Back

UXCO Build v0.1

UXCO Build v0.1

Turning design expertise into an AI agent capable of analyzing, critiquing, and building digital products directly on the canvas.

Access project on Github

About the project

UXCO Build is an AI-powered tool I am developing to operate as a specialized Product Design agent. Connected to Paper, it understands product context, analyzes interfaces, identifies problems, explores solutions, and works directly on the canvas to help designers make better decisions and accelerate their work.

Rather than simply generating screens, UXCO Build incorporates a proprietary design intelligence system that I developed through specialized methods, quality frameworks, skills, and structured workflows.

The tool is designed to understand the problem before designing, account for requirements and edge cases, challenge its own solutions, and maintain consistency throughout the design process.

The project represents my exploration of a new way of practicing Product Design: not simply using AI tools, but turning specialized knowledge into systems capable of reasoning and execution. With UXCO Build, I combine my experience in design, strategy, and leadership with AI agents, context engineering, MCPs, and AI-assisted development to build a tool that expands what designers can accomplish — without replacing the judgment that makes their work valuable.

How the UXCO Build harness works

  1. Instructions — CLAUDE.md: The entry point, always loaded. It defines the agent's identity, design principles, and now (§11) the operational
    aspects: which runtime to use, which commands to run, and the session start/end protocol (read PROGRESS.md, check
    git status, and finally run npm run verify and update the handoff).

  2. Tools — scripts/preflight.mjs: Is the environment ready for work? It checks Node, Git, Claude Code, and the
    Paper MCP endpoint.

  3. Environment — package.json + .nvmrc: Declares the runtime (Node ≥18, developed on Node 22) and explicitly states that the project has no external dependencies.

  4. State — PROGRESS.md: The project's short-term memory between sessions: where we stand, what is working, what is blocked, and the next step. It does not replace Git or CHANGELOG.md—it is simply a living summary of the current state.

  5. Feedback — npm test + npm run verify: Confirms whether behavior remains correct and if the repository is consistent enough to consider a task complete.

In practice: a new session reads CLAUDE.md → PROGRESS.md → runs npm run verify—and immediately knows the project's status and whether the existing state is reliable.