rails-ai

rails-ai 🚂

⚠️ EXPERIMENTAL: Under active development. APIs and architecture may change. Phase 2 of 4.

Rails domain expertise for Claude Code. Built on Superpowers workflows.

The Vision

Build Rails apps so autonomously they leave today's Next.js and React generators in the dust. We're not playing catch-up—we're setting the pace and showing what LLM-driven development really looks like when Rails is in the driver's seat.

The Problem

Right now, LLMs excel at Next.js and Node code. Ask them to build a Rails app and you get... okay results. Ask for a Next.js app and you get production-ready code.

Why? Training data. JavaScript/TypeScript dominates the training sets. Rails, despite being more mature and productive, is underrepresented.

The irony? Rails is actually better suited for AI code generation:

• Convention over Configuration — Less decision-making, more consistent patterns
• Strong opinions — Clear right/wrong answers
• Battle-tested — 20+ years of best practices
• Monolithic by default — Simpler mental model
• Solid Stack — Rails 8's batteries-included approach

Rails was built to make developers productive by removing decisions. That same philosophy makes it perfect for LLMs — fewer choices, more consistency, clearer patterns.

What It Does

Adds Rails-specific skills to Claude Code: ActiveRecord patterns, Hotwire, Minitest/TDD, security, Solid Stack (Rails 8), and 20+ team conventions.

The /rails-ai:architect command loads Superpowers workflows (process) and Rails-AI skills (domain knowledge), then coordinates general-purpose agents to build features end-to-end.

Installation

# 1. Install Superpowers (required)
claude
/plugin marketplace add obra/superpowers
/plugin install superpowers

# 2. Install rails-ai
/plugin marketplace add zerobearing2/rails-ai
/plugin install rails-ai

Local Development Install

To contribute or test changes locally:

# 1. Clone the repository
git clone https://github.com/zerobearing2/rails-ai.git
cd rails-ai

# 2. Install Superpowers (required dependency)
claude
/plugin marketplace add obra/superpowers
/plugin install superpowers

# 3. Add local directory as a marketplace and install
/plugin marketplace add /path/to/rails-ai
/plugin install rails-ai@rails-ai

# 4. Restart Claude Code to load the plugin

Verify installation:

/plugin
# Opens plugin management interface - verify rails-ai is listed

Run tests before submitting changes:

bin/ci

Usage

/rails-ai:architect add user authentication
/rails-ai:architect debug failing test
/rails-ai:architect refactor UserController

Local Development Note: If you're developing rails-ai locally (installed from a local directory), SessionStart hooks don't execute due to a Claude Code limitation. Manually use the skill once per session:

Use the skill: using-rails-ai

This uses the full protocol including Superpowers integration and skill-loading enforcement. Not needed when installed from GitHub.

Architecture

Two-layer system:

• Superpowers = HOW to work (brainstorming, planning, TDD, debugging, code review)
• Rails-AI = WHAT you're building (12 Rails domain skills + team rules)

The /rails-ai:architect slash command loads both layers, then dispatches general-purpose agents as workers to implement features.

Philosophy

Opinionated Rails development:

• 37signals philosophy (simple, pragmatic)
• Rails conventions (REST-only)
• Solid Stack (SolidQueue, SolidCache, SolidCable)
• Minitest, not RSpec
• TDD always

Roadmap

We're turning that gap into a roadmap instead of a wish list.

Phase 1 — Architect Coordinator ✅ Complete Clean architecture with /rails-ai:architect slash command that coordinates development: loads Superpowers workflows (HOW) and Rails-AI skills (WHAT), then dispatches general-purpose workers to implement. No complexity, clear separation between coordination and execution.

Phase 2 — Domain Skills & Team Rules 🚧 Current 12 focused domain skills, 20+ team conventions, and custom RuboCop cops. Every decision backed by Rails best practices in reusable, testable modules.

Phase 3 — Memory & Context 🔬 Next Index thousands of production Rails patterns, wire them into local memory + knowledge graph, and retrieve the right snippet at the right time with RAG + SQLite vector search.

Phase 4 — Fully Autonomous Rails 🚀 Goal Ship end-to-end features—auth, payments, background jobs, admin, APIs—complete with tests and security. Target: 85%+ first-pass success rate.