Human checkpoints in deploy scripts
You're letting an agent help rewrite a critical deploy/rollout script in Bash that manages a blue-green switch over a fleet of load balancers — drain connections, flip traffic, health-check, and roll back on failure. The agent is fast and the script is mostly mechanical, but a bug here takes production down for everyone. Decide the human-in-the-loop checkpoints: what you let it draft freely, what you review before it can run anywhere, and what it must never execute against prod without an explicit human action — and explain the reasoning for each line you draw.
plan_rollout_actions(lb_names: list[str], health_ok: list[bool]) → list[str][["lb1","lb2"],[true,true]]out["drain:lb1","flip:lb1","healthy:lb1","drain:lb2","flip:lb2","healthy:lb2","promoted"][["lb1","lb2","lb3"],[true,false,true]]out["drain:lb1","flip:lb1","healthy:lb1","drain:lb2","flip:lb2","unhealthy:lb2","restore:lb2","restore:lb1","aborted"][["lb1"],[]]out["drain:lb1","flip:lb1","unhealthy:lb1","restore:lb1","aborted"]Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.