Claude Code /design Skill: What the Research Preview Does

Claude Code’s /design skill, a research preview announced in Anthropic’s Week 34 digest (August 17 to 21, 2026), takes a brief and publishes a canvas of editable UI artboards as an artifact; you pick one, tweak it, and tell Claude to implement it. It runs in the CLI and Claude Code Desktop, needs v2.1.233 or later, and is available on Pro, Max, Team, and Enterprise.¹ Because it publishes to artifacts, it also falls under the artifacts prerequisites: a session signed in to a claude.ai account on the Anthropic API.⁵ The public record at launch runs to three short pages (the digest, the skill’s own description, and the artifacts availability page the digest links), so a first look stays short: what Anthropic says the skill does, what you get back, what it requires, and where the documented surface ends.

TL;DR

• One command, one canvas. /design <brief> makes Claude draft a multi-artboard design and publish it as an artifact; Claude prints the link when it finishes.¹
• Two tiers of result. Where your account allows saving, you get a visual editor and a Save that publishes a new version for everyone; everyone else gets a view-and-export preview (PNG/PDF).²
• The floor is v2.1.233 (August 14, 2026) on Pro, Max, Team, or Enterprise, in the CLI or Claude Code Desktop, plus the artifacts prerequisites: a claude.ai sign-in on the Anthropic API, with CMEK, HIPAA, and Zero Data Retention off for the organization.¹³⁵
• The preview stops at the canvas. Neither the digest nor the description lists flags or artboard limits; the description allows re-seeding without defining it, and names the published Artifact as the place to edit an existing canvas.²⁴

What Does the /design Skill Do?

Anthropic’s Week 34 digest describes it in one paragraph: the skill “brings Claude Design’s artboard workflow into the CLI and Claude Code Desktop, built on artifacts. Run it with a brief and Claude publishes a canvas of editable artboards for your UI. Pick one, tweak it, then have Claude implement it.”¹ The published example is a product brief, not a spec:

> /design redesign the composer based on what people actually use it for

Claude prints a link to the published canvas, and the digest’s card embeds a demo video.¹ You open the canvas, pick an artboard, and tell Claude which option to implement.¹ The skill owns only the first step, and its own description says so in its last sentence: “Only for CREATING or re-seeding a canvas; an existing one is edited in its published Artifact.”² The second step happens in the artifact, and the third is ordinary Claude Code work against your codebase.

The skill’s in-product description supplies the mechanics. It calls the output “a multi-artboard visual design published as an Artifact that runs Claude Design’s canvas editor (an early preview of Claude Design inside Claude Code),” drafted “as .dc.html artboards laid out on one pan/zoom canvas.”² One run puts every artboard on a single canvas, each artboard a .dc.html.

The description’s trigger clause also means you need not type the slash command. It tells Claude to use the skill “when someone wants a design, mockup, wireframe, UI or screen design, landing page, poster, flyer, brochure, banner, card, one-pager, or any visual layout they would rather tweak by hand than in code,” so a plain-language ask for a mockup can invoke /design on its own.²

What Do You Get Back?

The description splits the result by account. Where your account allows saving, “they refine every element visually (click-to-select, a properties panel, inline text editing, undo/redo) and Save publishes a new version for everyone.” Otherwise “they get a view-and-export (PNG/PDF) preview of your draft.”²

The description also reaches past app UI. It lists “UI mockups and screen flows, landing pages, marketing and social graphics, and print pieces,” with “posters, flyers, brochures as single-page artboards; memos and reports as one flowing artboard.”² The digest’s example is UI; the wider list is still Anthropic’s own scope statement.

What Does It Require?

Check your build with claude --version; the digest covers the v2.1.234 to v2.1.239 window but sets the requirement one release earlier.¹ The quickstart covers install and first-session setup.

The artifacts rows matter because the digest calls the skill “built on artifacts” and links the availability page from its /design card under “Where artifacts are available.”¹ Anthropic’s framing there is strict: “Artifacts require every condition below. When one is not met, Claude writes a local HTML file or says it cannot publish instead.”⁵ A session on an API key, or on Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry, meets the version floor and still cannot publish the canvas.⁵ The same page also describes a separate built-in design skill that styles every artifact Claude builds (v2.1.182 or later), which is not /design.⁵

Where Does the Preview Stop?

