Patterns
The engineering playbook that keeps your codebase consistent as it grows.
Consistency at scale
When one person writes code, it's consistent by default. When multiple agents build features over weeks, consistency requires structure. The framework ships with a pattern library that every agent follows.
These aren't abstract guidelines. They're concrete, copy-from-here documents with actual code examples. When the database agent creates a migration, it follows the migration pattern. When the designer builds a page, it follows the page pattern. When the connector wires up data fetching, it follows the caching pattern.
What's covered
Frontend patterns make up the largest set. Data flow from server to client. Caching strategies that prevent loading waterfalls. Component structure that keeps client bundles small. Form handling with validation. URL-based state management. Loading states, optimistic updates, drag and drop, virtual scrolling for large lists, data tables. There's also a visual quality checklist the designer reads before building any page.
Backend patterns cover how to write API procedures, how to structure background jobs, and how rate limiting works.
Quality patterns define the checklist agents run through before marking a feature complete, and how completed features get archived so the workspace stays clean.
Workspace templates give agents a consistent format for planning documents, task breakdowns, decision logs, and context files.
Why patterns matter more than you'd think
Without patterns, you get drift. Feature one fetches data one way. Feature five fetches it another way. By feature ten, you have three different loading state approaches and nobody remembers which one is "right."
The pattern library prevents this. Agents don't reinvent solutions. They apply the documented approach. Your tenth feature uses the same data flow pattern as your first, which means debugging the tenth feature feels just as familiar.
You can change them
The patterns are files in your repository. If you prefer a different form validation approach or want to change how loading states work, you edit the pattern document and every future build picks up the change. The agents read from these files, not from training data.