Loop/Prompt Library / Prompts from builders on X
💬 AI prompt: Before you say 'done': try to break it
You want the AI to check its own code for bugs and security holes before it reports success.
Copy this prompt
Add these rules to [instructions file], show me the change, and wait for my OK before saving: Before saying a task is done: 1) assume your first answer is wrong and look for evidence it isn't, 2) search the changed code for passwords or keys written into files, unchecked user input, and injection risks, 3) run the tests and the type checker, 4) write one test designed to break the new code, 5) report what you checked and what you didn't. If any check fails, fix it and run all checks again, at most [3] times, then stop and tell me what's still failing.
How to use it
- Open Claude (or Claude Code in your project folder).
- Paste the prompt and fill in every [bracket].
- Read what it changed or wrote before you keep it.
Not medical, legal or financial advice. The AI drafts; you decide and you press send.
Idea from @imcaiden on X — prompt rewritten by FixKit AI →
More prompts like this
Credit report check-up
You're about to apply for a loan or card and want to know what lenders will see.
Build a simple monthly budget sheet
You want a spending tracker that adds itself up, without learning spreadsheets.
Check a big medical bill before you pay
A hospital or doctor bill is big or confusing and you want to know what to question before paying.
Monthly tax check-in
You want to plan for taxes during the year instead of scrambling at filing time.
Make the AI show what it isn't sure of
You're about to rely on or publish something an AI wrote.
Résumé bullets with real numbers
You want a stronger résumé or LinkedIn profile without the AI making things up.