The three pages are the whole public record on the skill itself; Claude Design’s own product page describes the editor, not the skill. I would rather list what they leave out than guess.⁴

• No flags or options. The digest shows /design followed by prose. Neither the digest nor the description lists switches for artboard count, canvas size, device frame, or output tier.
• No stated artboard count. The digest’s “a canvas of editable artboards” and the description’s “multi-artboard” are the only quantities given.
• Re-seeding, undefined. The description allows “re-seeding a canvas” without saying what re-seeding takes, and names editing “in its published Artifact” as the path for an existing canvas.²
• No explanation of the saving gate. The description says “where saving is enabled for the user’s account” without naming which accounts, or how to check short of running it.² The artifacts page lists who can publish at all; whether the same conditions govern the editing tier stays unstated.⁵
• No account of implementation. The digest says you tell Claude which option to implement; whether Claude then reads the .dc.html artboard or works from your description goes unsaid.¹

The digest labels the skill a research preview, so each of those may change without notice.¹

What Else Shipped in Week 34?

Two other headline features share the digest. Concise is a new built-in output style (v2.1.237): Claude leads with the result and skips preamble while working as thoroughly as Default, and error reports, security warnings, and destructive-action confirmations stay complete; set "outputStyle": "Concise" in ~/.claude/settings.json or pick it in /config, then /clear or start a new session. Remote Control also left research preview: any machine running claude remote-control appears as a device card in the Code tab of the Claude mobile app, and tapping it starts a session there.¹

FAQ

What is the /design skill in Claude Code?

A research-preview skill that takes a brief, drafts a multi-artboard design, and publishes it as an artifact running Claude Design’s canvas editor. You pick an artboard and ask Claude to implement it.¹²

Which Claude Code version do I need for /design?

v2.1.233 or later, released August 14, 2026. Run claude --version to check.¹³

Which plans can use /design?

Pro, Max, Team, and Enterprise, in the CLI and Claude Code Desktop.¹

Can I edit the artboards /design produces?

Only where your account allows saving: then you get click-to-select, a properties panel, inline text editing, undo/redo, and a Save that publishes a new version for everyone. Otherwise you get a view-and-export preview (PNG/PDF).²

Does /design work with an API key or on Bedrock, Google Cloud, or Foundry?

Not as documented. The skill publishes to artifacts, and Anthropic’s artifacts page says sessions using an API key, gateway token, or cloud-provider credential cannot publish, and that artifacts are not available on Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry. You need a claude.ai sign-in on the Anthropic API. The artifacts page says Claude then writes a local HTML file or says it cannot publish; neither source says which /design does.⁵

Is /design only for app UI?

No. The description also lists landing pages, marketing and social graphics, print pieces such as posters, flyers, and brochures, and memos and reports as one flowing artboard.²

Key Takeaways

For product engineers: - Ask for a mockup in plain language on a real screen; the description’s trigger clause lets Claude invoke the skill from that ask, with no slash command typed.² - Write the brief as prose, as in Anthropic’s example; the documented example is prose, and neither the digest nor the description lists a flag.¹ - Keep your own record of which artboard you chose; the digest never says what Claude reads after you choose.¹

For designers: - Confirm whether your account allows saving before you plan a review round; without it, the canvas is view-and-export only.² - A one-page flyer or a social graphic is a fair test alongside a UI flow; both sit inside the stated scope.²

For team leads: - Verify v2.1.233 or later across the team, plus the artifacts prerequisites (claude.ai sign-in, Anthropic API, no CMEK, HIPAA, or Zero Data Retention), and expect the skill to change while it wears the research-preview label.¹³⁵ - Build process around the three documented steps only; flags, limits, and re-seeding remain undocumented.⁴

Once an artboard exists, designing for AI agent surfaces covers the interaction principles it still has to satisfy.