A surprisingly rich collection of Claude skills, MCP servers, and community resources already exists for Apple Human Interface Guidelines expertise, with at least a dozen dedicated tools that can differentiate between iOS, macOS, visionOS, and other Apple platforms. The most capable options combine SKILL.md files for Claude Code with MCP servers that accept platform parameters, enabling context-aware design guidance without manual switching. No single “official” Anthropic skill covers Apple HIG, but the community has filled this gap so thoroughly that the real challenge is choosing among competing solutions rather than building from scratch.
The Claude Code skills ecosystem has produced multiple purpose-built Apple HIG packages, each taking a different approach to encoding design knowledge. These install as SKILL.md files in your .claude/skills/ directory and auto-activate when Claude detects relevant design work.
ehmo/platform-design-skills (295 stars on GitHub) stands out as the most platform-aware option. It contains 450+ rules across separate skill files for every Apple platform — iPhone, iPad, macOS, watchOS, visionOS, and tvOS — plus Material Design 3 and WCAG 2.2 for cross-platform projects. Each platform file encodes specific conventions: iOS gets 67+ rules covering tab bars, sheets, and Dynamic Island; macOS covers menu bars, window management, and keyboard-driven interaction; visionOS addresses spatial UI, eye tracking, and ergonomic comfort zones. The skill was built by scraping the official HIG and distilling it into actionable rules.
Axiom (by charleswiltgen) takes the most comprehensive approach with 60+ skills for Apple platform development. It includes dedicated hig and hig-ref skills — the former for quick design decision frameworks, the latter as a comprehensive HIG API reference with code examples. Axiom also covers Liquid Glass (iOS 26), SwiftUI layout patterns, and accessibility. It has been featured on Hacker News and is actively maintained with extensive documentation at charleswiltgen.github.io/Axiom.
raintree-technology/apple-hig-skills packages the HIG into 14 focused agent skills covering platforms, foundations, and components (layout, controls, dialogs, menus, search, status, content). It uses progressive disclosure to stay token-efficient while providing deep knowledge on demand. Install is a single command: npx skills add raintree-technology/apple-hig-skills. The same team publishes hig-doctor, a CLI that audits any project for HIG compliance across 12 frameworks, detecting 349 anti-patterns in SwiftUI, UIKit, AppKit, and even React, Vue, and Flutter.
Other notable skills include axiaoge2/Apple-Hig-Designer (95 stars), which generates SwiftUI/UIKit components with HIG-validated design tokens and CSS custom properties; ceorkm/macos-design-skill, exclusively focused on macOS with exact spacing grids, corner radii, animation curves, and blur values; and oil-oil/ios-hig-design-guide-skill, which syncs official Apple HIG JSON source data and generates feature-level UI/UX specs with HIG section-path citations.
Claude skills, officially launched in October 2025 and published as an open standard at agentskills.io in December 2025, are folders containing a SKILL.md file with YAML frontmatter and markdown instructions. They use a three-stage progressive disclosure system: at startup, only the frontmatter metadata (~50–100 tokens) loads; when Claude determines the skill is relevant, the full instructions load (<5k tokens); supporting reference files load only when specifically needed. This means dozens of skills can coexist with minimal context window cost.
Auto-loading is semantic, not environment-based. Claude scans installed skills’ description and when_to_use fields, then automatically activates relevant skills based on conversation content. A paths field supports file-path glob patterns (e.g., src/ios/** or *.swift) to restrict activation to specific directories. Skills also support nested discovery in monorepos — editing packages/ios/ automatically checks packages/ios/.claude/skills/. However, there is no native mechanism for auto-loading based on target operating system. Platform differentiation must be achieved through one of three patterns:
- CLAUDE.md declaration: Set
## Platform: iOS 17+in your project’s CLAUDE.md file, which Claude reads at every session start - Separate skill files per platform: As ehmo/platform-design-skills implements, with distinct iOS, macOS, and visionOS skill files that Claude selectively loads based on conversational context
- MCP server with platform parameters: Tools like apple-dev-mcp accept an explicit
platformargument in their search functions
The skills ecosystem is now cross-platform. The Agent Skills open standard has been adopted by OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, and others — the same SKILL.md files work across all of them. Distribution happens through multiple community marketplaces including claudecodeplugins.io (2,793+ skills), SkillsMP (800,000+ skills), MCP Market, LobeHub, and skills.rest. Plugins, which launched in beta October 2025, bundle skills with MCP configurations, slash commands, and hooks into single installable packages.
Model Context Protocol servers provide the most dynamic approach to Apple HIG expertise, fetching current documentation rather than relying on static skill files.
