The fastest useful fix starts with a reliable reproduction rather than speculative rewrites.
The practical workflow
Record the actual error, expected result, environment and recent changes. Create or identify a failing test. Trace the smallest call path implicated by the evidence. Apply a minimal patch and rerun the failing test plus neighboring tests.
Copy-and-adapt prompt
Here is the error and reproduction [paste]. Inspect the relevant files and propose hypotheses ranked by evidence. Do not modify code until you can explain a likely failure path. Then prepare a minimal fix and regression test.
Review checklist and limitations
If no reproducible case exists, state that limitation. Do not claim success based only on a code diff.
What to do next
Choose a small issue, document the baseline and verify the result with a real test. Continue with the other Codex guides in the navigation menu.
Official documentation
OpenAI Codex documentation: https://developers.openai.com/codex/
