Sheshiyer Skill Clusters Native IOS Core

Native iOS Core

Shared model for the native-ios cluster. The concurrency, SwiftUI, persistence, AI, and design spokes all depend on the same baseline and the same safety rules — keep them consistent here so no spoke contradicts another.

1. The baseline (this cluster's defining decision)

Everything is written for one toolchain line:

Xcode 26 ──compiles──> Swift 6.2 ──targets──> iOS 26 / macOS 26 (with graceful fallback below)

The flagship capabilities the cluster exists to teach — Approachable Concurrency, Liquid Glass, and on-device FoundationModels — only exist on this line. So the first question for any task is "what's the deployment target, and is this API available there?" — not "how do I call it?". Get the baseline right and the spokes compose; get it wrong and they conflict.

2. Availability gating (non-negotiable)

A higher SDK does not mean every device can run the API. Gate, don't assume:

• Compile-time / OS gate — wrap iOS 26-only UI (Liquid Glass, newest SwiftUI) in if #available(iOS 26, *) with a real pre-26 branch. → liquid-glass-design
• Runtime capability gate — on-device AI is gated by hardware + user settings, not just OS. Always switch on SystemLanguageModel.default.availability (.available / .unavailable(...)) before opening a session. → foundation-models-on-device

Rule: never raise the whole app's minimum deployment target just to use one new affordance — gate the affordance and keep a fallback.

3. Data-race safety model (Swift 6.2)

Concurrency is isolated-by-default; you opt into parallelism, not out of it.

• Async code stays on the calling actor by default (no implicit background hop). Offload explicitly with @concurrent. → swift-concurrency-6-2
• MainActor is the default isolation for UI types; non-isolated protocols are satisfied via isolated conformances.
• Cross-isolation values must be Sendable; the compiler enforces it.
• Actors serialize access to shared mutable state — use them for persistence/caches instead of locks or DispatchQueue. → swift-actor-persistence
• Abstract I/O (file system, network, iCloud) behind small Sendable protocols so it's mockable and the isolation stays clean. → swift-protocol-di-testing

4. Shared SwiftUI & architecture conventions

• State: @Observable (Observation framework) over ObservableObject/@Published — property-level change tracking, fewer re-renders. → swiftui-patterns
• Pick the narrowest property wrapper that fits (@State → @Binding → @Bindable → @Environment).
• Inject dependencies via protocols (not concretes) so views are previewable and testable. → swift-protocol-di-testing
• Structured AI output uses @Generable types, not hand-parsed strings. → foundation-models-on-device
• Icons ship as asset-catalog imagesets (1x/2x/3x), generated to match existing project style. → ios-icon-gen

5. Version / tooling matrix

6. Shared guardrails

• Baseline first: confirm deployment target before reaching for an iOS 26 API.
• Gate every new API: if #available for OS features, .availability for on-device AI — always with a fallback.
• Don't widen the deployment target silently to use one affordance.
• Isolated-by-default: keep data-race safety; offload with @concurrent, share via Sendable, serialize state with actors — never ad-hoc background dispatch.
• @Observable over ObservableObject; narrowest property wrapper that works.
• Protocol-seam I/O: external dependencies behind Sendable protocols so tests stay deterministic.
• Native Swift vs cross-platform: pick this cluster for the newest first-party frameworks (Liquid Glass, on-device AI) and full platform fidelity; if the app must also ship Android/web from one codebase, weigh a cross-platform stack instead.

Source: Sheshiyer/skill-clusters — distributed by TomeVault.