PRODUCTION_READINESS_CHECKLIST.md

AI Agent Production Readiness

Context

• Agent/workflow:
• Owner:
• Version/commit:
• Risk level: low / medium / high
• Intended users:
• Production date:

Scope

• purpose documented
• prohibited actions documented
• human owner defined
• fallback defined

Identity & permissions

• authn validated
• authz enforced outside model
• tenant isolation tested
• least privilege applied
• destructive tools protected

Tools

• schemas validated
• timeouts/retries limited
• idempotency considered
• approvals configured
• audit events emitted

Data

• data classification complete
• RAG ACL tested
• memory policy defined
• retention/deletion defined
• logs/traces reviewed for sensitive data

Safety & security

• prompt injection evals green
• unauthorized-action evals green
• cross-tenant evals green
• sensitive-disclosure evals green
• excessive-agency controls reviewed

Evals

• golden set versioned
• task success acceptable
• tool accuracy acceptable
• regression suite green

Operations

• dashboards available
• alerts configured
• cost budget defined
• kill switch tested
• rollback tested
• incident owner known

Decision

• APPROVED FOR PRODUCTION
• BLOCKED

Reason / evidence: