AI Coding Prompt Generator

An AI coding prompt generator for bug reports, diffs and tests

This AI coding prompt generator packages a programming request the way a good bug report would: the language and versions, the code in a fenced block, the exact error, what you expected and what happened, the steps to reproduce, test cases, your constraints and the kind of answer you want back — a diff, a full file or an explanation.

Choose one of seven task types — write, refactor, debug, explain, write tests, review or convert — or one of nine templates. The checklist walks through a reproducible bug report, warns when a snippet looks like it contains a password or API key, and flags conflicting instructions. The prompt works with ChatGPT, Claude, Gemini or an assistant in your editor.

How to write a coding prompt

1. Pick a Task type or a template.
2. Enter the Language and Framework, libraries and versions, and describe what you want.
3. Paste the Code; for a bug, add the Error message or log, expected and actual behaviour and the steps to reproduce.
4. Add test cases, tick constraints such as No new dependencies and choose the Expected output.
5. Fix what the checklist flags, then copy the prompt as text, Markdown or JSON.

AI Coding Prompt Generator features

What makes a coding prompt reproducible

Stack Overflow’s guide to a “minimal, reproducible example” works for AI prompts too, because the model, like a stranger on a forum, knows only what you paste:

• Minimal — as little code as possible that still produces the same problem.
• Complete — everything needed to reproduce it, in the question itself.
• Reproducible — test the snippet, say what should happen, and give the exact error and the line that produces it.

The debug checklist follows that: error message, expected and actual behaviour, steps to reproduce, versions, and a warning past 150 lines of code.

Diff, full file or explanation?

Compare the AI’s version with yours in the diff checker before you commit it.

Vendor advice built in

Anthropic’s prompting guide (checked 14 October 2026) says some Claude models tend to over-engineer — extra files, abstractions and options nobody asked for — and suggests limiting changes to what was requested; Smallest change only adds that line. It also warns against code written just to pass the given tests, so listing test cases adds a request for code that is correct for all valid inputs. OpenAI’s Help Center notes that “leading words” such as import or SELECT steer the model to the right language; naming the language and tagging the fence does the same.

Never paste production secrets. The check looks for common key formats (sk-, AKIA, ghp_, xox, AIza), private-key headers and lines like password = "…" — a pattern match, not a guarantee.

Worked example

The start example — a Python function that crashes on an empty list — produces this prompt:

You are a senior Python developer. Find the root cause before changing anything, and explain it. If something you need is missing, ask for it instead of guessing.

Task:
Fix the crash when the list is empty.

Environment:
- Language: Python
- Framework, libraries and versions: Python 3.12

Code:
```python
def average(values):
    return sum(values) / len(values)

print(average([]))
```

Error message:
```text
Traceback (most recent call last):
  File "stats.py", line 4, in <module>
    print(average([]))
  File "stats.py", line 2, in average
    return sum(values) / len(values)
ZeroDivisionError: division by zero
```

Expected behaviour:
average([]) returns 0.0 instead of crashing.

Actual behaviour:
It raises ZeroDivisionError.

Steps to reproduce:
1. Run python stats.py

Test cases:
- average([2, 4]) → 3.0
- average([]) → 0.0

Constraints:
- Do not add new dependencies; use only the standard library and what the project already uses.
- Make the smallest change that does the job. Do not refactor unrelated code or add features nobody asked for.
- Make the code correct for all valid inputs, not only these test cases; never hard-code expected results.

Expected output:
Reply with a unified diff of your changes (as git diff would show them), then a short explanation of what you changed and why.