skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
lwlee2608/genesis79 installs

genesis-railway

Use when deploying a project scaffolded with genesis (services/<app>-server Go API, services/<app>-web React app, Postgres) to Railway — creating the Railway project, services, variables, volume, GitHub auto-deploy, and public domain via the Railway MCP server.

How do I install this agent skill?

npx skills add https://github.com/lwlee2608/genesis --skill genesis-railway
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill automates project deployments to Railway using the Railway CLI and MCP tools. It identifies necessary environment variables by parsing local source code and configuration files, which introduces a potential surface for indirect prompt injection from repository content. It also accesses the Railway CLI's local configuration to resolve project metadata.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

genesis-railway

Deploys a genesis monorepo to Railway using the Railway MCP tools. The layout it expects: services/<app>-server (Go API with a Dockerfile and /health), services/<app>-web (nginx-served React app with a Dockerfile), and Postgres.

Railway project: <app>
├── Postgres        # Railway Postgres template
├── <app>-server    # builds services/<app>-server/Dockerfile
└── <app>-web       # builds services/<app>-web/Dockerfile, gets the public domain
        web ──▶ server ──▶ Postgres   (private networking)

Prerequisites

Railway MCP server (mcp__railway__* tools). If they are not available, stop and tell the user — do not fall back to guessing CLI commands.

Railway CLI (npm i -g @railway/cli or brew install railway), used only for railway login, railway init, railway add, railway link, and reading resolved variables.

Rules

  1. Use MCP tools for every Railway mutation. Settings, variables, volumes, sources, and domains go through mcp__railway__*. The CLI is for login, link, and railway add --database postgres only.

  2. Never add railway.json / railway.toml. Config-as-code is deprecated and stops working 2026-12-01, and no MCP tool writes it. All build/deploy settings are set on the service via update-service (step 3). Its successor, .railway/railway.ts (railway config pull/plan/apply, CLI ≥ 5), is out of scope.

  3. Create services in dependency order. A ${{Service.VAR}} reference is empty until the referenced service exists: Postgres, then server, then web.

  4. Find required variables before creating anything. Grep cmd/<app>/main.go for config checks that os.Exit — each needs a variable. Config keys map to env with . → _ (http.port → HTTP_PORT).

  5. Use the Railway Postgres template. It ships with a volume, the Database tab, backups, and DATABASE_URL / PG* variables. Only if a migration runs CREATE EXTENSION vector fall back to a pgvector/pgvector image service (see pgvector exception) — the template image lacks it.

  6. Resolve the project and environment IDs first. If the repo root is linked, read ~/.railway/config.json (.projects["<abs repo path>"].project). Otherwise list-projects, or railway init --name <app> from the repo root. Resolve the target environment with list-services; use explicit project, environment, and service IDs for configuration and verification.

Steps

  1. Create services — railway add --database postgres from the linked repo root (the remote MCP has no template tool; the CLI names the service Postgres), then create-service <app>-server, then <app>-web (no image).
  2. Variables via set-variables with skipDeploys: true:
    • Server: PORT=8080, HTTP_PORT=${{PORT}} (both — the server reads HTTP_PORT; without it the healthcheck probes the wrong port), DB_URL=${{Postgres.DATABASE_URL}}?sslmode=disable, LOG_LEVEL=info, plus anything found in rule 4 (generate secrets, report them to the user).
    • Web: BACKEND_URL=http://${{<app>-server.RAILWAY_PRIVATE_DOMAIN}}:8080
  3. Settings via update-service on server and web: rootDirectory: services/<app>-{server,web}, dockerfilePath: Dockerfile, watchPatterns: ["services/<app>-{server,web}/**"] (repo-root relative), restartPolicyType: ON_FAILURE, restartPolicyMaxRetries: 1. Server also: healthcheckPath: /health, healthcheckTimeout: 60 (startup only; justify any increase).
  4. Read back both services' live config via get-service-config; confirm step 3 and server PORT=HTTP_PORT=8080 via list-variables. If values are hidden, use railway variables --json --project <project-id> --environment <environment-id> --service <server-id> without exposing secrets. Fix and recheck mismatches; stop if unverifiable.
  5. Source — after steps 2–4, connect-service-source on server, then web: repo: <owner>/<repo>, branch: main. This starts builds and GitHub auto-deploy. Record triggered deployment IDs from the response or scoped list-deployments, matching source and trigger time; report ambiguity rather than guess.
  6. Domain — generate-domain on web.
  7. Link subdirs so the CLI works from inside them (absolute paths for cd):
    cd <abs>/services/<app>-server && railway link -p <app> -e production -s <app>-server
    cd <abs>/services/<app>-web    && railway link -p <app> -e production -s <app>-web
    

Additional infra

If docker-compose.yml has more infra than Postgres (Redis, MinIO, etc.), create each from the official Railway template, not a plain image, and do it before the server so its variables can be referenced. Fall back to create-service with image: only when no template exists.

pgvector exception

Replaces step 1's Postgres and adds what the template would have provided:

  • create-service Postgres with image: pgvector/pgvector:<tag from docker-compose.yml>.
  • set-variables (skipDeploys: true): POSTGRES_DB=<app_underscore>, POSTGRES_USER=postgres, POSTGRES_PASSWORD=<openssl rand -hex 24>, PGDATA=/var/lib/postgresql/data/pgdata (volume root isn't empty), PGHOST=${{RAILWAY_PRIVATE_DOMAIN}}, PGPORT=5432, PGUSER=${{POSTGRES_USER}}, PGPASSWORD=${{POSTGRES_PASSWORD}}, PGDATABASE=${{POSTGRES_DB}} (the PG* set turns on the dashboard's Database tab), DATABASE_URL=postgres://${{POSTGRES_USER}}:${{POSTGRES_PASSWORD}}@${{RAILWAY_PRIVATE_DOMAIN}}:5432/${{POSTGRES_DB}}.
  • create-volume on Postgres, mountPath: /var/lib/postgresql/data.
  • No Database tab in the dashboard; that only comes with the template.

Verification procedure

  1. Poll scoped list-deployments for each recorded ID until SUCCESS; never substitute another deployment. On FAILED/CRASHED, fetch its get-logs with types: ["build", "deploy"] and report the cause. If blocked or polling stops, report the ID, state, and next action as unverified. Check Postgres with environment-status.
  2. curl --fail --max-time 10 https://<web domain>/health must return the expected backend health response; HTTP errors or unexpected bodies fail verification.
  3. For proxy failures, inspect resolved BACKEND_URL via MCP or scoped CLI. If http://:8080, redeploy web and repeat verification with the new deployment ID.

Common mistakes to watch for

  • Setting only PORT. The server reads HTTP_PORT; without it the healthcheck probes the wrong port and every deploy fails.
  • Skipping PGDATA on a pgvector image service. The volume root isn't empty, so Postgres refuses to init without a subdirectory.
  • Connecting the source before variables/settings. The first build fires immediately and runs with missing config.
  • Using a plain image service when the template would do. You lose the Database tab and backups for nothing.
  • Using the placeholder project-00 in watchPatterns or service names. References like ${{project-00-server.RAILWAY_PRIVATE_DOMAIN}} resolve to empty.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/lwlee2608/genesis/genesis-railway">View genesis-railway on skillZs</a>