Claude Design plugin by Anthropic: review, verdict and safety
Anthropic's Design plugin bundles seven Markdown design skills and nine MCP connectors for Cowork and Claude Code. We rate it Conditional, 66/100, at Low confidence, and it is not the Claude Design app. The skills give tidy structure to critiques, audits and handoff specs. Nobody has measured their value, and the accessibility skill carries a WCAG level error.
- Verdict
- Conditional
- Score
- 66/100
- Confidence
- Low
- Safety grade
- C (Markdown only, nine remote MCP connectors, two with empty URLs)
- Best for
- Designers and PMs in Cowork or Claude Code who want consistent critiques, handoff specs and copy reviews, and will verify output
- Poor fit
- Teams needing an accessibility audit they can cite for compliance without manual checking
- Reviewed version
- plugin.json 1.2.0 · commit da38ec1 (2026-09-24)
- Review date
- 2026-09-30
- Works in verified
- none
- Works in claimed
- Claude Cowork, Claude Code
- Evidence
- Thin (12 items)
- Status
- no change to design/ since review (checked 2026-10-21)
Get the skill: design folder on GitHub · SKILL.md at the reviewed commit · skills.sh listing
What is Anthropic's Claude Design plugin?
Anthropic's Design plugin is a free bundle of seven Markdown design skills for Claude Cowork and Claude Code. It covers critique, accessibility audits, design systems, UX copy, research and handoff, plus nine preconfigured MCP connectors.
Anthropic's plugins overview defines a plugin as one installable unit. It holds skills, agents, hooks or MCP servers. This one is named design and lives in Anthropic's knowledge-work-plugins repository. It is version 1.2.0 under the Apache-2.0 licence. Claude Cowork is Anthropic's agentic desktop app, and Claude Code is Anthropic's coding agent. Both load the plugin's Agent Skills, which are folders of instructions an agent reads on demand. Our guide to what Agent Skills are covers the format in full.
The repository holds seven skill folders at the reviewed commit:
| Skill | Job |
|---|---|
| accessibility-review | WCAG 2.1 AA audit of a design or page |
| design-critique | Structured critique of usability, hierarchy, consistency and accessibility |
| design-handoff | Developer spec with tokens, states, breakpoints and edge cases |
| design-system | Design-system audit for naming drift and hardcoded values |
| research-synthesis | Synthesis of transcripts, surveys and support tickets |
| user-research | User research support (no use case found in our review) |
| ux-copy | UX copy for errors, empty states and calls to action |
The Claude Marketplace listing shows four of these skills. The repository holds all seven.
How is the Design plugin different from Claude Design and frontend-design?
The Design plugin is not Claude Design. Claude Design is an Anthropic Labs app that builds designs and prototypes in chat. Frontend-design is a separate skill that generates interfaces, while this plugin critiques and documents them.
Three Anthropic products share the word "design".
| Entity | What it is | Where it runs | Our coverage |
|---|---|---|---|
| Design plugin | Seven Markdown skills plus nine connectors | Claude Cowork, Claude Code | This page |
| Claude Design | Anthropic Labs app for designs, prototypes and one-pagers | claude.ai, Artifacts tab, Claude Code, iOS, Android | No verdict (separate product) |
| frontend-design | Anthropic skill that generates UI code | Claude Code, via Anthropic's official plugin marketplace | Recommended, 76/100 |
Claude Design hands work to Claude Code through its /design and /design-sync commands, per the Help Center. It launched on Claude Opus 4.7, according to Anthropic's launch post. The X post "Claude Code can design now. The new /design" refers to Claude Design, not this plugin. Read Anthropic's frontend-design skill verdict for the generator side.
Is the Anthropic Design plugin worth installing?
Yes, but only as a starting structure. The Anthropic Design plugin scores 66/100, Conditional, at Low confidence. Its templates load cheaply, but nobody has measured them, one accessibility rule is mislabelled, and two connectors fail.
The main strength is compact templates. Each skill has clear output tables and works from a description or screenshot. The main weakness is generic content. The accessibility skill targets WCAG 2.1, lists 2.5.5 Target Size as AA, and two Google connectors ship empty.
Best for designers and PMs in Cowork or Claude Code who want consistent structure and will verify output. Poor fit for teams that need a compliance-grade accessibility audit without manual checks. Poor fit for users of several Anthropic role plugins who dislike duplicate connector sign-ins.
| Parameter | Score |
|---|---|
| Effectiveness | 68 |
| Reliability | 60 |
| Safety | 80 |
| Efficiency | 75 |
| Compatibility | 65 |
| Maintainability | 62 |
| Setup friction | 70 |
| Accessibility guidance accuracy | 55 |
| Connector hygiene | 50 |
| Output structure quality | 75 |
| Instruction weight vs value | 72 |
| Cost of adoption | 65 |
| Documentation consistency | 55 |
Evidence confidence is Thin. The full rubric explains how we score skills. Our ranking of the best design skills for Claude, ranked places this plugin among its peers.
How many tokens does the Design plugin use?
