Doctor

RefineUI Doctor inspects projects that use @refineui/* and reports setup gaps, token contract violations, and component API misuse — without modifying your code.

What it checks

Rule	Category	Description
package-setup	setup	App projects declare @refineui/react and @refineui/tokens
stylesheet-setup	setup	Global CSS imports tokens, icons, and refineui.css
foundation-contract	foundations	No hardcoded hex/rgba or arbitrary px in component styles
composition-api	components	No convenience props (title, description, …) on RefineUI components
component-imports	components	Import from @refineui/react root only

Diagnostics are read-only. Doctor writes a YAML report; it does not install packages or edit files unless you ask an agent to fix findings separately.

Install

npmyarnpnpmbun

npm install -D @refineui/doctor

CLI

refineui-doctor
refineui-doctor ./apps/web
refineui-doctor --workspace packages/my-app --output ./doctor-report.yaml
refineui-doctor --category setup --category foundations

Exit code 1 when error-severity findings exist.

Skill workflow

Install the RefineUI skill, then ask in natural language:

RefineUI 잘 쓰고 있나 봐줘
셋업만 진단해줘
이 모노레포의 RefineUI 워크스페이스를 전부 진단해줘

The skill loads doctor rules and routes to live docs via llm.txt — it does not duplicate documentation.

MCP

When @refineui/mcp is configured, agents can call run_doctor with an optional project path and workspace name. The tool returns JSON matching the YAML schema.

Report format

Schema v2 YAML written to a temp directory or --output:

schemaVersion: 2
meta:
  target: /path/to/project
  projectKinds: [app]
  refineui:
    declared:
      "@refineui/react": "^0.x"
summary:
  error: 0
  warn: 1
  info: 0
checks: [...]
findings: [...]

Each finding includes remediation text suitable for agent-guided fixes.

vs Skill vs MCP

	Doctor CLI	Skill	MCP run_doctor
Runs locally	Yes	Agent follows rules	Yes
Best for	CI / scripts	Natural language requests	Agent automation
Output	YAML file	Conversation + optional YAML	JSON

Use Skill for interactive diagnosis, CLI for repeatable checks, MCP when the agent should run diagnostics programmatically.