Skip to content

Ant Design Skills - Setup Guide

Source: ant-design/ant-design-cli and ant-design/antd-skill via skills.sh - 11.8K combined installs across 12 indexed listings; first seen Feb 1, 2026 (ant-design) and Mar 24, 2026 (antd) GitHub: ant-design/ant-design-cli (260 stars; MIT; pushed 2026-10-21; layout skills/antd/SKILL.md beside the CLI source and 71 bundled data files) and ant-design/antd-skill (136 stars; MIT; pushed 2026-03-27; layout skills/ant-design/SKILL.md, skills/ant-design/references/antd-cli.md, skills/antd/SKILL.md) Category: Frontend / React Design System Quality Tier: 🟢 Production - official Ant Design org, MIT on both repos, a vitest suite covering every CLI command in the CLI repo, sampled verdicts Pass on all three scanners, and a CLI push on the day of this sweep

Ant Design is the React design system behind Ant Group's enterprise products and one of the most installed component libraries in the ecosystem. This is the official publisher's own agent layer: a CLI that carries every antd prop, token, demo and changelog offline, shipped with a skill that teaches an agent when to call it, plus a sibling repo holding a decision guide for antd 6, Ant Design Pro 5 and Ant Design X v2.

The pair solves the most common failure in agent-written frontend code: the model writes an API from memory and gets a prop name, a default or a version boundary slightly wrong. Here the lookup is local and version-pinned. The CLI bundles 67 per-version snapshots (v3 2, v4 26, v5 31, v6 8) plus design-v6.md, so antd info Button --version 5.12.0 --format json answers for that exact release instead of the latest one.

Installation

# The CLI skill, from the repo that ships the CLI
npx skills add ant-design/ant-design-cli

# The decision-guide skills, from the sibling repo
npx skills add ant-design/antd-skill

# The CLI itself, which both skills call on every task
npm install -g @ant-design/cli

# Other package managers for the CLI
pnpm add -g @ant-design/cli
bun add -g @ant-design/cli

antd setup --client claude (also cursor, vscode, codex, github-actions) writes the client's MCP config or installs the bundled skill, and --dry-run previews the write.

What It Provides

Skill Installs Use For
antd 5,979 Offline antd workflows: query props, tokens, demos and docs, lint a file, run doctor checks, analyze usage, and migrate v3 to v6
ant-design 4,098 Decision guide for antd 6, Pro 5 and Ant Design X v2: component choice, theming and tokens, SSR, accessibility, performance, Pro routing and CRUD

The remaining 10 indexed listings range from 3 to 1,329 installs: antd 1,329 (an older copy of the same CLI skill, indexed from antd-skill), antd 121 and pro-upgrade 118 from the ant-design-pro repo, five maintainer workflow skills from the main ant-design/ant-design repo (antd-commit-msg 31, antd-create-pr 23, antd-issue-reply 20, antd-version-release 16, antd-test-review 15), and ant-design-pro 7 plus ant-design-x 3 from antd-skill.

Why This Matters for Hermes Agents

Frontend work is where a Hermes session burns the most retries: write a component, run the build, watch it fail on a prop that changed two minor versions ago, fix, repeat. An offline lookup that answers per version removes that loop at the source. The agent checks the API before writing the first line, which is the same discipline the skill's own mandatory rules encode.

The second gain is verification. antd lint reports deprecated APIs, accessibility gaps and performance issues in files the agent just touched, and antd doctor runs ten project-level checks covering React compatibility, duplicate packages, peer dependencies, SSR and babel plugin conflicts. Those are cheap, deterministic gates an agent can run before claiming work is done, and they speak JSON.

The third is migration. antd migrate 4 5 returns a checklist split into auto-fixable and manual items, and --apply ./src generates an agent-ready prompt without touching files, so a version upgrade becomes a reviewed plan rather than a blind rewrite. The same command set covers v3 to v4, v4 to v5 and v5 to v6.

Finally it runs as an MCP server. antd mcp exposes eight lookup tools and two prompts over stdio, which means a Hermes agent can consume antd knowledge as tools instead of shelling out, and Cursor, Claude Code or VS Code can share the same source of truth.

