Debug Code With AI: 7 Copy-and-Paste Prompts That Work

Published 2026-10-15 · by AgentKit Co

AI assistants can be excellent debugging partners, but only if you ask well. "Fix my code" gets you a rewrite you don't understand. The prompts below get you an explanation, a small fix and a way to make sure the bug stays gone.

Each prompt has blanks in [square brackets]. Fill them in, paste, and send. They work in any language and with any modern chat assistant or AI coding tool.

Before you start: give the AI context

Start the chat with one short block so every answer fits your setup:

Context: I'm using [language + version] on [OS/framework].
My level: [beginner/intermediate]. I want you to explain your
reasoning briefly, make the smallest change that works, and say
so when you're unsure instead of guessing. Never invent functions
or libraries.

That last sentence matters. It reduces the chance of the assistant referencing a function that doesn't exist.

1. The error decoder

Here is the full error, including the traceback:
[paste everything]
Here is the code around it: [paste 20-40 lines]
In plain English: what is this error saying, which line is the
real cause (not just the last line shown), and what are the three
most likely reasons? Rank them.

Why it works: the line where a program crashes often isn't the line that caused the problem. Asking for the real cause and a ranked list steers the assistant away from a first guess.

2. Expected vs. actual (for bugs with no error)

This code should [expected behavior] but instead [actual behavior].
Input: [input]. Expected output: [output]. Actual output: [output].
Walk through the code step by step with this input and tell me
where the actual behavior first diverges from what I expect.

Why it works: most "wrong answer, no error" bugs are logic bugs. Making the AI trace a specific input forces it to find where reality and your expectation split.

3. The rubber duck

Don't give me the answer yet. Ask me up to five questions, one at
a time, that would help me find this bug myself. Here's the code
and the symptom: [paste]

Why it works: sometimes you learn more by being guided. This is the best prompt if your goal is to improve, not just to get unstuck.

4. The minimal reproducible example

Help me shrink this problem into the smallest program that still
shows the bug. Remove everything unrelated and tell me what you
removed and why. Here's the code: [paste]

Why it works: half of all debugging is isolating the problem. A minimal example is also what you need if you ask a forum or open an issue.

5. "It worked yesterday"

This worked before and now it doesn't. What I changed since then:
[list changes, new packages, updates]. Here's the error or
symptom: [paste]. List the most likely culprits in the order I
should check them, and how to check each one quickly.

Why it works: listing what changed turns the AI into a detective with clues instead of a guesser.

6. The smallest-diff fix

Fix it with the smallest change possible. Show only the changed
lines plus 2 lines of context. Explain in one or two sentences
why this fixes it. Don't refactor or rename anything else.

Why it works: big rewrites can fix one bug and quietly add three more. Small diffs are easy to review and easy to undo.

7. The "safe to run?" check

Before I run this command or script: explain what it does line by
line. Flag anything that deletes or overwrites files, installs
software, needs admin rights (sudo), changes settings, or sends
data over the internet. How can I test it safely first?

Why it works: AI suggestions sometimes include commands with side effects. For example, previewing what git clean would delete before actually running it is a good habit, and this prompt builds it in.

Bonus habit: turn every bug into a test

After you fix a bug, ask:

Write a test that would have FAILED because of this bug and
passes now. Explain what it checks.

That's how the bug stays fixed. We cover this in more detail in our guide to AI prompts for learning to code.

A word on trust

AI can be wrong, with total confidence. Always read the suggested change, run your tests, and never run commands you don't understand. If an answer references a function or option you can't find in the official docs, treat it as suspect.

Get all 40

These seven are a taste. The Prompt Pack for Coding has 40 copy-paste prompts across five sections: learning, debugging, refactoring, testing and shipping. Each has a "when to use" note, the full prompt and a short example, plus a "safe to run?" checklist and a two-page quick reference.

Prompt Pack for Coding

40 prompts to learn, debug and ship small projects. 22-page PDF, $9.

Keep reading

AI Prompts for Learning to Code: 8 Prompts to Learn FasterEight AI prompts that turn an assistant into a patient coding tutor: trace tables, three-way explanations, adaptive quizzes and more.