Top prompts for fixing vibe-coded apps

So you have been vibe coding. The AI built most of the app. Now something is broken, and you are staring at code you did not write.

These prompts are designed for that moment. Copy one, fill in the brackets, and paste the relevant error, code, or screenshots.

1. Explain this like I am not a programmer

Prompt

Explain what this code does in plain English. Assume I have never coded before.
Here is the code:
[paste code]  
Also tell me what inputs it expects and what output it produces.

Use this when you do not know what you are looking at yet.

2. What does this error mean?

Prompt

I am seeing this error:
[paste full error and stack trace]  
Explain what it means in plain English. List the 3 most likely causes in my situation and how to confirm each one.

Tip: include the first error in the console, not only the last one. The first error is often the real cause.

3. Why is it not doing what I expected?

Prompt

Expected: [what should happen]
Actual: [what happens instead]
Steps to reproduce: [1, 2, 3]
Relevant code:
[paste code]  
What are the most likely reasons for the mismatch? Give me a short checklist to narrow it down.

This is the fastest way to turn “it is broken” into something debuggable.

4. Find the problem in this code

Prompt

Something is wrong with my [button/form/page/API call]. Here is the code:
[paste code]  
Find likely bugs, edge cases, or missing wiring. For each issue, show the exact change you would make.

If the AI suggests large refactors, ask it to propose the smallest change first.

5. What assumptions does this code make?

Prompt

What does this code assume is true for it to work?
Examples: required fields, auth state, network availability, non-null data, timing.
Here is the code:
[paste code]  
Then tell me which assumptions are most likely false given this symptom: [describe symptom].

Most “random” bugs are violated assumptions.

6. Ask me questions until we find it

Prompt

I am stuck. I will describe what I see, and you ask me one question at a time until we find the cause.
Start by asking for the minimum info you need.
Symptom: [describe]
Environment: [browser, OS, device, local vs deployed]
Error (if any): [paste]

This is useful when you do not know what to paste yet.

7. What changed between working and broken?

Prompt

This worked before and now it does not.
Old version (worked):
[paste]  
New version (broken):
[paste]  
Identify the meaningful differences and rank them by likelihood of causing the bug. Suggest a minimal fix.

If you do not have the old version, use your git history or any saved export.

8. What could break this in real use?

Prompt

Here is the feature and code:
Feature: [what it does]
Code:
[paste]  
List unusual but realistic cases that could break it (empty input, double clicks, slow network, weird characters, time zones, permissions). For each case, tell me how to reproduce and how to guard against it.

This is how you find bugs before users do.

9. Will my fix work, and what could it break?

Prompt

I think the cause is: [your guess]
I want to fix it by: [your plan]
Here is the relevant code:
[paste]  
Evaluate this fix. What could it break? Propose a safer alternative if needed. Include a quick test plan.

This saves you from “fixing” one thing and breaking three others.

10. How do I stop this from happening again?

Prompt

We fixed the bug: [describe the fix].
How do I prevent this class of bug in the future?
Suggest: validation, better error messages, logging, tests, and UI safeguards.
Keep changes small and show code examples where possible.

Ask for a minimal “guardrail” first, then add tests if you have time.

Tips for better debugging conversations

The full error message and stack trace
The smallest code snippet that still reproduces the bug
Steps to reproduce, starting from a fresh reload
What you expected vs what happened

Ask for minimal changes

If the AI proposes rewriting multiple files, ask:

Can you fix this with the smallest possible diff? Show only the lines that change.

Get a test plan every time

Add this line to any prompt:

After the fix, give me 5 manual checks to confirm it is solved and nothing else broke.

Next step

Pick one broken behavior and run prompts #2 and #3 with the exact error and reproduction steps. That usually narrows the problem to one file and one change.