When it breaks, narrow the next move.
A failed check is information, but several speculative edits can bury it. Pause the feature work long enough to preserve the error and decide what kind of problem it represents. Then choose one small diagnostic action.
- 01
Keep the useful failure
Save the exact command, relevant output, and current repository state. Remove secrets from anything shared. A missing runtime, unavailable endpoint, and failing assertion belong to different layers. Avoid modifying application logic to compensate for an environment that cannot run the intended check.
- 02
Choose one explanation to test
State the leading hypothesis and the observation that would weaken it. Compare with an alternative cause when possible. Make a narrow check before attempting a broad rewrite. If the result is unchanged, record that result and update the theory instead of repeating the same action.
- 03
Resume from a verified point
After addressing the cause, rerun the relevant failed check and inspect any new changes. Remove temporary diagnostic code when it is no longer needed. Return to the original acceptance condition. The debugging detour is complete when the iteration can continue with a known state.
A useful finish
The failure has a supported explanation, and the next action follows from new evidence rather than another guess.
A diagnostic request
Diagnose [redacted failure] before changing code. Identify whether it is a code, environment, access, or provider issue. Give the strongest evidence for the leading cause and propose one small check that could disprove it. Preserve existing work.