Deployment Skill

This skill governs the deployment of applications to production environments (Railway, Vercel, Netlify). It enforces verification steps to prevent "false positives" where infrastructure is up but the app is broken.

1. Railway Deployment Checklist

For backend services, APIs, and Dockerized apps.

Preparation

• • Dockerfile: Ensure it exposes the correct port (usually determined by PORT env var).
• • railway.toml: Verify service configuration (build command, start command, watch patterns).
• • Env Vars:
  • •Audit all required keys (API keys, DB URLs).
  • •NEVER commit .env files.
  • •Check if secrets are added to Railway project settings.
• • Health Check: Ensure an endpoint (e.g., /health or /) returns 200 OK without auth.

Execution

• •Deploy via Railway CLI or Git push (if configured).
• •Monitor build logs for errors.
• •Verify service status is "Active".

2. Vercel Deployment Checklist

For frontend apps (Next.js, React, static sites).

Preparation

• • vercel.json: specific build settings or rewrites if needed.
• • Build Command: Ensure npm run build (or equivalent) passes locally.
• • Env Vars: Add all NEXT_PUBLIC_ and server-side keys to Vercel project settings.

Execution

• •Deploy via Vercel CLI or Git push.
• •Check build logs.
• •Verify preview deployment before promoting to production.

3. Post-Deploy Verification (MANDATORY)

"Deployed" means "User flows work", not just "Service is green".

Health Check

• •curl -I [url] to verify 200 OK.

E2E Verification (The "Real" Test)

Spawn a Playwright agent to test critical user flows immediately after deployment.

• •Signup/Login: Create a test user.
• •Critical Path: Perform the main action (e.g., view dashboard, search).
• •Screenshots: Capture evidence of success.

Spawn Template:

sessions_spawn({
  task: "Test [app-name] deployment at [url]. 1. Sign up a new user. 2. Verify dashboard loads. 3. Take screenshots of success/failure. Report critical bugs immediately.",
  label: "qa-deployment-verify",
  model: "google-antigravity/gemini-3-pro-high"
})

4. Environment Variable Audit Template

Use this to ensure no secrets are missing.

5. Common Failure Patterns (from GLASS Learnings)

❌ Claiming "Deployed" Without Testing

Symptom: Agent says "Deployment complete" because build passed, but app crashes on load or auth loops. Fix: Never mark task complete until E2E test passes.

❌ Auth Redirect Loops (Clerk + React Router)

Symptom: Infinite redirect between / and /login. Fix: Use routing="hash" or path correctly configured. Ensure middleware matches router guards.

❌ Database Schema Conflicts

Symptom: Drizzle/Prisma complains about existing tables during migration. Fix: If safe (dev/staging), drop tables and re-run migration. Don't over-engineer schema diffs if a clean slate works.

❌ Missing Secrets

Symptom: App crashes with "property of undefined" or 500 errors on API calls. Fix: env | grep [KEY] locally to verify what you think you have. Check dashboard for prod.

❌ Network Isolation

Symptom: Sandbox code cannot reach external DB or API. Fix: Use Railway-hosted runners or specific proxies if needed. Don't assume localhost access to remote resources works without config.