If you use Claude Code, Cursor, or Windsurf for Rails development, you've probably noticed the AI doesn't always follow your team's conventions. It might generate service objects when you prefer rich models, or scaffold a test suite in RSpec when your codebase uses Minitest.
Developer Julian Rubisch released three free, installable "skills" that solve this by teaching AI coding agents specific Rails patterns. You install them with a single terminal command (npx skills add julianrubisch/skills), no signup required.
The three skills each target a different workflow:
- jr-rails-classic follows 37signals conventions: rich models, thin controllers, Minitest with fixtures, and no service objects
- jr-rails-new walks the AI through setting up production-ready Rails apps with Phlex or ERB templating, Pundit authorization, Solid Queue, and AASM state machines
- jr-rails-phlex focuses on reusable UI components with Stimulus and Turbo integration plus caching patterns
The concept behind "skills" (sometimes called rules or instructions files) is straightforward: you give your AI coding agent a set of written conventions, and it follows them instead of guessing. Think of it as a .editorconfig for AI behavior. The difference here is that someone has pre-written detailed Rails-specific conventions so you don't have to.
This is a niche release, but it points to a pattern worth watching in AI-assisted development. The gap between "AI can write code" and "AI can write code the way my team writes code" is where most frustration lives. Shared, installable convention packs are a practical answer to that problem.