When Should a Prompt Change Trigger Human Reapproval?

A prompt edit can look small in review and still change what an agent returns—or what it is allowed to do next. Rechecking every change with a person may slow delivery; relying only on automated tests can miss a consequence the tests were never designed to catch.

What evidence should determine whether a prompt revision needs human reapproval before release? Would you base that decision on the affected workflow, the permissions involved, changes in test results, or the possible impact of an unexpected output? And who should own that release decision when several teams depend on the same prompt?