Collect a repair request. Return JSON {"what_i_tried":"what you did before the error","error_text":"full text of the error as is","expected_result":"what you expected to see"}. In error_text, paste the entire red error text - copy it from the place where it appeared (output window, terminal, under the code), do not retell it. You can take a realistic example like "TypeError: can't convert str to int, line 4".
A mistake is not a failure, but a hint: the computer honestly reports where it stumbled, but in its own language. Your job is to convey this message to the AI in its entirety and without distortion. It’s easy to take the “full text of the error”: this is the same red text that popped up where you ran the code - in the output window, terminal, or right under the code. Select it with the mouse from the first to the last line and copy it without rewriting anything by hand. It looks something like this: “TypeError: can't convert str to int, line 4” - the type of problem and often the line number. For the model, this is a map: using it, she finds a place and repairs it, and the retelling (“doesn’t work, red”) erases the map. In your request, indicate what you were doing before the error, the full text as is, what you expected and what you saw. Copy the entire message, including the unclear lines at the bottom - the real reason is usually hidden there, and the first line is often just the effect. If after fixing the error repeats, add “the previous solution did not help, here is a new error” and attach it. The worst thing is to write “gives an error” without text: it’s like telling the doctor “it hurts” without showing where.
Unlock access to submit solutions for instant AI review.