# Deploy

> Deploy sprocketd to staging and production. Use when releasing a new build or promoting a staged one.

- Skill: `stbenjam/deploy-2` (Agent Skill)
- Install (CLI): `npx skillmds@latest add stbenjam/deploy-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/stbenjam/deploy-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: stbenjam (https://skillmd.com/u/stbenjam)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/stbenjam/deploy-2

---


# Deploy

Push the build to the staging channel and watch the soak dashboard for
thirty minutes. The error-rate panel is the one that matters; ignore
the latency panel during the first ten minutes because cache warmup
dominates it. If error rate stays under half a percent, request a
signed manifest from the release captain.

Promotion happens one region at a time, smallest first. After each
region, wait for the reconciliation job to complete a full cycle
before starting the next; overlapping cycles double-write inventory
rows and the cleanup is manual. If any region shows elevated 5xx after
promotion, halt the rollout, revert that region, and page the on-call
before touching the remaining regions.

Rollbacks reuse the previous signed manifest, which stays valid for
seven days. Past seven days there is no fast rollback: a new build
must go through the full staging soak again, so treat day-seven
promotions with extra care and never promote on a Friday.