SwiftUI Expert Skill

What it is

SwiftUI Expert Skill is one skill for Claude and other coding agents, written by Antoine van der Lee and Omar Elsayed. It gives the agent a short workflow and a router to 33 reference files on SwiftUI, from state and view structure to Liquid Glass and macOS windows. The agent opens only the files a task needs. The skill pushes no architecture or code style, and it focuses on correctness and performance. It also ships two Python scripts that record and read Instruments traces.

1. You install the skill once
2. You ask your agent to write, review or refactor SwiftUI
3. The agent follows the matching workflow in SKILL.md
4. The agent opens only the reference files that topic needs
5. The agent reports findings or makes the change you asked for

Who it suits

iOS and macOS developers who write SwiftUI with an AI agent and want current APIs and sound data flow. It also suits people who review or clean up an existing SwiftUI codebase, and anyone chasing hangs or dropped frames.

Where to use it

Any coding agent that reads Agent Skills, in a SwiftUI project for iOS or macOS.

• Claude Code: the plugin, or the skills CLI
• Codex and ChatGPT: the OpenAI Plugins Directory
• Cursor: a local plugin folder, or the Cursor Marketplace once the plugin is listed there
• pi and Gemini CLI: one command each
• Autohand Code: a clone, then a copy of the skill folder

How to set it up

Get it saves a zip of Mindket’s setup notes in your downloads folder, with every route the maker documents. The skill itself installs from the maker’s repository.

1. Choose a route. Claude Code takes the two plugin lines below. Most other agents use the skills CLI, which needs Node.
2. Run the lines for your route. The two Claude Code lines run inside a Claude Code session. The CLI line runs in a terminal.
3. Ask for the skill by name. Type the sentence from the Example below in your agent, in a SwiftUI project.

/plugin marketplace add AvdLee/SwiftUI-Agent-Skill
/plugin install swiftui-expert@swiftui-expert-skill

Claude Code’s docs say /plugin install in a session opens the plugin’s details first, so you review it and pick a scope. An installed plugin loads on the next start, or after /reload-plugins in a session that is already open. Plugin skills carry the plugin’s name, so a slash command would read /swiftui-expert:swiftui-expert-skill, and the README asks in plain words instead.

The README recommends the skills CLI, which works for most other agents. Run its line in a terminal.

npx skills@latest add https://github.com/avdlee/swiftui-agent-skill --skill swiftui-expert-skill

For Codex and ChatGPT, open Plugins in the app and search for SwiftUI Expert. The README gives no command for that route. The kit also holds the lines for Gemini CLI, pi and Autohand Code.

How to use it

Name the skill and say what you want. SKILL.md holds five workflows, and the agent follows the one that fits.

Example

Your feed screen drops frames when you scroll, and you want a second opinion on its state and updates.

Use the swiftui expert skill and review the current SwiftUI code for state-management and performance improvements

Your agent reads the code, checks it against the state, list and performance references, replaces hard-deprecated APIs and flags soft-deprecated ones.

Per SKILL.md it offers each speed-up as a suggestion, not a requirement. For a broad review it splits the codebase into smaller focus areas and presents them one at a time.

For newcomers and experts

If you are new. A skill is a folder of instructions that your agent loads when a request matches it. Start with one SwiftUI file you know well. Ask for a review, read what comes back, then say yes to the changes you want. The skill adopts Liquid Glass only when you ask for it.

For experts. SKILL.md’s operating rules treat each View type as an invalidation boundary, prefer native SwiftUI over UIKit bridging, and gate version-specific APIs with #available and a fallback. During feature work the skill replaces hard-deprecated APIs and only flags soft-deprecated ones, leaving them in place.

The trace scripts read five lanes: Time Profiler, Hangs, Animation Hitches, SwiftUI updates and the SwiftUI cause graph. On each correlation, main_running_coverage_pct under 25 percent means the main thread was blocked, and 75 percent or more means it was busy with CPU work. The two cases need different fixes. The SwiftUI template fills its lane on a real device or the host Mac, and on the iOS Simulator the notes switch to Time Profiler.

A repository can offer the plugin to a whole team through .claude/settings.json. The kit has the README’s settings and the limit Claude Code’s docs put on them.

Good to know

It is advice for your agent plus two trace scripts.

• Maker and license: MIT, by Antoine van der Lee and Omar Elsayed. A community project, not made or endorsed by Apple
• Traces need a Mac: the scripts call xctrace, which ships with Xcode. The review and write workflows are text for your agent
• System-wide recording: a recording of all processes can capture activity from other apps. The skill’s notes tell the agent to explain that and get your explicit approval first, then pass a flag that notes it in the command
• Imports: Mindket read record_trace.py, analyze_trace.py and the parser modules they load, and none imports a network library
• The skills CLI: its docs say it sends anonymous telemetry by default, the skill name, files and a timestamp, and that DISABLE_TELEMETRY=1 turns it off
• Older installs: the README says npx skills@latest update migrates an install made before the plugin format. The skill folder moved to skills/swiftui-expert-skill/ in the repository
• Other skills by the maker: Swift Concurrency, Core Data, Swift Testing and Xcode build optimization are separate repositories, not part of this one