A skilled operator reading a runbook would seem less likely to exacerbate the mess than a naively written script. The operator can think, "wait a minute, this isn't right." Maybe they won't when they're sleeping off a night of drinking at 3am, but they could. The Python interpreter can't.
To make the script less naive, you'd need a development environment with the same infrastructure components as production, and the freedom to make them fail so you can develop your script against their failure states.
To make the script less naive, you'd need a development environment with the same infrastructure components as production, and the freedom to make them fail so you can develop your script against their failure states.