PromptForge
Student Hub

Debug a Real Bug With AI

ProjectIntermediate75 minClaudeChatGPT

Practise the debugging loop that works — evidence first, hypothesis second, fix last — on a bug you actually have. The habit this builds is refusing to accept a fix you cannot explain.

Why this one

The common failure is pasting an error and applying whatever comes back. Sometimes it works, and you have learned nothing and possibly hidden the real bug. This project enforces a loop where you understand the cause before touching the code.

0 / 5 steps

Steps

  1. 01Write the reproduction before you ask anything

    State in writing: what you expected, what happened, the exact error, and the smallest sequence that triggers it. Doing this by hand solves a surprising share of bugs outright, and when it does not, it is the input the model needs anyway.

  2. 02Ask for hypotheses, not a fix

    Requesting causes ranked by likelihood, with a way to test each, keeps you in control of the diagnosis. A fix offered before diagnosis is a guess with syntax highlighting.

    Prompt
    Here is a bug. Do NOT give me a fix yet.

List the 3-5 most likely causes, ranked, and for each: why it would produce exactly this symptom, and the single cheapest check that would confirm or eliminate it.

EXPECTED: [what should happen]
ACTUAL: [what happens]
ERROR: [exact text]
REPRODUCTION: [steps]
RELEVANT CODE:
[paste]

  3. 03Run the cheapest check first

    Work down the list. Each check should eliminate a hypothesis or confirm one. Resist skipping to the interesting theory — the cheap checks are cheap precisely because they rule out the boring causes that turn out to be right most of the time.

  4. 04Ask for the fix, then interrogate it

    Once you know the cause, ask for a fix — and then ask what it would break. The question "what else calls this, and what assumption does my change violate?" catches the fixes that trade one bug for two.

  5. 05Write the one-line explanation

    In your own words, in one sentence: what was wrong and why the fix works. If you cannot write that sentence, you have not fixed the bug, you have moved it. This is the whole point of the project.

You should end up with

A fixed bug and a one-sentence explanation of the root cause.

Done when

  • You eliminated at least two hypotheses with evidence before fixing anything
  • You can explain the root cause without referring back to the conversation
  • You checked what else the fix might affect

If you want to go further

  • Add a regression test that fails before your fix and passes after
  • Do the same exercise on a bug in an unfamiliar codebase

Try next

Topics

AI for Coding

Related across the site

Prompt

Chunking: Why Split Size Matters

Documents are split into chunks before embedding.

Prompt

Code Review Pass

Pre-review pass before submitting a pull request

Prompt

Custom GPTs & Claude Projects

Both ChatGPT (Custom GPTs) and Claude (Projects) let you upload documents and set standing instructions — a no-code RAG setup.

Glossary

Function calling

Function calling lets a model request that your code run a specific function with specific arguments. You expose tool definitions; the model returns a structured call, your code executes it, and the result goes back into the conversation.

Article

GPT vs Claude vs Gemini: An Honest Comparison

A practical comparison of the three leading AI models across writing, coding, reasoning, long documents and cost — including which one to reach for on which task, and where each genuinely falls short.

Glossary

Model drift

Model drift is when a hosted model's behaviour changes over time as the provider updates it, causing prompts that worked reliably to degrade without any change on your side.