railwayapp/railway-skills

Railway Skills – What It Is

Railway Skills is a collection of agent skills that let large‑language‑model (LLM) agents talk to Railway, a cloud platform for deploying and managing apps. The repo follows the open‑source Agent Skills format and bundles ready‑to‑use plugins for a handful of popular LLM‑powered environments:

ChatGPT / OpenAI Codex
Claude (and Claude Code)
Cursor
Grok Build (xAI)
Google Antigravity

Each plugin packages the same core skill – use‑railway – together with a small “MCP” (Message‑Control‑Protocol) server configuration that lets the LLM call Railway’s APIs securely (OAuth/OIDC). The plugins are published to the respective marketplaces, and the repo also provides the raw manifests so you can install the exact version directly from GitHub.

How to Get Started

Install the Railway CLI and enable agent support

curl -fsSL agents.railway.com | sh   # installs CLI + agent bits
railway login                       # if you aren’t already logged in

If you already have the CLI, just run railway setup agent.

Pick your LLM environment and follow the short instructions in the README:
- ChatGPT / Codex – add the official Railway plugin from the ChatGPT plugin directory, or add this repo as a marketplace source.
- Claude / Claude Code – install via the Claude connector or the Anthropic marketplace commands shown.
- Cursor – add the plugin from the Cursor Marketplace or paste the repo URL in Cursor’s plugin settings.
- Grok Build – install from the xAI marketplace inside the Grok TUI.
- Google Antigravity – install from a custom URL inside Antigravity.
Authenticate when the plugin prompts you; it links your Railway account to the LLM so the agent can create services, run commands, read logs, etc.

What the Skill Actually Does

The repository ships a single installable skill:

use-railway – defined in plugins/railway/skills/use-railway/SKILL.md.

The skill is route‑first: the LLM first matches the user’s intent (e.g., "deploy my Node app" or "show the logs for service X") to a predefined route in SKILL.md. Once the route is selected, the skill executes a series of low‑level actions (API calls to Railway’s backend) that perform the requested operation.

Because the routing logic lives in a markdown file, developers can extend or customize the skill without writing code – just add new intent patterns and map them to existing or new Railway API actions.

Who Might Use This?

LLM‑powered dev assistants that need to spin up or manage Railway projects on the fly.
Teams building custom AI agents (e.g., internal bots that can provision environments, fetch logs, or trigger deployments).
Tool makers who want a ready‑made Railway connector for the major LLM platforms.

License

MIT – you can freely reuse, modify, and redistribute the skill definitions and plugin manifests.