deploy-checklist
Produces a deployment checklist specific to the repository, with a GO/NO-GO verdict before the deploy and checks after it, covering validation, migrations, feature flags, monitoring, communication and rollback.
- Group
- Engineering
- Kind
- Action
- Writes
- Nothing
- Asks before
- Sending any communication: it only drafts it
When it runs
When you plan a deployment, a release to production or another shipping event:
- we are deploying to production
- deploy checklist
- can we deploy
What it does
An item whose evidence is unavailable (no CI status, no dashboard access) is marked UNVERIFIED, never assumed to pass.
- Validation. It detects the package manager from the lockfile and runs the repository's real test, lint, type-check and build commands, and requires green CI on the intended release branch.
- Migrations. It checks that they ran locally or on staging, are backward compatible, have a backup, an expected lock time and runtime, and a tested rollback.
- Feature flags. It checks defaults and rollout percentages for risky behavior.
- Monitoring. It confirms dashboards, error tracking, alerts and explicit abort thresholds.
- Communication. It drafts deployment, client or status messages when needed, without sending them.
- Rollback. It writes the exact rollback command, the recovery steps and the metrics that trigger it. Provider-specific commands come from the repository's configuration; if the target is unknown, it asks.
- Window and owner. It confirms a suitable deployment window and an owner available for at least an hour of monitoring.
- Verdict.
GO, orNO-GOwhen any item above failed or is UNVERIFIED, with what would turn it intoGO. - After the deploy. It monitors error rate and latency against the abort thresholds, inspects new exceptions, runs a smoke test of the critical flow, and records the result: stable, or rolled back with the trigger. It offers brain-capture when the deployment taught a lasting lesson.
Boundaries
Stops when
Before the deployment, at the GO/NO-GO verdict; after it, at the recorded result. You run the deployment.
Never
- Deploy, run a production migration, or change a feature flag.
- Send deployment, client or status communication without your explicit authorization.
- Mark an item passed without evidence.
- Invent provider-specific commands.
Related
- bug-triage takes over when the deploy goes wrong.
- pr-cycle gets the change to a merge verdict first.
- monitoring-conventions covers what the abort thresholds watch.