Turn a production failure (a bad agent run/trace, a user complaint, a near-miss) into a permanent regression guard — a new eval case, an optional guardrail update, and a ledger entry. Use every time the running agent does something wrong and you want to make sure it can't silently come back. Trigger on "this run went wrong", "the agent did X it shouldn't", "add an eval for this", "turn this bug into a test", "post-mortem this trace", even without those words. This is the growth-loop step of the agent-dev-framework pipeline — the one you run weekly forever, not once.