The Design plugin uses few tokens. Each skill loads about 450 to 1,400 tokens by our estimate. A one-screen critique costs about 4,938 input and 1,500 output tokens, our estimate for a typical task.
| Figure | Label | Value |
|---|---|---|
| One SKILL.md (user-research) | Measured file size per load | 438 tokens |
| One-screen critique or audit | Our estimate for a typical task | 3,500 input, 1,500 output (4,938 input with the skill loaded) |
The per-skill range comes from byte counts across 33 KB of skill files. An enabled plugin keeps each skill's name and description in context on every turn, per Anthropic's plugins overview. The larger cost is sign-ins. Each connector needs its own OAuth, repeated for every Anthropic role plugin (#1031).
Is the Anthropic Design plugin safe to install?
Yes, with care: the Anthropic Design plugin is safe to install at safety grade C. Its skills are plain Markdown with no scripts. The bundle preconfigures nine remote MCP servers, and two ship with empty URLs.
Grade C here reflects network reach, not hostile code. What we found in the design folder:
- Files: 11 in total, made of 7 SKILL.md files, README.md, CONNECTORS.md, plugin.json and .mcp.json.
- Code: no scripts, hooks or executable files anywhere in the folder.
- Servers: seven HTTP MCP servers at their vendors' official endpoints.
- Empty entries: Gmail and Google Calendar have no URL and fail to connect (#933).
- Slack: a static OAuth clientId and callback port 3118, added in PR #189.
- Secrets: no environment variables or API keys requested.
- Writes: tickets and wiki pages change only through connectors you sign into.
- Injection: no guidance on untrusted content pulled from Figma, Intercom or support tickets.
- Audits: skills.sh lists accessibility-review as passing Gen Agent Trust Hub, Socket and Snyk checks.
Our checklist for vetting a plugin's connectors before install walks through each of these checks.
What does the community report about the Anthropic Design plugin?
The community reports little about the Anthropic Design plugin, and no independent test exists. GitHub reports focus on connectors. Empty Gmail and Calendar entries fail every Claude Code session, and sign-ins repeat per plugin.
| Claim | Source | Date | Sentiment |
|---|---|---|---|
| Design is among 12 plugins with empty Gmail and Calendar URLs. | Issue #933 | 2026-08 | − |
| A community placeholder fix was auto-closed. | PR #1161 | 2026-09 | − |
| Shared connectors force separate OAuth per plugin. | Issue #1031 | 2026-08 | − |
| Maintainers added a Slack OAuth clientId after sign-in failures. | PR #189 | 2026-04 | + |
| Gmail and Calendar URLs were cleared to placeholders on purpose. | PR #184 | 2026-04 | neutral |
| A wrong install command in the READMEs was fixed. | PR #1209 | 2026-09 | + |
| The marketplace served an older sibling plugin version. | Issue #178 | 2026-04 | − |
| Anthropic plugins failed to install on the Linux desktop build. | claude-desktop-debian #396 | 2026-04 | − |
| CI auto-closes PRs from non-members. | Issue #1181 | 2026-09 | − |
| The accessibility skill lists 2.5.5 as AA and targets WCAG 2.1. | accessibility-review SKILL.md | 2026-09 | − |
| README command names do not match the skill folders. | design README | 2026-09 | − |
| accessibility-review has 3.8K installs and passes three audits. | skills.sh | 2026-09 | + |
What are the Anthropic Design plugin's known failures?
The Anthropic Design plugin has three known failures at commit da38ec1. The accessibility skill labels a AAA criterion as AA and ignores WCAG 2.2. Two connectors have empty URLs, and README command names do not match.
- Level error: the skill lists 2.5.5 Target Size (44px) as AA, but WCAG rates it AAA.
- Old standard: the skill targets WCAG 2.1, so 2.2's AA criterion 2.5.8 and focus criteria are absent.
- Empty connectors: Gmail and Google Calendar fail in every Claude Code session, and #933 stays open.
- Stale names: the README lists /critique, /handoff, /accessibility, design-system-management and ux-writing, none of which ship.
- Closed fixes: outside fixes get auto-closed by CI (#1161, #1181).
Our review names the worst case in these words:
An accessibility audit that looks authoritative but applies the wrong conformance level or misses issues only real assistive-technology testing would catch, leading a team to ship non-compliant UI.
The skill does tell users to test with VoiceOver or NVDA.
Which agents does the Anthropic Design plugin work with?
The Anthropic Design plugin works with Claude Cowork and Claude Code, by Anthropic's claim. We have verified neither agent. The skills.sh directory distributes single skills such as accessibility-review to other agents, which we have not verified.
| Agent | Status | Evidence |
|---|---|---|
| Claude Cowork | Claimed | README; marketplace "Install in Cowork" button |
| Claude Code | Claimed | README install commands |
| Other agents via skills.sh | Claimed, single skills only | skills.sh skill pages |
One report shows Anthropic plugins, including Design, failing to install on the community Linux desktop build (#396, April 2026). No verified agent exists for this plugin at review.
