debugging

Reproduce the failure first and capture the exact symptom: command, input, output, and stack trace. Then test one hypothesis at a time with the smallest experiment possible, and only edit code after stating the likely cause and the evidence supporting it.

State a failure model naming the likely cause, the evidence for it, and the observation that would disprove it. The root cause should explain every observed symptom in one sentence and name the evidence that ruled out the main alternatives.

Investigate it. A flake is a bug, so identify whether the test, the code, or the environment failed rather than retrying it away. Retrying hides the defect and lets it resurface later in production.

Skip it for planned refactors with no failing behavior, which belong in a refactoring workflow, and for pure performance symptoms, which fit a performance workflow. Git bisect and reflog mechanics pair better with a git-workflow skill.

Sleeps mask race conditions without proving the cause. Timing-sensitive bugs should be verified with non-invasive observation or replay, showing interleaving, shared state, or timing evidence before touching concurrency code.