How do we test a prompt before it goes live?

Prompt edits have non-obvious blast radius. A clarification added for one case can change refusal behaviour, formatting or tone across many others, and the regression may not surface for weeks because nothing errors — the answers simply become subtly different.

Because prompts and skills are versioned records, a change can be staged and compared against the previous version rather than replacing it irreversibly. The Interaction Log retains what each execution assembled, so a before-and-after comparison reflects the full context — which skills loaded, what was retrieved — rather than only the final text. Rollback is a matter of reverting a record.