Layered Defense for a 'Never Touch Prod DB' Org Rule

Your organization has one rule that must hold with zero exceptions: Claude Code sessions must never run any command against the production PostgreSQL database, identified by the hostname prod-db.internal.example.com. Engineers routinely use Claude Code against staging and local databases with similar tooling (psql, migration scripts), so a blanket ban on database commands isn't practical.

1. Design a layered defense for this rule using at least three distinct mechanisms from this subject, explaining what each layer catches that the others might miss.
2. Explain specifically why relying on a single permissions.deny rule alone would be an incomplete defense here, even if written carefully.
3. Where does sandboxing fit into this design, and what does it add that permission rules and hooks don't?