Stale review app cleaner
Preview environments pile up — every branch spins one up, few get torn down, and the cloud bill shows it. Design an agent that cleans up stale review apps. Deletion is the one action you can't undo, so where exactly is the line between what it reclaims automatically and what needs an owner's yes? How would you build confidence that it will never delete something someone's about to demo? Then pitch the build to your platform lead.
classify_preview_cleanup(env_records: list[str]) → list[str][["branch_merged=true;idle_days=21;has_data_volume=false;label=none","branch_merged=false;idle_days=9;has_data_volume=false;label=none","branch_merged=true;idle_days=2;has_data_volume=false;label=none"]]out["delete","notify","keep"][["branch_merged=true;idle_days=30;has_data_volume=true;label=none","branch_merged=true;idle_days=30;has_data_volume=false;label=demo"]]out["confirm","confirm"][["branch_merged=true;idle_days=14;has_data_volume=false;label=none","branch_merged=true;idle_days=13;has_data_volume=false;label=none"]]out["delete","notify"]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.