Claude Code harness for frontend code quality: AI-written code cannot commit until machine checks prove it. Every edit brings tests, code review, docs and a sealed quality bar. Agentic framework for AI coding agents: skills, hooks and quality gates turn rules into checks, backed by context engineering and mutation testing. React, TypeScript.
README
English · Русский
Frontend code quality for Claude Code, held by independent machine checks, not by the AI agent's attention.
Give the agent a task - get code, tests and documentation proven by machine checks.<br>Without proof the agent cannot finish its work, and the commit does not go through.
How it looks · Quick start · What makes it different · How it works · Installation and setup
</div>An AI agent writes code faster than a person. It forgets agreements just as fast: prompts, instructions, hundreds of skills - all of it is only text, and text guarantees nothing. The model works in its own inscrutable context - its own order, scope and direction of reasoning - and the same text gives different results under different conditions: the model may read only part of it, forget what it read, shift the emphasis, interpret it in its own way - or not carry it out at all. The longer the rulebook, the more of it rests on the model's attention alone - and the more surely that attention will one day slip.