The Production Readiness Review

Review this agent system for production readiness.

System: [DESCRIBE]
Current state: [PROTOTYPE / INTERNAL / LIMITED RELEASE]
Users: [WHO AND HOW MANY]
Failure impact: [DESCRIBE]

Assess against:
1. Correctness — is there an eval set, does it run, what does it
   cover
2. Safety — guardrails, approval gates, blast radius limits
3. Observability — could someone debug a bad run
4. Reliability — retries, timeouts, circuit breakers, graceful
   degradation
5. Cost — is spend bounded, is it monitored, what happens on a
   runaway
6. Security — credential handling, permission scoping, injection
   defences
7. Data — what is retained, for how long, under what basis
8. Operations — who is on call, what alerts fire, what the runbook
   says
9. Change management — how prompts are versioned, how model
   changes are handled
10. Rollback — how to turn it off

For each: ready, gap, or blocker. For each gap, the specific work
required.

Then: the three things I must fix before this touches real users,
in order.

Be strict. Agent systems fail in ways that are hard to detect and
expensive to explain.