop version.Security prompts to run before other people sign up
Look for exposed secrets
When to use it: before you publish, and after adding any paid service. A .env file is a private settings file that stays on your computer and is never uploaded.
Search the whole project for secret keys, passwords and
tokens. Tell me if any are in code that runs in the
browser or would be uploaded to GitHub. Move them to a
private .env file and make sure .env is in .gitignore.Check users can't see each other's data
When to use it: any app with logins. An API route, mentioned in the prompt, is a hidden address your app's pages use to fetch and save data.
Check that every page and every API route only returns
data that belongs to the logged-in user. Look for places
where changing an ID in the address bar would show someone
else's data. List problems worst first. Don't fix yet.For the full list of risks and what each one means, see our vibe coding security guide.
Launch prompts for the last day
Pre-launch review
When to use it: the day before you share the link.
We're about to launch. Review the app for: bugs that could
lose data, slow pages, broken links, missing error messages,
and anything a first-time visitor would find confusing.
List problems worst first in plain English.Write the help and privacy text
When to use it: you need a short FAQ or privacy page that matches what the app really does.
Write a short privacy page for this app. Base it only on
what the code really does: what data we collect, where it's
stored and who can see it. Flag anything I should check
with a professional.Prompts for Claude Code and other coding agents
All of the templates above work as prompts for Claude Code. A few extra habits help with coding agents that run in a terminal:
- Run
/initin Claude Code to have it write a startingCLAUDE.mdnotes file for your project, then edit it with the rules from the planning section. - Use plan mode for the planning prompts, so nothing changes until you approve.
- Codex reads
AGENTS.mdand Gemini CLI readsGEMINI.mdfor standing rules. See our Codex CLI and Gemini CLI guides. - Rules you repeat in every message belong in the notes file. Then your daily prompts can stay short.
Anatomy of a good vibe coding prompt
Every template above is built from the same few parts. Once you know them you can write your own.
- Who you are
- "I'm not a developer" makes the AI explain more and assume less.
- The goal
- What you want, described from the user's side, not in technical words.
- Done means
- What you will see when it works. This gives the AI a finish line.
- Limits
- What not to touch, what to leave out, how big the change should be.
- The real evidence
- The full error message or the exact steps, never "it's broken".
- Stop point
- "Then stop and wait for me" keeps you in charge of each step.
- Output format
- "List problems worst first" or "list every file you changed" makes the answer easy to check.
Weak prompts and stronger rewrites
| Weak prompt | Why it fails | Stronger version |
|---|---|---|
| Build me an app like Uber. | Far too big. The AI guesses hundreds of details. | Plan a page where a customer books one ride. Ask me questions first. |
| It doesn't work. | The AI can't see your screen. | I clicked Save. I expected a message. The page went blank. Here is the error. |
| Fix everything. | Leads to big changes that break other parts. | Fix only the blank page after Save. Don't touch anything else. |
| Make it look better. | Nobody knows what better means. | Make the header smaller and the buttons bigger on phones. |
| Here's my API key: sk-... | Your secret is now in the chat history. | My key is in .env as PAYMENTS_KEY. Use that name. |
| Add login and payments and emails. | Three risky features at once, hard to test. | Add email login only. Stop when it works. |
Prompting questions
What are the best prompts for vibe coding?
+
The ones that ask for a plan first, build one small step at a time, and give the AI a clear finish line. Start with the planning template above and use the bug report template whenever something breaks.
Do these prompts work in Lovable, Bolt and Claude Code?
+
Yes. They are plain English, so they work in browser builders and coding agents alike. Only the name of the notes file changes from tool to tool.
How long should a vibe coding prompt be?
+
Usually 3 to 10 lines. Long enough to say what you want, what done looks like and what to leave alone. Very long prompts that ask for many things at once tend to get messy results.
Sources
Run up to 12 AI coding agents in one window.
Claude Code, Codex, Gemini, Grok and more, side by side. Windows and Mac. From $29 a month.
Get startedKeep reading
Weekend tutorial
How to vibe code: build your first real app in one weekend
Habits and rules
Vibe coding best practices: the 10 mistakes that sink AI-built apps
24 numbered tips
Claude Code tips and tricks: 24 habits that pay off
Before and after
Claude Code plan mode: one task, with and without a plan
Vibe coding
The best vibe coding app depends on which of two roads you take