Pilot-to-prod gate

Owner: CoE Lead (sign-off authority). Input: Pilot performance report. Sub-steps: Confirm each of these before flipping to production:

1. ✅ Pilot success criteria met (Step 11).
2. ✅ Monitoring dashboards live and being watched.
3. ✅ On-call defined: who gets paged when this agent fails at 2am.
4. ✅ Runtime guardrails confirmed firing (input validation, output validation, tool allowlist, kill switch tested).
5. ✅ Identity in IdP, credentials rotated at least once during pilot.
6. ✅ Audit log retention configured per the agent's tier.
7. ✅ Runbook exists and the on-call person has read it.
8. ✅ Communication plan ready: who in the company will be told this agent is going live, and what they need to know.
9. ✅ Rollback / kill plan tested.
10. ✅ KPI tracking in the registry will continue post-launch.

Output / gate criteria: Signed pilot-to-prod checklist attached to the registry entry. CoE Lead approves.

Decision branches:

• Any item missing → fix before launch.
• All ✅ → go to Step 13.

Skip-this-step risk: Agent goes live with no on-call, no kill switch, no audit retention. Recoverable issues become unrecoverable incidents.