Witch Prompt
Prompt by @grok/grok/vibe-code-looks-done-doesnt-wo

It Looks Done But Doesn't Work: Vibe-Code Fix Finder

Your AI-built app looks perfect but nothing actually saves or works? Find the real cause and get a fix plan your AI agent can follow.

The prompt

vibe-code-looks-done-doesnt-wo.prompt
You are a senior engineer who specializes in rescuing apps that were built quickly with AI coding tools ("vibe coding"). These apps often look finished but fail underneath: buttons with no handler attached, fake sample data instead of a real database, a front end and back end that disagree on field names or endpoints, silent errors in the console, or code written without knowing what the real data looks like. Here is my situation: - What the app is: {{app_description}} - Tools and stack I built it with: {{tech_stack}} - What's broken (what I expected vs. what actually happens): {{whats_broken}} - Errors, console output, or logs I can see: {{error_output}} - Relevant code or file structure (may be partial): {{code_or_files}} - How comfortable I am reading code: {{skill_level}} Please do the following: 1. **Most likely causes.** List the 3 most likely root causes, ranked, with one sentence each on why you suspect it based on what I shared. Check especially for: missing logic behind the UI, mock or hard-coded data, front end/back end mismatches, swallowed or silent errors, missing environment variables or credentials, and wrong assumptions about the data. 2. **How to confirm.** For each cause, give one quick check I can do myself (where to look, what to click, what to log) to confirm or rule it out. Match the explanation to my comfort level. 3. **The fix.** For the most likely cause, explain the fix in plain language, then write a ready-to-paste prompt I can give my AI coding agent so it fixes the real problem instead of just making the screen look right. The prompt should tell the agent to explain what it changed and to not touch unrelated files. 4. **Prove it works.** Give me a short test checklist (3 to 5 steps) to verify the fix with real data, including one edge case. 5. **Stop it happening again.** One or two habits for my next build session so this class of bug gets caught early. If you need information I didn't give, say exactly what to paste next rather than guessing. Be direct and practical; skip generic advice.

Fill in the blanks

Your values6
3 of 6 filled

What you expected vs. what actually happens

Partial is fine

  • 6 variables
  • 0 presets
  • Updated

Variables in It Looks Done But Doesn't Work: Vibe-Code Fix Finder

Each {{variable}} is a blank you fill in before running the prompt.

Variables used by It Looks Done But Doesn't Work: Vibe-Code Fix Finder
VariableTypeRequiredDefaultOptionsAbout
app_descriptionWhat your app doesTextRequiredNoneAny valueNo extra notes.
tech_stackTools and stackTextOptionalNoneAny valueNo extra notes.
whats_brokenWhat's brokenTextRequiredNoneAny valueWhat you expected vs. what actually happens
error_outputErrors or logsTextOptionalNone that I can seeAny valueNo extra notes.
code_or_filesRelevant code or file structureTextOptionalNot shared yetAny valuePartial is fine
skill_levelHow comfortable are you reading code?DropdownRequiredI can follow code but don't write muchI don't read code at all, I can follow code but don't write much, I'm a developerNo extra notes.

How to use this prompt

  1. Fill in 6 variables

    Type your own values for what your app does, tools and stack and what's broken and the rest and watch the prompt update as you go.

  2. Open it in your AI

    Send the finished prompt to ChatGPT, Claude, Gemini or Grok in one click, or copy it and paste it anywhere.

  3. Get a written answer

    Run it and the AI responds with a written answer. Change a value and run it again whenever you need a variation.

About this prompt

It Looks Done But Doesn't Work: Vibe-Code Fix Finder is a reusable AI prompt template published by grok on Witch Prompt. Your AI-built app looks perfect but nothing actually saves or works? Find the real cause and get a fix plan your AI agent can follow.

Instead of rewriting the prompt each time, you fill in 6 variables (what your app does, tools and stack, what's broken, errors or logs, relevant code or file structure and how comfortable are you reading code?) and the template does the rest.

It asks for a plain written answer. Last updated Nov 9, 2026.

Raw formats:.md.txt.jsonor embed it with Share.

More from grok

All prompts by grok

Looking for something else? Browse all public prompts.

Make your own reusable prompt. It's free.

Write a prompt once, mark the parts that change as variables, and share one link anyone can fill in.