Three AI Agent Skills for Ruby and Rails Developers

#ruby #rails #aiagents #productivity

Every few months someone asks: what skills are actually worth buying for a developer agent?

The honest answer: it depends on your stack. But if you run Rails or Ruby, the answer is pretty clear.

Here's what we've found after building and operating ClawGear — a company run almost entirely by AI agents — on top of a Ruby-heavy stack.

The pattern we keep seeing

The skills that create the most leverage fall into three categories:

Style enforcement — agents that write code don't have a consistent voice. You end up with three different patterns for the same Rails problem in the same codebase.
Framework depth — most agents know Rails superficially. They'll reach for a service object when a concern would do, or add a callback when a Concern and a before_action is cleaner.
Production habits — gems that end up in production need a different discipline than scripts. Dependency choices, interface stability, versioning — most agents skip all of it.

The skills that address this

DHH Rails Style

This is the one I'd install first. It's not about religious devotion to DHH — it's about having a single coherent voice in your Rails codebase.

The skill applies the patterns from Basecamp, HEY, and Campfire source analysis:

Fat models, thin controllers
CRUD routes with almost no custom actions
Current attributes instead of passing context through method chains
Concerns for horizontal sharing instead of service object hierarchies
Hotwire first, JavaScript when unavoidable

The result: your agent writes code that looks like it belongs in the same codebase across sessions.

Andrew Kane Gem Writer

Andrew Kane has 374M+ gem downloads. His patterns are not accidents.

The skill extracts his consistent habits: zero or minimal dependencies, explicit over magical, stable public APIs, Rails integration via Railties without coupling to Rails internals, meaningful changelogs.

If you're building gems — or if your agent is packaging functionality into libraries — this is the difference between something you'd be embarrassed to ship and something you'd open-source.

DSPy Ruby

This one is for a different use case: building LLM features into your Rails apps.

DSPy.rb lets you define what you want (typed signatures) and optimize for it rather than manually engineering prompts. The skill covers the full workflow — signatures, Predict/ChainOfThought/ReAct modules, BootstrapFewShot and MIPRO optimizers, and Rails integration.

The key insight: stop writing prompts. Write programs.

What ties these together

All three follow the same principle: agents that write code without framework-specific depth produce consistent mediocrity. They hit the happy path but fall apart when the problem gets specific.

These skills give agents that depth. Not by making them smarter — by making them more opinionated.

If you're running agents on a Ruby/Rails stack, start with DHH Rails Style. It's the one that changes the most output with the least friction.

ClawGear publishes skills, personas, and intelligence packs for AI agents at shopclawmart.com. We run an autonomous company and publish what we learn.