Usage

You say What happens
"Add a filterable table with server-side pagination" Runs antd info Table --format json and antd demo Table basic, then writes the component against the real prop surface
"This Select does not behave right on 5.12.0" Pins the lookup with antd info Select --version 5.12.0 and checks the prop against that release rather than latest
"Upgrade this project from antd 4 to 5" antd changelog 4.24.0 5.0.0 for the breaking list, antd migrate 4 5 for the checklist, then antd lint on the changed files
"Theme the Form to match our brand" antd token Form --format json and antd semantic Form --format json return component tokens and the classNames structure, tokens first instead of global overrides
"Which component should I use for a multi-step form?" The ant-design decision guide maps the need to Steps plus Form, and names the provider and validation baseline
"Something in this repo is misconfigured" antd doctor runs the ten diagnostic checks and antd env collects the full environment snapshot
"Report this antd bug" antd bug previews the issue with the collected environment, waits for confirmation, then submits with --submit

Verification

Confirm what the skills CLI installed:

npx skills list | grep -i antd

Read both skills before installing. All four URLs below returned 200 when checked on Oct 21, 2026:

curl -sL https://raw.githubusercontent.com/ant-design/ant-design-cli/main/skills/antd/SKILL.md
curl -sL https://raw.githubusercontent.com/ant-design/antd-skill/main/skills/ant-design/SKILL.md
curl -sL https://raw.githubusercontent.com/ant-design/antd-skill/main/skills/antd/SKILL.md
curl -sL https://raw.githubusercontent.com/ant-design/antd-skill/main/skills/ant-design/references/antd-cli.md

Prove the offline lookup works before trusting it in a task:

which antd || npm install -g @ant-design/cli
antd info Button --format json | head -20

Security

skills.sh verdicts for sampled skills (verified Oct 21, 2026) - re-check the security pages on skills.sh before production use:

Skill Gen Agent Trust Hub Socket Snyk
antd Pass Pass Pass
ant-design Pass Pass Pass

No sampled verdict reports malware, injection or data exfiltration, and the CLI ships its knowledge inside the npm package, so ordinary lookups make no network calls.

Read the permission block before installing, because three behaviours are writes. The skill grants Bash(antd *), Bash(antd bug*), Bash(antd bug-cli*), Bash(npm install -g @ant-design/cli*) and Bash(which antd), and the CLI copy also grants Bash(antd upgrade*). antd bug --submit and antd bug-cli --submit open a public GitHub issue, which publishes whatever the report contains. antd setup --client ... writes MCP config files into the project. antd upgrade invokes the detected package manager, which can reach the network. Set ANTD_NO_AUTO_REPORT=1 to stop the agent from suggesting bug reports at all, and always review a antd bug preview before --submit.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations

  • The CLI is the load-bearing half and it needs Node plus a global npm install. The two skills still work without it, but every workflow they describe assumes antd is on PATH.
  • The antd skill ships twice and the two copies differ: the ant-design-cli copy allows antd upgrade and claims v3 to v6 migration coverage, while the antd-skill copy allows neither and stops at v4 to v6 metadata. Install the CLI repo copy if you want the newer instructions.
  • ant-design-pro (7 installs) and ant-design-x (3 installs) are indexed from antd-skill, but that repo's tree holds only skills/ant-design/ and skills/antd/, so those two listings have no SKILL.md today.
  • The five maintainer skills are indexed from the main ant-design/ant-design repo. Probes for skills/<name>/SKILL.md, .claude/skills/<name>/SKILL.md, .agents/skills/<name>/SKILL.md and agent/skills/<name>/SKILL.md on both master and main all returned 404, so their path in that repo is unconfirmed at this snapshot.
  • Component knowledge stops where the bundled snapshots stop. A brand new release can lag until antd upgrade refreshes the data, and v3 projects are handled through antd migrate 3 4 rather than direct v3 lookups.
  • Snapshot data, verified Oct 21, 2026: 11,760 combined installs across 12 indexed listings; 260 GitHub stars on ant-design-cli and 136 on antd-skill; MIT on both; last pushed 2026-10-21 and 2026-03-27. Counts drift over time.

Pro Tips

  • Let the agent query first. The skill's own rule is antd info <Component> before writing code, and it is the single change that removes most broken component output.
  • Always pass --format json. Text output is for humans, and a parser against it breaks on the next release.
  • Pin --version to the project's installed antd when the answer matters. Without it the CLI answers for its newest snapshot.
  • Use antd migrate 4 5 --apply ./src to get a plan, not an edit. It emits a prompt and leaves files untouched, so the upgrade stays reviewable.
  • Wire antd mcp into the agent config when the session does a lot of component work. Eight tools over stdio beat shelling out per question, and they return the same offline data.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing