Incident communication drafter
During incidents, engineers either go silent on the status page or overshare speculation — both hurt. Design an agent that drafts incident communications: status-page updates, customer emails, the follow-up summary. Where exactly is the human gate, and which words can it never publish without review — think legal exposure, blame, root-cause claims? How would you evaluate the drafts before a real sev-one, and how do you convince the incident commander this helps rather than adds a step?
route_incident_draft(draft: str, postmortem_published: bool) → str["Checkout is degraded for some requests. We are investigating. Next update in 30 minutes.",false]out"queue_for_commander"["We have identified the root cause: a bad config push. Next update in 20 minutes.",false]out"block_root_cause_claim"["We are investigating possible data exposure in the export job. Next update in 15 minutes.",false]out"hold_for_legal_and_security"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.