SwiftUI Design Skill
Make SwiftUI interfaces look intentionally crafted, not AI-generated: six iron rules, design direction consulting, brand asset protocol, and a five-dimension review in one skill.
Pure-text design guidance: no code execution, no permissions, no telemetry; data flow largely transparent. Deductions: the Brand Asset Protocol step 3 instructs the agent to download brand asset files (PNG/SVG/fonts) from the web without source verification, malicious-file safeguards, or explicit user confirmation; publisher identity is unverified, so attribution rests on self-declared metadata.
SKILL.md is internally consistent: prerequisites, six rules, workflows and checklist cohere, and it declares dependencies on references/ and templates/. But this static review cannot verify those reference files' actual content or compilability of the Swift snippets; no tests, failure-feedback paths, or abnormal-input handling are documented, so the static ceiling applies.
Clear fit and non-fit boundaries (When to Use / When NOT to Use, routing to swiftui-expert-skill) and precise semantic triggers (e.g., 'UI looks generic'). Deductions: body text is English-only with a Chinese README only; brand protocol depends on reaching overseas brand sites, though the core design function runs offline, limiting impact.
MIT License, version 1.0.0, release notes, bilingual READMEs, and disclosed file structure with line counts; progressive disclosure (SKILL.md → references → templates) is well designed. Deductions: no explicit maintenance ownership or update path, no FAQ or known-limitations section, publisher identity unverified.
Capability matrix, typical timings, 5-Dimension Review and deliverables are concrete; marginal value is clear (an actionable anti-AI-slop rule set is rare in prompt-only solutions). But static review cannot verify outputs (compilable SwiftUI code, review radar) are directly usable; marketing-style time promises ('3–30 min') lack supporting evidence.
Only two Apple documentation links cited; the referenced references/ files are not present in this evidence set, so consistency with SKILL.md claims cannot be cross-checked. No tests, no third-party execution evidence, no real output samples — key claims remain author self-assertions.
- The Brand Asset Protocol directs the agent to download brand asset files from the web; users should explicitly confirm this step and verify file provenance themselves.
- The referenced references/ and templates/ files were outside this review scope; their quality and consistency with SKILL.md claims are unverified.
- Code snippets (e.g., the Color(hex:) extension) depend on the swift-extensions reference; compilability is unverified.
- Publisher is unverified; no explicit commitment to long-term maintenance or updates.
- Body text is English-only; Chinese users depend on the agent's cross-language comprehension.
What does this skill do, and when should you use it?
A visual design skill for SwiftUI front-ends, aimed at producing iOS/macOS interfaces that don't look AI-generated. It bundles six anti-AI-slop iron rules, a direction-consulting workflow across five design schools, a five-step brand asset protocol, and a five-dimension design review. When requirements are vague it recommends 2-3 differentiated directions and produces design tokens, typography hierarchies, and layout code. The repo states it works across AI coding agents including Claude Code, Cursor, Codex, and OpenCode.
Reads the user's product context and brand materials; collects and verifies brand colors, fonts, and assets via a five-step protocol, generating a brand-spec.md; recommends 3 directions from five design schools (Informational, Editorial, Expressive, Functional, Warm Minimal) with visual anchors; defines design tokens (colors, fonts, 8pt spacing grid, corner radii); produces compilable SwiftUI layout code (drawing on 9 patterns in references/layout-patterns.md); and before delivery scores the design on five dimensions (philosophy consistency, visual hierarchy, detail execution, functionality, originality, each ≥7/10) with a fix checklist.
- An indie iOS developer designing a new screen from scratch who wants to avoid AI-slop looks like purple-blue gradients and emoji icons
- A developer told 'the UI looks generic' who needs an item-by-item audit and fixes for banned patterns
- A developer building an app for a specific brand who needs brand colors and typography turned into brand-spec.md and design tokens
- A team with vague requirements that needs 3 design directions with rationale before choosing a visual style
- Prototyping scenarios that need fast, compilable SwiftUI mockups including animations and 9 common layouts
- Pre-ship UI quality review yielding radar-chart scores and a Keep/Fix/Quick Wins checklist
What are this skill's strengths and limitations?
- Rules are concrete and actionable: banned AI-slop visual patterns are listed explicitly with approved alternatives
- The five-dimension review provides a quantified gate (each ≥7/10) for pre-ship quality control
- Agent-agnostic; claims support for Claude Code, Cursor, Codex, OpenCode and more
- MIT-licensed, plain Markdown structure (SKILL.md + references + templates) — easy to audit and customize
- Clearly divided from swiftui-expert-skill, covering 'what to build' vs 'how to build it right'
- No automated test suite; design quality depends on how well the model actually follows the rules
- The brand asset protocol's search-and-download steps require network access through the agent; the skill itself does not enforce it
- Covers only visual design: no compile fixes, no performance work, no standalone accessibility compliance
- Five-dimension and radar scores are subjective heuristics, not objective measurements
- Requires Xcode 16+ and iOS 18+/macOS 15+ SDK; some patterns may not apply to older projects
How do you install this skill?
The README gives: npx skills add wholiver/swiftui-design-skill -g -y. Alternatively, clone https://github.com/Wholiver/swiftui-design-skill and place the folder containing SKILL.md into your agent's skills directory (the exact directory name is not documented in the source; check your platform's docs). Prerequisite: Xcode 16+ with iOS 18+/macOS 15+ SDK.
How do you use this skill?
Trigger it when designing new SwiftUI views, reviewing UI quality, choosing a visual style, or when the interface 'looks AI-generated'. Example prompt: 'Design a warm-minimal SwiftUI screen for my iOS budgeting app using my brand color #E85D3A.' The skill asks about context and brand assets first, then proposes directions, tokens, and compilable code, finishing with the five-dimension review. Note: it does not fix SwiftUI compilation errors or do performance optimization — that is swiftui-expert-skill's scope.
How does this skill compare with similar options?
The repo explicitly contrasts this skill with the same author's swiftui-expert-skill: this one answers 'how does it look?' (visual design, direction, brand feel, review), while swiftui-expert-skill answers 'is the code correct?' (code review, Instruments analysis, API modernization). They are complementary: use Design to set direction first, then Expert to ensure code quality.