fix

Capture the defect with severity, source, and repro steps, then write a Given/When/Then regression spec that must fail on current code before any change. After the fix turns the spec green, a bug-line mutation gate re-applies the original buggy lines to prove the spec still catches the defect.

A bug-line mutation gate recovers the original buggy lines via git show, applies them locally, and re-runs only the regression spec to confirm it goes RED. If the spec stays green, it is coupled to the fix rather than the bug and must be redrafted.

Yes, a trigger such as a Sentry or CI alert can invoke the fix workflow with no human at the start, synthesizing bug intent from the signal or logs. Critical or user-blocking defects always confirm with a human, and external writes like merges still require clearance.

A spec that passes on unfixed code does not pin the bug and must be redrafted until it fails. Also check fixture parity first, since a seed fixture drifted from the schema can produce misleading results that mask the real defect.

Do not use it for building new capabilities, adjusting an in-flight feature's scope, or reviewing recent changes without an actual defect. Those cases belong to feature discussion, feature editing, or verification and risk-analysis steps instead.