Ryugi gives coding agents a drawing board instead of a chat window

Ryugi is a new artifact system for coding agents, from the developer Innei, with 403 stars on GitHub. It was created on 19 October 2026, so the traction is real inside a 30-day window. It is a fork of Superpowers v7.0.0 by Jesse Vincent and Prime Radiant. It works with Claude Code, Codex, Cursor, Gemini CLI and OpenCode. The README also lists Antigravity, Devin CLI, Factory Droid, Kimi Code, Pi, Qwen Code, Hermes Agent and Muse.

The breakdown it names

The README puts the point in one line. The project says: "Ask an agent for a settings page and you get three paragraphs describing three layouts". Ask for a design and you get a wall of Markdown you cannot point at. The breakdown is the chat window itself. The agent describes its options. The user has to build a picture in their head to pick one. Every design discussion becomes a thread of guessing.

The five surfaces

Ryugi gives the agent five drawing surfaces instead of chat, each a specific tool for a specific task.

• Artboard. The agent draws every option side by side and marks the one it recommends.
• Document. The agent asks its questions inside the doc and links the boards it drew.
• Whiteboard. You both draw on it. Select what you drew. Ask about it. The answer lands next to your sketch.
• HTML. A one-page bench whose sliders keep their values when the agent rewrites the page.
• Verify. The built page sits next to the design you approved. Pass it, or reject it and say what is off.

A workflow stitches the surfaces together. The agent goes from idea to brainstorm to artifacts to spec to a test-driven build. A verify step compares the result to the approved design. A pass or a reject sends a specific note to the agent.

A note on the license

Ryugi is not MIT. The repository uses the Ryugi License 1.0, which is source-available, not open source. A team cannot pick it up and use it in a commercial product without reading the licence first. The fork has carried over the Superpowers v7.0.0 history on MIT, so historical contributions keep their original licence. The npm artifacts application is proprietary. Read this before adding the tool to a team's workflow.

Where the real gain is

Most of the review time on an agent-written feature goes into questions the agent did not know it should ask. A chat with an agent produces text. The text is shared, read and argued about. Nobody has pointed at a layout. Ryugi makes the agent draw the layouts, which turns the review question from "what did it mean" to "which one". That is the step that has been missing in the agentic-coding workflow for designers and product managers. 400 stars inside 30 days say the breakdown is widely felt.

Source

Innei/ryugi on GitHub.