Catching config drift with AI

After GreenGrid's config migration, a teammate adds a new alert_channels field to the production YAML so that alerts can be sent to Slack in addition to email. The pipeline doesn't break — but the new field is silently ignored because the config schema doesn't know about it.

A week later, the ops team wonders why Slack alerts never worked. You want to use AI to catch these kinds of mismatches before they reach production.

Which prompt approach gives AI the best chance of detecting the drift?