ryoo-ghee · [ɾʲɯːɡʲi] · Japanese for “a school's own way of doing things”
Artifacts for coding agents.
Your agent draws its options instead of describing them. You pick, comment and pass.
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 can't point at.
Ryugi gives the agent pages to put that work on, and gives you a way to answer on them:
- Artboard — every option drawn side by side; pick one.
- Document — a design doc with the agent's questions inline; comment on any sentence.
- Whiteboard — boxes and arrows you both draw on.
- HTML — one page with real sliders to tune a component or a motion.
- Verify — the built screen next to what you approved, and a report of what the agent proved by running it; pass or reject.
Your picks, comments and verdicts go straight back to the agent session that made the page. Every artifact lives in your project, opens from one local server, and becomes a link when you want to share it.
npx ryugi-artifacts installThe terminal wizard installs a persistent ryugi command, lets you select one or more
harnesses, then asks which content to install:
- Artifacts only — send
Show me three directions for a settings pageand an artboard opens. - The full workflow — artifacts plus skills that brainstorm, spec, build under TDD and verify.
Send
Let's make a react todo listand the agent opens an artboard before it writes code.
Harnesses · CLI · Full workflow · Ryugi Cloud
Use it when there is more than one way it could look.
The agent draws every option side by side and marks the one it recommends. Pick, comment on any pixel, and Submit sends your answer back.
Use it when the decision is structure, architecture or tradeoffs.
The agent asks its questions inside the doc and links the boards it drew. Comment on a sentence or edit in place; your edits and the agent's merge live.
Use it when the parts and arrows are still the question.
You both draw on it. Select what you drew, ask about it, and the answer lands next to your sketch.
Use it when one thing needs tuning: a component, a type scale, a motion.
A one-page bench whose sliders keep their values when the agent rewrites the page.
Use it when the work is built and you decide whether it matches.
The built page sits next to the design you approved. Pass it, or reject it and say what is off; it comes back fixed.
What a screen can't show goes in a verify report: one line per claim the agent checked by running something — a test run, an API response, a migration count, which build is deployed — with the raw output one click away, and a "Not verified" list of what it couldn't check and why. You read it top to bottom and pass, reject or waive; screens still get judged on the artboard, and the report links to them.
Use it when you'd rather point at the thing than describe it in the terminal.
Pin a comment on a board, a sentence or a sketch, and it reaches the agent session that opened the page. In Claude Code it arrives on its own, with no message from you; the agent applies it, replies under your comment and updates the page while you watch. Questions, picks and verdicts travel the same way.
Use it when you want last week's mockup back, or a link to share.
npx ryugi-artifactsLocal first, no subscription, no agent session needed. One server covers every project on the
machine, with ⌘K to jump and links that survive a restart. publish shares one to
Ryugi Cloud.
For a persistent ryugi command without a harness plugin:
npx ryugi-artifacts install --mode artifacts --harness none
ryugi
ryugi /path/to/projectNode.js 20.12 or newer is required. The installer keeps a complete runtime outside
the npx cache and creates ~/.local/bin/ryugi (a command shim on Windows). It prints
PATH instructions for your shell. npm install -g ryugi-artifacts remains available;
package-manager installations keep their original update channel. ryugi update
updates the CLI and recorded plugins; ryugi doctor checks the install and
ryugi uninstall removes owned entries while retaining artifacts and user data.
See the CLI guide for commands and options.
Artifacts work on their own. If you also want the agent to build from what you approved, pick the
full workflow at install (or run ryugi install --mode full --harness codex,pi later). Its skills make
the agent reach for the artifacts at the right moment, and they fire on their own:
flowchart LR
I[Idea] --> B[Brainstorm] --> A[Artifacts] --> S[Spec] --> T[Build · TDD] --> V[Verify] --> L[Land]
After you sign off, the agent slices the spec into tasks, proves each one red-green, keeps a ledger of judgment calls across compaction, gets one fresh review of the branch, and reports how to land it. Debugging and worktrees are covered too. How it works →
Have Node.js and npm available in the agent's environment. The plugin calls its own launcher,
which downloads ryugi-artifacts on first use and caches each release under
~/.cache/ryugi/artifacts/ by default. It checks for updates at most once a day and can reuse a
compatible cached release offline. The first run needs network access and a writable cache.
Ryugi replaces Claude's hosted Artifacts with the Artboard, documents, whiteboards and HTML
benches. Turn the hosted ones off by adding this key to the top level of ~/.claude/settings.json —
keep your other settings:
{
"disableArtifact": true
}That is the only flag Ryugi needs; leave the rest of your Claude Code settings alone.
Other harnesses
| Harness | Command |
|---|---|
| Antigravity | agy plugin install https://github.com/Innei/ryugi |
| Codex, Cursor | Add this repo as a local plugin source (detail) |
| Devin CLI | devin plugins install Innei/ryugi |
| Factory Droid | droid plugin marketplace add https://github.com/Innei/ryugi then droid plugin install ryugi@ryugi |
| Gemini CLI | gemini extensions install https://github.com/Innei/ryugi |
| Kimi Code | /plugins install https://github.com/Innei/ryugi |
| OpenCode | .opencode/INSTALL.md |
| Pi | pi install git:github.com/Innei/ryugi |
| Qwen Code | qwen extensions install Innei/ryugi |
| Hermes Agent | hermes plugins install Innei/ryugi --enable |
| Muse | muse plugins install ./ then muse plugins approve ryugi |
Updates and caveats: docs/install.md.
Fork of Superpowers v7.0.0 by Jesse Vincent and Prime Radiant (UPSTREAM.md). Issues and PRs welcome; AGENTS.md has the rules.
Skills, harness integrations and docs: Ryugi License 1.0 — free to use, modify and redistribute, with two exceptions. The Restricted Parties in Schedule A (Tencent, Alibaba, ByteDance, Baidu, Huawei, Xiaomi, Meituan, JD, PDD, NetEase, Kuaishou, Didi, Ant Group, iFlytek, SenseTime, and products such as WorkBuddy, Qwen, Qoder and Trae) receive no rights at all, and nobody may scrape the skills into a marketplace. This is source-available, not an open source license; material derived from Superpowers and contributions made before 2026-11-04 stay MIT (Exhibit B).
The artifacts application, published to npm as ryugi-artifacts, is proprietary (npm license: UNLICENSED) and not part of this repository. Releases up to 1.51.2 were published under AGPL-3.0-only.
