What is design-os?

design-os is a personal design system meant for AI coding agents like Claude or Codex, not for humans writing code by hand. It exists to fix a common problem: when an AI agent builds a user interface without any reference, it tends to produce a generic look, a centered hero section, a gradient headline, three cards, and a few emoji, the kind of page that feels like nobody in particular made it. The project addresses this with plain text files that any agent able to read local files can load. The system is organized into four working parts plus optional extras. First, a judgment section lists concrete standards for what counts as good design and what counts as an obvious AI look, along with a checklist the agent runs before calling work finished. Second, a techniques section offers layered recipes and cards describing specific interaction patterns. Third, a references section holds an engine that requires the agent to gather quality reference examples before writing any code, and it can connect to a personal style library the user builds themselves. Fourth, a workflow section defines a fixed collaboration process and a preferences file the agent updates automatically each time the user corrects or approves its work, so the system gradually learns that person's taste. A fifth, optional section contains material meant for humans who want to train their own design eye. Setup involves cloning the repository, installing a short skill file into the agent's skills folder with a path placeholder filled in, then simply asking the agent to start a project using design-os. It is not a component library or CSS framework and is not tied to any particular tech stack. The project itself ships no third-party design assets such as fonts or screenshots, since those carry their own licensing, users are expected to build their own optional reference library separately. The methodology, protocols, and research notes are released under the MIT license.