PROMPTHERE GUIDE
한국어로 읽기Coding prompts that connect bug reproduction to verification
‘Fix this bug’ leaves out the evidence needed to diagnose it. A coding prompt should include the error, reproduction steps, expected behavior, actual behavior and the parts of the system that may change. That context makes both the patch and its verification easier to review.
Updated · Prompthere
Prepare a reproducible description
Record the language and framework versions, command and input that trigger the issue. Share relevant code and the exact error rather than an unexplained dump of the entire project. Remove tokens, passwords and personal information from logs while keeping the material needed to understand the failure.
Separate a cause hypothesis from a finding
Suggesting a possible cause is different from confirming it. Ask for a log check or reproduction step alongside each hypothesis. Require a distinction between verification that was actually executed and verification merely proposed. A confident explanation is not evidence that a test ran.
Specify boundaries and completion conditions
State whether public APIs or data formats must stay compatible. Define the input that should now work and consider relevant edge cases. Request the reason for the change and any remaining limits along with the patch, so a reviewer can assess its effect on the surrounding code.
An example to make your own
A debugging request that keeps hypotheses visible
Replace the bracketed fields, then paste into your AI tool.
Environment: [language, framework, versions] Reproduction: [commands and input] Expected behavior: [expected result] Actual behavior: [error and logs] Relevant code: [code] Allowed changes: [scope] List cause hypotheses with a way to check each. Propose a fix based on available evidence, explain the change and give verification steps. Do not claim tests passed unless they actually ran. Explain any required public-interface change before making it.
Before you run the prompt
- Provide steps and input that reproduce the failure.
- Name the environment and expected behavior.
- Distinguish executed checks from proposed tests.
Frequently asked questions
Can I start with only an error message?
Yes, but it may not establish the cause. Add reproduction steps, environment information and relevant code. Keep unconfirmed explanations labeled as hypotheses.
Is an AI statement that tests passed enough?
Check the actual command, environment and result. When the assistant cannot execute tests, treat its test suggestions as a verification list to run yourself.