Claude Code design plugin with Agent Skill and MCP tools · Early-stage versioned project

referodesign/refero_skill

MIT-licensed Claude Code plugin and portable Agent Skill that requires reference-led design research, routes work through bundled craft guidance and a read-only Refero MCP service, and uses repository-reported access to 150,000+ app screens and 6,000+ user flows for evidence-backed UI, product, and web design.

UI designUX designProduct designWeb designDesign researchFrontend stylingDesign systemsAccessibilityCopywritingVisual QA
Routing score
80.0
Readiness
Usable with service and verification caveats
License
MIT
Maintenance
active
Components
20
Revision
0

Selection

Select when

  • You want a packaged Claude Code plugin combining a design skill with read-only MCP research.
  • The task needs evidence from real interfaces rather than generic model taste.
  • A landing page or redesign needs a distinctive, explicitly locked visual direction.
  • A product screen needs concrete component, hierarchy, state, or copy references.
  • A multi-step workflow needs examples of journey sequencing and system responses.
  • You want major design decisions traced to sources or user constraints.
  • You need bundled craft guidance even when live Refero access is unavailable.
  • You want a workflow that includes rendered visual comparison before handoff.

Boundaries

Avoid when

  • You cannot send research queries to Refero and bundled methodology alone is insufficient.
  • You need free live access because the repository states live research requires a paid plan.
  • You need an in-app or iOS-specific style system rather than screen-level examples.
  • You need direct manipulation of Figma or another design editor.
  • You want a minimal coding-only task with no meaningful design decision.
  • You require independently audited corpus provenance, MCP security, or privacy behavior.
  • You cannot accommodate mandatory research and reference-lock steps.
  • You need deterministic design outcomes without human visual review.

Strengths

Capabilities

Installs as a native Claude Code plugin through the documented plugin marketplace flow.Packages one canonical refero-design SKILL.md with progressively disclosed craft references.Connects Claude Code plugin installs to the Refero MCP production endpoint through repository manifests.Provides read-only MCP research for visual styles, real product screens, and multi-step user flows.Uses repository-reported access to 150,000+ real app screens and 6,000+ user flows.Works without a Refero account as a bundled research-first methodology and craft reference.Routes work among direct build, visual exploration, audit, and asset-generation workflows.Requires visual work to begin with style research when live Refero MCP tools are available.Uses screens to investigate concrete layouts, components, states, copy, and calls to action.Uses flows to investigate onboarding, checkout, cancellation, settings, and other journeys.Creates a reference lock that records dominant traits, bounded borrowed details, token roles, and media strategy.Creates a decision ledger tying major choices to research, user constraints, or craft rules.Provides bundled guidance for typography, color, spacing, motion, icons, accessibility, performance, and copywriting.Includes anti-averaging and anti-AI-slop quality gates intended to preserve distinctive reference traits.Requires multiple references to be synthesized rather than copying one source.Requires substantial rendered work to be compared with the locked target before handoff.Supports manual or Skills CLI installation of the standalone skill when MCP is already configured or not needed.Also packages first-party metadata for Codex, Cursor, Gemini CLI, and the MCP Registry.Documents release checks for manifests, versions, links, endpoint consistency, package boundaries, and common secret leaks.Provides a private vulnerability-reporting channel with a stated 48-hour response target.

Risk profile

Risks and limitations

  • Uncertainty: only selected first-party files were supplied, so MCP manifests, bundled references, release scripts, platform validation results, and server implementation were not inspected.
  • Uncertainty: repository metadata is dated in 2026 relative to this analysis context, so activity, adoption, and recency cannot be independently reconciled.
  • Uncertainty: claims of 150,000+ screens, 6,000+ flows, curation quality, and well-designed source products were not independently verified.
  • Uncertainty: the README calls the MCP connection read-only, but no server code, tool schema, or independent security assessment was supplied.
  • Live research requires a paid Refero plan and OAuth authentication, while pricing and service availability can change independently.
  • Refero styles currently cover web marketing and product pages, not in-app dashboards, authentication, settings, or iOS screens as complete style systems.
  • Bundled craft references can support offline methodology, but they do not provide the same live corpus evidence as MCP research.
  • Mandatory multi-reference research, synthesis, and visual QA may be excessive for trivial or time-sensitive edits.
  • Generated assets and raw screenshot inspection depend on host capabilities and tools not established by the supplied files.
  • No minimum Claude Code version, compatibility matrix, automated host test results, telemetry disclosure, or independent audit was supplied.
  • Design evidence can still be misapplied, and accessibility, originality, conversion claims, and visual quality require qualified human review.
  • The plugin depends on a hosted production endpoint, so outages, API changes, account state, or corpus changes can affect live functionality.

First-party files describe the MCP connection as read-only, state that manifests contain no access tokens, and use OAuth for live access. SECURITY.md directs private reports to support@refero.design and states a 48-hour response target. The documented release check covers manifest consistency and common secret leaks. Uncertainty: the MCP configuration, server implementation, release script, platform validation, and any audit evidence were not supplied, so read-only enforcement, OAuth storage, transport handling, and server-side controls could not be verified.

