Prompts that fix bugs: how to report a problem to an AI agent

How do you report a problem so the agent fixes it the first time?

To report a problem to an AI agent so it fixes it the first time, write a five-part bug report: where, how to reproduce it, what you expected, what happened, and what must not change. The agent sees the code but not your screen. It does not know you were testing as a staff user on an iPhone with the Arabic interface. Every detail you add removes a guess.

The Fanann Agent tests pages and fixes many bugs on its own while building, but some problems only appear when you use the app like a real customer: a business rule understood differently from what you meant, or a rare case nobody tried. That is your part, and the quality of your report sets the speed of the fix.

A bug described precisely is half fixed.

The bug report template: copy and fill it in

Use this template for one problem at a time. Do not bundle three bugs in one message; each deserves its own report and its own test.

Bug or change request?

Before writing the report, ask yourself: did I ask for this behavior in the first place? If the rule was in your prompt and is not applied, it is a bug; use the template above. If you have discovered a new rule you never mentioned, such as deposits being non-refundable during Eid, it is a change request. Write it as a clear new feature, 'Add a rule that…', not as a complaint that 'booking has a problem'. The agent then adds the missing logic instead of hunting for a fault that does not exist.

Weak reports vs strong reports: realistic examples

Strong reports need no technical terms. They contain numbers, a specific input format and a comparison of expected and actual. That is all the agent needs to know where to look.

How do you gather evidence if you are not a developer?

1. 1
   Reproduce it again

   If it repeats with the same steps, it is easy to fix. If not, say so and give the approximate time.

2. 2
   Change one variable

   Try the other language, another role or another device. 'Only happens in Arabic' is gold.

3. 3
   Copy the text, do not summarize it

   Paste the error message as shown, even if it is in English and means nothing to you.

4. 4
   Take a screenshot

   Especially for design and direction issues; one image replaces a paragraph.

5. 5
   Name the test data

   The order number, product name or test account you used, so the agent can find the same case.

What if the fix does not work?

Sometimes a fix goes in circles: the agent fixes one thing and another breaks, or the same bug returns in a new form. The practical rule: after two failed attempts, change approach instead of repeating 'still not fixed'.

1. Ask for diagnosis without edits: 'Do not change anything yet. Explain step by step how this feature works, where it could fail, and suggest two ways to fix it.'
2. Split the problem: instead of 'the booking page is full of problems', start with the smallest part: 'focus only on showing available slots for Friday'.
3. Roll back to a good version: if the problem started after a specific change, return to the version before it and request the change again more clearly. Fanann offers instant rollback to an earlier version.
4. Restart from a summary: in very long chats, context can get lost. Write a short summary of what the feature must do and start from it.

Why context gets lost and how to handle it: why AI forgets in long projects. For the broader guide for non-developers, see how to fix bugs in an AI-built app.

Stop bugs coming back: test after every fix

The real danger of a fix is breaking something else unnoticed. Keep a short list of your app's core scenarios and ask for them to be retested after every important fix:

• New customer sign-up with phone number, and login.
• A full order or booking with each payment method.
• Cancellation and refund.
• Staff and manager login, with permissions checked.
• Main pages on mobile in Arabic and English.

If real customers already use your app, try big fixes in a test environment first; see staging vs production. For a full pre-launch quality list, read can you trust AI-generated code?.

Ready prompts for common Gulf app bugs

These bugs recur in apps serving customers in Saudi Arabia and the UAE because they involve local calendars, digits, tax and time zones. Copy the one closest to your problem and adjust the details:

• Hijri date: 'The Hijri date on the booking page is one day off the official Saudi calendar. Tell me which calculation you use and align it with the Umm al-Qura calendar.'
• Phone numbers: 'If a customer types their number in Arabic-Indic digits, with spaces, or as +966, normalize it to one format before saving and sending the code.'
• VAT: 'On the invoice, calculate VAT on the order total rather than rounding per item, and show the pre-tax amount, VAT and total clearly. Test with 3 items at fractional prices.'
• After payment: 'If the customer closes the page right after paying, the order stays unpaid. Make sure payment confirmation is recorded from the gateway notification, not only from the thank-you page.'
• Direction: 'On the Arabic checkout page the email field types from the right. Make email and card fields always left-to-right.'
• WhatsApp: 'The WhatsApp confirmation does not arrive for orders placed after midnight. Check the time zone in use (Riyadh time).'

Frequently asked questions

Terms in this article