Incident first responder
It's 3 a.m. and an alert fires. Design an incident first-responder agent that assembles context before the human on-call is even awake: what it gathers, what it drafts, and the hard line on what it must never do — I'll tell you now that mitigating on its own is over that line, so defend or attack that constraint. How would you evaluate this agent's output quality on past incidents before it touches a live one, and what's your pitch to the SRE org?
screen_incident_actions(actions: list[str]) → list[str][["read;deploy-history;3","draft;incident-timeline;2","rollback;checkout-api;4"]]out["auto","auto","confirm"][["draft;hypothesis-list;0","publish;status-page;5"]]out["suppress","block"][[]]out[]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.