Master Production Deployment Procedures

This skill teaches deployment principles and decision-making for safe production releases - understanding the why behind each step and adapting procedures to the platform, rather than memorizing fixed scripts.

Its platform selection decision tree routes by what's being deployed: static sites/JAMstack to Vercel/Netlify/Cloudflare Pages, simple web apps to managed platforms (Railway, Render, Fly.io) or self-managed VPS with PM2/Docker, microservices to container orchestration, and serverless to edge functions/Lambda - each platform having a distinct deployment method (git push auto-deploy, SSH plus manual steps, image push, or kubectl apply).

Its pre-deployment principles cover four verification categories (code quality: tests/linting/review; build: production build works cleanly; environment: env vars and secrets current; safety: backup done and rollback plan ready) and an eight-item checklist covering tests, review, build, env vars, migrations, rollback plan, team notification, and monitoring readiness.

Its five-phase deployment workflow is Prepare (never deploy untested code) to Backup (can't rollback without one) to Deploy (watch it happen, don't walk away) to Verify (trust but verify) to Confirm-or-Rollback (have the rollback trigger ready). Post-deployment verification checks the health endpoint, error logs, key user flows, and performance, on a verification window of active monitoring for the first 5 minutes, stability confirmation at 15 minutes, final verification at 1 hour, and a metrics review the next day.

Its rollback principles define when to rollback (service down or critical errors: rollback immediately; performance degraded over 50%: consider it; minor issues: fix forward if quick) and platform-specific rollback methods (redeploy previous commit for Vercel/Netlify, dashboard rollback for Railway/Render, restore backup and restart for VPS+PM2, previous image tag for Docker, kubectl rollout undo for Kubernetes) - guided by speed over perfection, not compounding errors with multiple simultaneous changes, communicating what happened, and a post-mortem once stable.

Its zero-downtime deployment strategies are rolling (replace instances one by one, standard release), blue-green (switch traffic between environments, for high-risk changes needing easy rollback), and canary (gradual traffic shift, for validating with real traffic). Its emergency procedures for a service-down incident: assess the symptom, try a quick restart if unclear, roll back if the restart doesn't help, then investigate once stable - checking logs, resources (disk/memory), network (DNS/firewall), and dependencies (database/APIs) in that order.

Its anti-patterns table contrasts risky habits against safer alternatives: deploying on Friday (deploy early in the week instead), rushing deployment (follow the process), skipping staging (always test first), deploying without backup, walking away after deploying (monitor for 15+ minutes), and bundling multiple changes into one deploy (one change at a time). Its best practices favor small frequent deploys over big releases, feature flags for risky changes, automating repetitive steps, documenting every deployment, reviewing what went wrong after issues, and testing the rollback path before it's actually needed - because every deployment is a risk minimized through preparation, not speed.

Use this skill when making deployment decisions for safe production releases - selecting a platform-appropriate procedure, verifying readiness before deploying, choosing a zero-downtime strategy, planning rollback, or handling a deployment emergency.

Inputs: a deployment target (platform, application type) and its current readiness state (tests, build, environment, backup).

Outputs: a platform-appropriate deployment procedure, a pre-deployment readiness checklist, a selected zero-downtime strategy, a documented rollback plan and trigger criteria, and a post-deployment verification schedule.

Vercel, Netlify, Cloudflare Pages, Railway, Render, Fly.io, PM2, Docker, Kubernetes (kubectl rollout undo).

Engineers making deployment decisions for production releases who need platform-appropriate procedures, rollback discipline, and zero-downtime strategy selection.