Skip to content

deploy-ops

  • Group: Operations and deployment
  • When it activates: Deploying and running any stack in production: Netlify/Vercel/Forge/VPS, Docker, CI/CD with GitHub Actions, per-environment secrets, queues and cron in prod, backups with tested restores, monitoring and incidents
  • Installed: Always: comes with every project.

Putting things in production and KEEPING them there, for any stack. Building and verifying belongs to the other skills (/verify, /launch); this one starts where the “READY” of /launch ends: publish, operate and keep it from going down — and if it goes down, roll back in minutes.

Task Read only
Deploy a project (where and how, by stack) references/deploy-by-stack.md
Dockerize (Dockerfile, dev compose, prod image) references/docker.md
CI/CD: GitHub Actions pipeline, previews, deploy with approval references/ci-cd.md
Hand over the project to the client: manual, access, maintenance (/handoff) references/entrega-cliente.md
Environments and secrets (.env, staging, rotation, leaks) references/envs-secrets.md
Queues, cron, workers, storage and logs IN production references/production-runtime.md
Backups (with tested restore), uptime, Sentry, /health, incidents references/backups-monitoring.md
Run a specific deploy (checklist with evidence) references/deploy-checklist.md (command /deploy)
Verify the WEBSITE before publishing (SEO, legal, measurement) ui-verify §references/launch-checklist.md (/launch)
Comply with a website’s legal side (cookies, privacy, legal notice, banner, GDPR in depth, AI/AI Act) references/legal-web.md
Queue/job design (the WHAT; here only running it in prod) code-quality §references/jobs-and-queues.md
  1. Deploy to production NEVER without explicit approval from the user at that moment (the guard blocks a direct --prod; the SENZU_ALLOW_DEPLOY=1 escape only after approval, documented in the devlog).
  2. Before touching prod: a fresh VERIFIED backup + a written rollback plan. Without that, there is no deploy.
  3. Secrets never in the repo or in logs: they live in the platform (Netlify/Vercel/Forge/GH Environments); .env.example kept up to date; committed secret = rotate it NOW (protocol in envs-secrets §leak).
  4. Reproducible or it does not count: the deploy is a script/pipeline, not a list of steps in someone’s head. If it only works “when X does it by hand”, it is broken.
  5. After the deploy you WATCH: smoke test (home + /health + 1 critical flow) and logs for ~10 min. Deploying and closing the laptop is not finishing.
  6. Staging if the project has it; never “test in prod” risky changes (destructive migrations, payments, auth).
  7. Infra changes are documented in the devlog like any code (what, why, how to revert it).
  8. Public website = legal pages, no exceptions: legal notice, privacy and cookies reachable from the footer
    • a real consent banner (visible reject button, non-essential cookies blocked until accepted). The guard blocks the deploy to prod if there is NO legal page at all in the repo; the SENZU_ALLOW_SIN_LEGAL=1 escape only if the legal side is managed outside the repo (CMS, external template). WHAT each page must contain and how to set up the banner, in references/legal-web.md; the checklist, in /launch.

Default recommendations (agency, no over-engineering)

Section titled “Default recommendations (agency, no over-engineering)”
  • Static website/Astro/SPA → Netlify or Cloudflare Pages (deploy via git, free previews, zero servers).
  • Next.js → Vercel unless there is a reason (cost/data) to self-host.
  • Laravel → VPS + Forge/Ploi (or Docker if the team already masters it). Shared hosting without SSH: no.
  • Python/agents → Docker on a VPS or Fly.io/Railway.
  • CI: GitHub Actions mirroring verify-build on every PR; prod only from main with approval.

/launch verifies the website → deploy-ops publishes and maintains it · queues: design in code-quality, prod here · app security in security-owasp; here, deployment security (secrets, surfaces, access).