// tool · 05 / 05
iOS-toolkit
Current iOS knowledge for Claude Code, on demand. 200+ Apple-platform references plus Swift guardrails, so the agent stops writing last year's APIs.
// what is it
A single iOS skill that routes Claude Code to current Apple-platform references.
iOS-toolkit is a Claude Code plugin built around one skill the agent consults before any iOS or Swift work. It routes to 200+ reference files on demand — SwiftUI, Swift Concurrency, SwiftData, networking, security, media, App Store shipping — each scoped to current Apple platforms: iOS 26, Swift 6.2, Liquid Glass, Foundation Models.
Beyond references, four hooks keep the work honest: iOS context loads at session start, Swift writes are formatted and checked on save, common build and concurrency errors are flagged in the terminal, and a validation pass runs before the agent reports done. Curated from battle-tested open-source iOS skills, wired into one plugin. Open source, MIT-licensed.
// the problem
Claude Code's iOS knowledge is frozen at its training cutoff.
Apple ships a major platform every year, and Claude Code's training stopped early in 2025. So the agent reaches for what it remembers: UIKit where SwiftUI fits, StoreKit 1 patterns, deprecated networking, concurrency that races under Swift 6. It compiles, then breaks on device — or clears the simulator and gets rejected at review.
iOS-toolkit fixes the input. The agent fetches current references — iOS 26, Swift 6.2, SwiftData, Liquid Glass, the App Review guidelines — at the moment it writes, instead of guessing from a year-old memory of Apple's docs. Hooks catch the predictable failures: a Swift write is formatted and run through guardrails, a failed build or a data-race warning is surfaced in the terminal, a validation pass runs before the agent stops.
The agent stops writing last year's iOS. It looks up this year's.
// how it works
One skill. The agent looks up the rest.
- // 01
Install the plugin.
Add the marketplace and install — two lines in Claude Code:
/plugin marketplace add petbrains/IOS-Plugin, then/plugin install ios. No keys, no config; the skill and its references ship with it. - // 02
The agent consults it first.
On any iOS or Swift task — a build error, a SwiftUI view, a schema change — the skill activates before the agent answers, even for work that looks simple. iOS 26 and Swift 6.2 are the baseline, not the agent's year-old default.
- // 03
It loads only the references it needs.
The skill routes the task to specific references — SwiftUI data flow, actor isolation, SwiftData migration, StoreKit 2 — and reads just those. Tasks that span domains pull from every domain that applies, not the first match.
- // 04
Hooks guard every write.
Swift files are formatted and run through guardrails on save; failed builds, data races, and risky migrations are flagged in the terminal; a validation pass runs before the agent reports done.
- // 05
Audits and workflows for the hard parts.
Twenty-two audit methodologies and operational workflows — build-fix, crash analysis, simulator testing, a full health check — cover the second-day work, not just the first draft.
// who it's for
iOS developers shipping real apps with Claude Code.
iOS-toolkit is for developers already building Apple-platform apps with Claude Code — SwiftUI front to back, SwiftData or Core Data underneath, headed for the App Store — who keep correcting the agent's dated Swift. If the agent is fast but its iOS is a year behind, this closes the gap.
Best fit: real apps targeting current iOS, where deprecated APIs and missed platform features cost real time. Worst fit: tiny scripts, non-Apple platforms, or codebases pinned to old SDKs — the current-references angle is wasted if you can't adopt them.
// compatibility
Compatibility
// get started
Install the plugin, write current iOS.
Open source, MIT-licensed. Add the petbrains marketplace and install — two lines in Claude Code, no account, no config.
Get notified when iOS-toolkit ships a major update.
// subscribed
Check your inbox for the confirmation link.