Component inventory

20 documented components

Claude Code plugin

Refero for Claude Code

Installs the Refero design skill and read-only Refero MCP connection as a native Claude Code plugin.

.claude-plugin/plugin.json
Agent Skill

refero-design

Research-first UI and product-design methodology with craft references, reference locks, decision ledgers, and visual quality gates.

skills/refero-design/SKILL.md
read-only MCP service

Refero MCP

OAuth-backed live research connection for Refero styles, real product screens, and user flows.

README.md
MCP tool

refero_search_styles

Searches semantic style references extracted from real web marketing and product pages.

skills/refero-design/SKILL.md
MCP tool

refero_get_style

Retrieves a full style reference with visual thesis, tokens, typography, composition, components, imagery, and implementation guidance.

skills/refero-design/SKILL.md
MCP tool

refero_search_screens

Searches real interface screens for concrete layouts, components, states, copy, and CTA patterns.

skills/refero-design/SKILL.md
MCP tool

refero_get_screen

Retrieves full details for a selected product screen.

skills/refero-design/SKILL.md
MCP tool

refero_get_similar_screens

Expands research from a strong screen to related interface examples.

skills/refero-design/SKILL.md
MCP tool

refero_get_screen_image

Retrieves a raw screenshot image for direct visual inspection.

skills/refero-design/SKILL.md
MCP tool

refero_search_flows

Searches real multi-step journeys such as onboarding, checkout, cancellation, deletion, and password reset.

skills/refero-design/SKILL.md
MCP tool

refero_get_flow

Retrieves step-by-step goals, actions, system responses, and completion states for a selected flow.

skills/refero-design/SKILL.md
design workflow

Direct build

Researches and locks a clear direction, then proceeds directly to implementation.

skills/refero-design/SKILL.md
design workflow

Visual exploration

Produces reference-locked visual options before implementation, defaulting to three alternatives for selection.

skills/refero-design/SKILL.md
design workflow

Audit

Uses screenshots, Refero screens, or flows as evidence for interface critique.

skills/refero-design/SKILL.md
design workflow

Asset generation

Generates bitmap media only when a locked reference requires assets that code, icons, or existing media cannot provide.

skills/refero-design/SKILL.md
ChatGPT/Codex plugin

Refero OpenAI plugin

Directory-installed OpenAI plugin bundling Refero MCP and Refero Skill with OAuth sign-in.

.codex-plugin/plugin.json
Codex app/CLI plugin

Refero Codex plugin

GitHub-installed Codex plugin exposing the Refero skill and MCP server.

.codex-plugin/plugin.json
Cursor plugin

Refero Cursor plugin

Cursor package metadata points to the shared skills directory and Cursor MCP manifest.

.cursor-plugin/plugin.json
Gemini CLI extension

Refero Gemini CLI extension

Gemini CLI distribution of the Refero package through its extension manifest.

gemini-extension.json
maintainer command

Release validation

Checks manifests, versions, local links, the canonical MCP URL, package boundaries, and common secret leaks.

scripts/check-release

Technical profile

Requirements and configuration

Repository
referodesign/refero_skill
License
MIT
MetadataLanguage
Python
PluginVersion
1.0.2
DefaultBranch
master
ClaudePackaging
.claude-plugin/plugin.json plus skills/refero-design/ and .mcp.json
SkillEntry
skills/refero-design/SKILL.md
McpEndpoint
https://api.refero.design/mcp
McpTransport
HTTP as shown in the documented manual Claude Code setup
Authentication
OAuth browser sign-in for live Refero access
ServiceAccess
Paid Refero plan required for live research
ToolAccess
Described as read-only
OfflineMode
Bundled methodology and craft references work without an account
Validation
README documents scripts/check-release and separate native clean-profile install tests
RepositoryStatus
Public, unarchived, 249 stars, 21 forks, and 2 open issues in supplied metadata

Classification

How it enters the stack

Context InjectionTool SurfaceOrchestration

Evidence: a Claude Code plugin manifest packages the refero-design SKILL.md and MCP configuration; the skill injects design methodology, the read-only MCP exposes research tools, and the workflow routes among styles, screens, flows, and build paths. Live use adds OAuth and a paid plan.

Claude Extension · low setup effort · high confidence · automated

Evidence and risk

Primary sources

first_party_fileREADME.mdhttps://github.com/referodesign/refero_skill/blob/master/README.md
first_party_fileSECURITY.mdhttps://github.com/referodesign/refero_skill/blob/master/SECURITY.md

Routing context

Conflicts, complements, and synergies

overlaps_with

gh_owl_listener_designer_skills

Both extend Claude Code with design guidance, but Refero centers one research-first skill and hosted visual corpus while designer-skills offers a broad multi-plugin design-practice library.

high confidence
overlaps_with

gh_mengto_skills

Both provide portable design and frontend skills, but Refero adds a native Claude plugin and live read-only interface research while MengTo/Skills spans a much broader skill collection.

medium confidence
complements

gh_dzhng_skills

Refero supplies research locks and design evidence, while dzhng/skills includes broader implementation loops and screenshot-oriented visual review workflows.

medium confidence