deploy

---

name: deploy description: Deploy Rails applications to Railway. Handles first-time setup and re-deploys idempotently using Railway CLI. Trigger on: "deploy", "deploy to railway", "railway deploy", "发布", "部署", "上线". agent: coding disable-model-invocation: false

Deploy Rails App to Railway

Deploy the current Rails project to Railway using the Railway CLI. Works for both first-time deploys and re-deploys.

Prerequisites Check

Before starting, verify:

# 1. Railway CLI installed?
railway --version

# 2. Logged in?
railway whoami

If not logged in, instruct the user:

Please run: railway login
Then retry deployment.

Step 0: Prepare for Linux Build

Railway runs on Linux. Ensure Gemfile.lock includes the linux platform:

bundle lock --add-platform x86_64-linux

If the project uses a Dockerfile builder (check railway.toml for builder = "DOCKERFILE"), no Procfile is needed — skip creating one.

Only create a Procfile if there is no Dockerfile:

web: bundle exec puma -C config/puma.rb

Step 1: Check Link Status → Deploy Immediately if Already Linked

First: check if already linked:

railway status 2>&1

If output contains Project: → project is already linked. Skip Steps 2–5 entirely and jump to Step 6 (Deploy).

If output contains "not linked" or an error → not linked yet. Try linking to an existing project first — list available projects:

railway list 2>&1 | grep -i "<app-name>"

If a matching project is found, link it:

railway link --project <project-name> --service <service-name> 2>&1

Only if no existing project is found, init a new one:

railway init -n <app-name>

The app-name should match the current directory name (e.g., my-rails-app).

⚠️ NEVER run railway init when already linked or when an existing project exists. It silently creates a brand-new Railway project. If this happens by mistake:

1. Find the correct project name from railway list
2. Re-link: railway link --project <project-name> --service <service-name>

Step 2: Set Environment Variables

Set required Rails production variables. Use --skip-deploys to avoid triggering premature deploys:

# Generate a secret key base
SECRET_KEY_BASE=$(bundle exec rails secret)

railway variable set SECRET_KEY_BASE=$SECRET_KEY_BASE --skip-deploys
railway variable set RAILS_ENV=production --skip-deploys
railway variable set RAILS_LOG_TO_STDOUT=true --skip-deploys
railway variable set RAILS_SERVE_STATIC_FILES=true --skip-deploys

If the project uses any other env vars (check .env.example or config/application.yml.example if they exist), prompt the user to provide values and set them too.

If the project uses config/application.yml (Figaro gem), read it and set all values as Railway variables:

# Read application.yml and set each key=value pair
ruby -ryaml -e "
  data = YAML.safe_load(File.read('config/application.yml')) || {}
  data.each { |k, v| puts %(railway variable set #{k}=#{v} --skip-deploys) unless v.to_s.empty? }
" | bash

Step 3: Ensure PostgreSQL Service (Idempotent)

Check if Postgres already exists:

railway status --json

Parse the JSON output. If a service with type postgres or name containing postgres/Postgres is already found, skip with: ✅ PostgreSQL already provisioned

⚠️ IMPORTANT: railway add --database postgres has a known CLI bug that always returns Unauthorized. Do NOT attempt to run this command. Instead, instruct the user to add PostgreSQL manually via the Railway Web UI:

1. Open your Railway project: https://railway.com/project/<project-id> (Get the project ID from the Railway dashboard or cat .railway/config.json)
2. Click "+ New" → "Database" → "PostgreSQL"
3. Wait for the database to provision
4. Come back and continue

After Postgres is provisioned, set the DATABASE_URL variable:

railway variable set DATABASE_URL='${{Postgres.DATABASE_URL}}' --skip-deploys

Step 4: Get Domain (Idempotent)

Check if a domain is already set:

railway domain --json

If no domain exists yet:

railway domain

Capture and display the domain URL to the user. Also set it as PUBLIC_HOST:

railway variable set PUBLIC_HOST=<domain-without-https> --skip-deploys

Step 5: Configure Storage Bucket (if needed)

Check if the project uses S3-compatible storage by reading config/storage.yml. If it contains an amazon or s3 service section, storage bucket configuration is required.

⚠️ Storage bucket requires Railway Hobby plan ($5/month minimum). Confirm with the user before proceeding.

Check if bucket env vars are already set:

railway variables --json | grep STORAGE_BUCKET

If not set, create a bucket and configure the variables:

# Create bucket (choose region: iad=US East, sjc=US West, ams=EU, sin=Asia)
railway bucket create <app-name>-storage --region iad --json

# Get credentials
railway bucket credentials --bucket <app-name>-storage --json

The credentials JSON will contain: accessKeyId, secretAccessKey, region, endpoint, bucketName.

Set them as environment variables:

railway variables set \
  STORAGE_BUCKET_ACCESS_KEY_ID=<accessKeyId> \
  STORAGE_BUCKET_SECRET_ACCESS_KEY=<secretAccessKey> \
  STORAGE_BUCKET_REGION=<region> \
  STORAGE_BUCKET_NAME=<bucketName> \
  STORAGE_BUCKET_ENDPOINT=<endpoint> \
  --skip-deploys

⚠️ Missing these variables will cause a hard crash at boot (Aws::Errors::MissingRegionError) because the AWS SDK initializes at startup, not lazily.

If the project does not use S3 storage, skip this step entirely.

Step 6: Deploy

Upload and deploy the project:

railway up --detach

Show the user the deployment is in progress and they can monitor it with:

railway logs

No manual migration needed. The bin/docker-entrypoint script runs rails db:prepare automatically on container startup. Just wait for the deployment to complete.

Step 7: Verify Deployment

After deployment completes (wait ~30 seconds), verify the app is running:

# Should return 200
curl -s -o /dev/null -w "%{http_code}" https://<domain>/

If it returns 200, deployment is successful. If not, check logs:

railway logs --tail 50

Step 8: Done

Print a summary:

✅ Deployment complete!
🌐 Platform URL: https://<domain>
📋 Monitor: railway logs
🔄 Re-deploy: just run deploy again

---

Notes

• Idempotency: Running this skill multiple times is safe. Each step checks current state before acting.
• Link detection: Use railway status to check if already linked — it's more reliable than checking .railway/config.json (works across machines and fresh clones).
• Re-deploy: On subsequent runs, Steps 1–5 are all skipped or no-ops. Only Step 6 (upload) actually runs.
• Secret key: Only set SECRET_KEY_BASE if not already set (check with railway variable list).
• Database migrations: Handled automatically by bin/docker-entrypoint via rails db:prepare — never run railway run bundle exec rails db:migrate as Railway's internal DB IP is not accessible from local machine.
• PostgreSQL CLI bug: railway add --database postgres always fails with Unauthorized — always use the Web UI instead.