Image: GitHub
Why it mattersCoding agents write the markup. They rarely see the result. A tool that feeds them screenshots closes that loop in minutes.
A coding agent can write every line of a page and never look at the page. The CSS compiles. The component renders. The agent moves on. The person reviewing the diff is doing the only visual check in the loop.
ShuaiMXu released HappyHands on 10 November on GitHub. The repository held 487 stars on 13 November. The README describes the tool in one sentence: "open-source, AI-assisted UI design review for coding agents. Inspect screenshots, scope code changes and compare results in a local workbench." The licence is MIT. The project is in alpha, with four releases between 10 and 12 November.
What it does
The tool runs on the developer's own machine. The agent takes a desktop or phone screenshot of the page. It picks a scope of the markup to change. It writes the change in an isolated copy of the code. Then it renders the page again. HappyHands lays the before and after side by side. The agent judges the result against the first screenshot. The author lists three things for the agent to look at. These are information hierarchy, interaction feedback and user-task flow. Pixel-perfect alignment is not the point.
Which agents
The README names Codex CLI as the built-in integration. Claude Code has an adapter. Cursor, Kimi Code, WorkBuddy and the DeepSeek Harness can drive the tool through its CLI. The agent needs to know the HappyHands commands. The author also publishes a short skill file. The agent loads that file at the start of the session.
What it needs to run
HappyHands is a Node.js tool. It needs Node 22.15 or later and pnpm 11.19. The install is a single npx call that points at the latest release. The README holds a 37-second demo video that shows the full loop on a real page.
The alpha caveat is loud in the release notes. The author rejects HTML viewers, redirects and binary input. Every design reference the agent uses must be a bounded, hashed file. That is a short list of things the tool will refuse. It is not a list of what it supports. Expect rough edges on anything beyond a plain web page.
The reason to read it anyway is the gap it names. A coding agent that writes UI without seeing the UI is the shape of most AI coding work today. A local screenshot loop driven by the agent is one of the few ways to close that gap. If it holds up in practice, the team that adopts it writes smaller review tickets.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.