Debug by Pasting the Error, Not the Vibe

A debugging discipline: give ChatGPT the exact error message, the stack trace, and the smallest snippet that reproduces it — rather than describing the symptom in your own words.

The error text is the single most information-dense clue you have. Paraphrasing it throws that away and forces ChatGPT to guess, which is where confident wrong fixes come from.

Every time you have an actual error, exception, or failing test. The more literal the paste — exact message, line numbers, versions — the sharper the fix.

When there is no error and the problem is 'the output is subtly wrong', pasting a trace won't help — there describe the expected vs actual behaviour with examples instead.

This fails. Here is the exact error and a minimal reproduction.

Error:
<paste full message + stack trace>

Code:
<smallest snippet that reproduces it>

Environment: <language/runtime + versions>.

Explain the root cause first, then give the minimal fix. Don't rewrite unrelated code.

Pasting a full TypeError trace plus the six lines around it, ChatGPT pinpoints an off-by-one on line 3 and returns a two-character fix — no rewrite of the surrounding module.

Ask ChatGPT to first state what the error means in general, then map each part of the trace to your code. Understanding beats a patch you can't defend later.

• Retyping the error from memory and dropping the part that mattered.
• Pasting 400 lines of context instead of a minimal reproduction.
• Applying the fix without asking why it failed, so the next instance surprises you too.