Sosumi.ai is the strongest current option — a remote MCP server at https://sosumi.ai/mcp that translates Apple’s JavaScript-rendered developer documentation (including HIG) into AI-friendly Markdown in real-time. Its fetchAppleDocumentation tool accepts specific HIG paths like design/human-interface-guidelines/foundations/color, enabling platform-specific queries. Setup requires no local installation: claude mcp add --transport http sosumi https://sosumi.ai/mcp. It also provides a companion SKILL.md file.
tmaasen/apple-dev-mcp was purpose-built for HIG with 113+ pre-processed sections, smart caching, and dedicated platform-aware tools. Its search_human_interface_guidelines function accepts a platform parameter (iOS, macOS, watchOS, tvOS, visionOS), and a compare_platforms tool directly compares how components behave across platforms. It also exposes platform-specific resource URIs (hig://ios/navigation, hig://macos/navigation). This server was archived in August 2025 but remains functional and installable via npm.
apple-docs-mcp-server (by bbssppllvv) takes a RAG approach with OpenAI embeddings over 16,253+ Apple developer documents in a 260MB pre-built database, supporting natural-language semantic search across the entire documentation corpus including HIG.
For cutting-edge content, appledeepdoc-mcp exposes Apple’s hidden Xcode documentation including advanced SwiftUI patterns and Liquid Glass design guides for iOS 26+ that aren’t available on Apple’s public developer site.
If existing options don’t fit your needs, several resources make building a custom HIG skill straightforward.
justinwetch/HIGAgentSkills is the single best foundation — 150 distilled Markdown files totaling ~130k tokens, purpose-built for agent context windows. Each file has YAML frontmatter with topic, tier, platforms, category, triggers, and related fields. The corpus uses a tiered loading system: 14 foundational files always load (~21.4k tokens), 7 platform overview files load when a specific platform is mentioned, 100 component/pattern files load on demand, and 29 niche files cover edge cases. The distillation achieved 75% word reduction while preserving all specific rules, measurements, API names, and platform distinctions.
Apple’s official HIG at developer.apple.com/design/human-interface-guidelines is organized into six categories — Platforms, Foundations, Patterns, Components, Inputs, and Technologies — with platform-specific callouts inline. Apple does not provide any API, JSON feed, or machine-readable export of HIG content. The documentation is HTML-only. However, swiftlysingh/HIGs-PDF provides a Python scraper that compiles the entire HIG into a comprehensive PDF with table of contents and bookmarks, suitable for uploading to Claude Projects.
Community supplements worth incorporating include sindresorhus/human-interface-guidelines-extras (236 stars), which covers details Apple doesn’t specify — macOS settings window conventions, menu item best practices, drag-and-drop patterns with SwiftUI examples — and the NeonWatty iOS HIG workflow testing skills, which encode anti-pattern detection tables that catch cross-platform pollution like hamburger menus, floating action buttons, and web-style dropdowns appearing in iOS apps.
The optimal architecture combines three layers, each addressing a different need:
Layer 1 — Persistent context via CLAUDE.md and platform skills. Set your target platform in the project’s CLAUDE.md file and install ehmo/platform-design-skills for rule-based guidance. This gives Claude a permanent understanding of which platform’s conventions to follow, with 450+ encoded rules available on demand. For comprehensive Apple development coverage beyond just HIG, add Axiom’s full skill suite.
Layer 2 — Live documentation via MCP. Add Sosumi.ai as a remote MCP server (claude mcp add --transport http sosumi https://sosumi.ai/mcp) for real-time access to current Apple documentation. This ensures guidance stays current even when HIG content changes, and it covers API documentation alongside design guidelines.
Layer 3 — Compliance auditing via specialized tools. Install raintree-technology/hig-doctor for automated HIG compliance scoring across your codebase. For iOS projects specifically, add the NeonWatty workflow skills that systematically interrogate your app against platform conventions.
For teams using Claude.ai rather than Claude Code, the most effective approach is creating separate Claude Projects — one for iOS design work and one for macOS — each with platform-specific custom instructions and uploaded HIG documentation generated via the HIGs-PDF scraper or the justinwetch/HIGAgentSkills distilled corpus. Claude Projects use enhanced RAG (Contextual Retriever) for uploaded documents, making this viable even with the full HIG corpus.
The Claude HIG ecosystem has matured rapidly since skills launched in late 2025. The gap between “no tools exist” and “comprehensive multi-platform HIG coverage” has been fully bridged by community efforts. The most significant insight is that platform-aware design guidance doesn’t require a single monolithic tool — it emerges from layering persistent context (CLAUDE.md), encoded rules (platform-specific skills), and live documentation access (MCP servers). The ehmo/platform-design-skills package with its per-platform rule files, combined with Sosumi.ai for live documentation, provides the most complete solution available today. For those wanting a single-install experience, Axiom’s skill suite offers the broadest coverage with the least configuration overhead. Apple’s own integration of Claude into Xcode 26.3 — with system prompts instructing Claude to search for Liquid Glass and use Apple developer documentation via MCP — confirms this layered architecture as the direction the ecosystem is heading.