// the find
twostraws/SwiftUI-Agent-Skill
SwiftUI agent skill for Claude Code, Codex, and other AI tools.
A set of SwiftUI review rules packaged as an agent skill for developers who already work in Claude Code, Codex, or similar tools on Apple-platform apps. It is a prompt bundle rather than a library, so its value depends on whether the rules catch what models actually get wrong.
The rules are narrow and concrete, covering deprecated API, controls hidden from VoiceOver, and performance traps, and they are split into topic files (accessibility, navigation, performance, views, data and so on) so an agent can load only what a given review needs. The packaging uses the open Agent Skills format with an npx installer and a Claude Code plugin marketplace entry, so one set of content works across several tools without forking. The MIT license and the contributing rule to skip what models already know are the right constraints for a skill that has to stay small.
The listing shows no evals, test fixtures, or before and after examples, so the claim that the rules target mistakes LLMs actually make rests on the author's experience rather than on measured output. The README says the skill is designed for iOS 26 and Swift 6.2, which means advice may be wrong for projects with older deployment targets, and the README does not say how to handle that. The contributing notes concede that the skill costs tokens, and the tree shows SKILL.md in two places (swiftui-pro/ and swiftui-pro/skills/swiftui-pro/), which leaves it unclear which copy is canonical.