How do you install the Anthropic Design plugin?
Install the Anthropic Design plugin in Claude Code with two commands: add the knowledge-work-plugins marketplace, then install design. In Cowork, install it from its Claude Marketplace listing. Every skill works before any connector sign-in.
How do you install it in Claude Code?
- Run
claude plugin marketplace add anthropics/knowledge-work-pluginsto add the marketplace. - Run
claude plugin install design@knowledge-work-pluginsto install the plugin. - Invoke a skill by its namespaced name, such as
/design:accessibility-review.
PR #1209 corrected an earlier wrong install command in the README. Our walkthrough of Claude Code skills setup explains scopes and namespacing.
How do you install it in Cowork?
- Open the Design listing on the Claude Marketplace.
- Choose Install in Cowork.
- Sign in to a connector only when a task needs it.
The guide to skills in the Claude apps shows where installed skills appear.
How do you use the Anthropic Design plugin?
Use the Anthropic Design plugin by giving it a screenshot, description or Figma link. Then invoke a skill such as accessibility-review. It returns table-based critiques, audits or handoff specs that you check yourself.
Three tasks carry the clearest evidence:
- Critique: run design-critique on a mockup for usability, hierarchy and consistency (primary use).
- Audit: run accessibility-review for a WCAG 2.1 AA report with severity and fixes (primary use).
- Handoff: run design-handoff for tokens, states, breakpoints and edge cases (secondary use).
Design-system audits and research synthesis rest on the publisher's README only. Nobody has reported real use of them.
Test it yourself before you trust the audit:
- Run /design:accessibility-review on a screenshot.
- Include known contrast and target-size failures in that screenshot.
- Check the levels and criteria it reports against WCAG 2.2.
What are the alternatives to the Anthropic Design plugin?
The strongest alternative to the Anthropic Design plugin is the Designer Skills Collection, Recommended at 78. It covers critique, specs and research. Web Design Guidelines, Conditional at 66, audits interfaces against UI and accessibility rules.
| Alternative | Verdict | Score | When it is better |
|---|---|---|---|
| Designer Skills Collection verdict | Recommended | 78 | Broader design process coverage |
| Emil Kowalski's design engineering skill | Recommended | 76 | Interaction and motion polish |
| Frontend Design | Recommended | 76 | Generates UI rather than critiques it |
| Web Design Guidelines review | Conditional | 66 | Rule-based UI and accessibility audit |
The design hub ranks the full field by score.
Does the Anthropic Design plugin use MCP connectors?
Yes, the Anthropic Design plugin uses nine MCP connectors declared in .mcp.json. They are Slack, Figma, Linear, Asana, Atlassian, Notion, Intercom, Gmail and Google Calendar. Each needs OAuth sign-in, and the Gmail and Calendar entries are empty.
| Connector | Endpoint status | Auth |
|---|---|---|
| Slack | Official vendor endpoint | OAuth (static clientId) |
| Figma | Official vendor endpoint | OAuth |
| Linear | Official vendor endpoint | OAuth |
| Asana | Official vendor endpoint | OAuth |
| Atlassian | Official vendor endpoint | OAuth |
| Notion | Official vendor endpoint | OAuth |
| Intercom | Official vendor endpoint | OAuth |
| Gmail | Empty URL, fails to connect (#933) | None |
| Google Calendar | Empty URL, fails to connect (#933) | None |
The skills hold the method, and the connectors fetch files or post results. Our explainer on how skills and MCP servers differ covers that split in general terms.
Where is the Anthropic Design plugin on GitHub?
The Anthropic Design plugin lives in the anthropics/knowledge-work-plugins GitHub repository, inside the design folder. It is Apache-2.0 licensed at version 1.2.0, with no releases or tags. We reviewed commit da38ec1 on 2026-09-30.
- Repository: anthropics/knowledge-work-plugins, folder design/.
- Size: 11 files, plugin.json version 1.2.0, no releases or tags.
- Popularity at review: 25,935 stars, 3,049 forks and 32 contributors.
- Reviewed commit: da38ec1, dated 2026-09-24, reviewed on 2026-09-30.
- Last change to design/: c71b6f3 on 2026-09-21, which fixed the install command.
One directory page shows version 1.0.0 and 4 contributors. The repository itself shows 1.2.0 and 32.
Common questions
Does Claude Design support plugins?
The Help Center guide lists exports and Claude Code handoff, not plugins. This Design plugin runs in Cowork and Claude Code instead.
Is the Claude design plugin free?
Yes, the plugin is free under Apache-2.0. Claude Design plans differ between the Help Center (Free and paid) and the April 2026 launch post (paid only).
Is there an app for Claude Design?
The Help Center lists claude.ai, the Artifacts tab, Claude Code, iOS and Android. The standalone claude.ai/design site closes on 2026-12-14.
Do I need Figma for the Design plugin?
No. Every skill works from a description or screenshot, according to the publisher's README.
Does the accessibility skill follow WCAG 2.2?
No. It targets WCAG 2.1 AA at commit da38ec1.