I open-sourced 16 production iOS skills — designed to make AI coding agents actually good at Swift.

AI agents will happily write you a 2,000-line ViewController with Singletons everywhere. Without a structured codebase scaffold and rigid rules, ensuring a high quality codebase can be really challenging with generative AI. Agents can generate a lot of unscalable code resulting in unstructured prompts. In the era of AI, creating a high quality codebase with every aspect from architecture to the robust error handling is more important than ever.

So I extracted every architectural decision from an iOS app I've been building — 10+ Tuist modules, on-device LLM, StoreKit 2, EventKit–CoreData sync, custom design system — and turned them into 16 self-contained skill documents.

**iOS Agent Skills **is a free, open-source collection of battle-tested iOS patterns. Point your AI agent at them and it stops guessing.

What's inside:

Architecture (5 skills)

• **Tuist-Based Modular Architecture **— Project.swift with Core/Feature layers, strict dependency boundaries at compile time, no more Xcode merge conflicts

• **Protocol-Driven DI **— Complete dependency injection in ~80 lines of Swift. No third-party frameworks. Thread-safe, KeyPath-based, testable

• **UI Factory Pattern **— Feature modules expose a single factory. No service locator calls from Views. No AnyView

• **Actor-Based Concurrency **— When to use actor vs @MainActor vs @unchecked Sendable with NSLock. Prevent data races at compile time

• **Typed Error System **— Every error case maps to a title, message, SF Symbol, and recovery action. Includes AI safety classification with localized refusal responses

• Makefile for iOS Workflows— One make command to rule them all. make run launches Pro Debug, make free launches Free tier, make snapshots runs all visual regression tests, make fresh resets data and relaunches. Self-documenting help, simulator override, CI-ready with CODE_SIGNING_ALLOWED=NO

UI (3 skills)

• **Design System as Core Module **— Tokens, reusable components, animations, haptics. Dark-mode-only, 4pt grid, OLED-optimized

• **Custom Tab Bar & Navigation **— Replace TabView entirely. Re-tap scroll-to-top, spring physics, keyboard avoidance, splash → onboarding → main app flow

• **Localization & Multi-Language **— 43 languages in a modular Tuist app. Bilingual AI intent classification. Localized safety responses

Testing & Debugging (2 skills)

• **Swift Testing & TDD **— Modern @Suite/@Test/[#expect](https://x.com/search?q=%23expect&src=hashtag_click) over XCTest. Five test categories. Regression tests from real bug screenshots

• **Debug Modes & Mock Services **— Three-tier mock system. Secret gesture activation with SHA256-hashed codes for QA testers without Xcode

Platform Frameworks (5 skills)

• **StoreKit 2 Subscriptions **— Intelligence-based trial that ends when the AI proves its value (not a fixed timer). Trial manager state machine, daily query tracking

• **Privacy-First Analytics **— Typed event factory (no free-form strings). Calendar titles, locations, and names are NEVER sent. Pluggable multi-backend with OSLog fallback

• **Notifications with Deep Linking **— LLM-generated local notifications. Deep links into Zoom, Meet, Teams, Webex (11 platforms). Engagement-gated permission requests

• **EventKit–CoreData Sync **— Three-layer sync pipeline. Programmatic CoreData models (no .xcdatamodeld). Upsert by ID. 90-day lookback + 7-day forward

• **WidgetKit & App Intents **— Home Screen widgets as a separate Tuist target with zero framework dependencies. Siri Shortcuts via AppIntent

AI & Intelligence (1 skill)

• **On-Device LLM with Apple Foundation Models **— 100% on-device inference. No API key. No data leaves the device. Dynamic prompt gene pool with fitness-weighted selection, feedback-driven evolution, and safety guardrails

---

How to use:

Point your AI agent at any skill file:

@skills/architecture/01-tuist-modular-architecture.md

Or load an entire category:

@skills/architecture/

Every skill is self-contained. Context → Pattern → Anti-Patterns. The agent knows *when *to apply it, *how *to implement it, and what mistakes to avoid.

Built with Swift 6+, SwiftUI, Tuist, Swift Testing, CoreData, Apple Foundation Models, StoreKit 2, and TelemetryDeck.

MIT licensed. Fork it, add your own skills, make your agents smarter.

🔗 github.com/alpozcan/iOSAgentSkills

PRs and contributions are welcome.