buildanything 1.7.1 → 2.0.0

package/skills/ios/apple-on-device-ai/references/mlx-swift.md

@@ -0,0 +1,285 @@
+# MLX Swift & llama.cpp Reference
+
+Complete reference for running open-source LLMs on Apple platforms using
+MLX Swift and llama.cpp.
+
+## Contents
+
+- [MLX Swift](#mlx-swift)
+- [llama.cpp](#llamacpp)
+- [Multi-Backend Architecture](#multi-backend-architecture)
+- [Built-in Apple Frameworks](#built-in-apple-frameworks)
+- [Performance Best Practices](#performance-best-practices)
+- [Review Checklist](#review-checklist)
+
+## MLX Swift
+
+Apple's ML framework for Swift. Highest sustained generation throughput on
+Apple Silicon via unified memory architecture.
+
+### Key Characteristics
+
+- Unified memory: operations run on CPU or GPU without data transfer
+- Lazy computation: operations computed only when needed
+- Automatic differentiation for training
+- Metal GPU acceleration
+- Research-oriented but increasingly used in production
+
+### Loading and Running LLMs
+
+```swift
+import MLX
+import MLXLLM
+
+let config = ModelConfiguration(
+    id: "mlx-community/Mistral-7B-Instruct-v0.3-4bit"
+)
+let model = try await LLMModelFactory.shared.loadContainer(
+    configuration: config
+)
+
+try await model.perform { context in
+    let input = try await context.processor.prepare(
+        input: UserInput(prompt: "Hello")
+    )
+    let stream = try generate(
+        input: input,
+        parameters: GenerateParameters(temperature: 0.0),
+        context: context
+    )
+    for await part in stream {
+        print(part.chunk ?? "", terminator: "")
+    }
+}
+```
+
+### Recommended Models by Device
+
+| Device | RAM | Recommended Model | Disk Size | RAM Usage |
+|---|---|---|---|---|
+| iPhone 12-14 | 4-6 GB | SmolLM2-135M or Qwen 2.5 0.5B | ~278 MB | ~0.3 GB |
+| iPhone 15 Pro+ | 8 GB | Gemma 3n E4B 4-bit | ~2.7 GB | ~3.5 GB |
+| Mac 8 GB | 8 GB | Llama 3.2 3B 4-bit | ~1.8 GB | ~3 GB |
+| Mac 16 GB+ | 16 GB+ | Mistral 7B 4-bit | ~4 GB | ~6 GB |
+
+### Memory Management Rules
+
+1. Never exceed 60% of total RAM on iOS
+2. Set GPU cache limits:
+   ```swift
+   MLX.GPU.set(cacheLimit: 512 * 1024 * 1024) // 512 MB
+   ```
+3. Monitor memory pressure and reduce cache under pressure
+4. Unload models on app backgrounding
+5. Use "Increased Memory Limit" entitlement for larger models on iOS
+6. Pre-flight memory checks before loading models
+7. Physical device required (no simulator support for Metal GPU)
+
+### Model Lifecycle Management
+
+```swift
+@Observable
+class ModelManager {
+    private var model: LLMModelContainer?
+    private var generationCount = 0
+
+    func loadModel() async throws {
+        let config = ModelConfiguration(
+            id: "mlx-community/Llama-3.2-3B-Instruct-4bit"
+        )
+        model = try await LLMModelFactory.shared.loadContainer(
+            configuration: config
+        )
+    }
+
+    func unloadModel() {
+        model = nil
+        MLX.GPU.set(cacheLimit: 0)
+    }
+}
+```
+
+Key lifecycle patterns:
+- Track active generation count to distinguish "loaded but idle" from
+  "generating"
+- Unconditional cancellation on app backgrounding
+- 5-second delayed force-unload after backgrounding
+- Platform-specific memory monitoring (UIKit on iOS, DispatchSource on macOS)
+
+### Background Handling
+
+```swift
+// iOS: Observe app lifecycle
+NotificationCenter.default.addObserver(
+    forName: UIApplication.didEnterBackgroundNotification,
+    object: nil, queue: .main
+) { _ in
+    modelManager.cancelGeneration()
+    Task {
+        try await Task.sleep(for: .seconds(5))
+        modelManager.unloadModel()
+    }
+}
+```
+
+## llama.cpp
+
+C/C++ LLM inference engine. Best cross-platform support. Uses GGUF model format.
+
+### Swift Integration (swift-llama-cpp)
+
+```swift
+import SwiftLlamaCpp
+
+let service = LlamaService(
+    modelUrl: modelURL,
+    config: .init(
+        batchSize: 256,
+        maxTokenCount: 4096,
+        useGPU: true
+    )
+)
+
+let messages = [
+    LlamaChatMessage(role: .system, content: "You are helpful."),
+    LlamaChatMessage(role: .user, content: "Hello")
+]
+
+let stream = try await service.streamCompletion(
+    of: messages,
+    samplingConfig: .init(temperature: 0.8)
+)
+for try await token in stream {
+    print(token, terminator: "")
+}
+```
+
+### GGUF Quantization Levels
+
+| Level | Quality | Size | Use Case |
+|---|---|---|---|
+| Q2_K | Lowest | Smallest | Extreme memory constraints |
+| Q4_K_M | Good | Balanced | Mobile devices (recommended) |
+| Q5_K_M | Higher | Larger | When quality matters more |
+| Q8_0 | Near-original | Largest | Desktop with ample RAM |
+
+### llama.cpp vs MLX Swift
+
+| Aspect | llama.cpp | MLX Swift |
+|---|---|---|
+| Model format | GGUF | Hugging Face / MLX format |
+| Platform support | Cross-platform | Apple only |
+| Throughput (Apple Silicon) | Good | Best |
+| Model ecosystem | Broadest | mlx-community models |
+| Maturity | Very mature | Evolving |
+| Memory efficiency | Excellent | Good |
+
+## Multi-Backend Architecture
+
+When an app needs multiple AI backends:
+
+### Fallback Chain Pattern
+
+```swift
+func respond(to prompt: String) async throws -> String {
+    // Try Foundation Models first (zero setup, best integration)
+    if SystemLanguageModel.default.isAvailable {
+        return try await foundationModelsRespond(prompt)
+    }
+
+    // Fall back to MLX Swift (best throughput)
+    if canLoadMLXModel() {
+        return try await mlxRespond(prompt)
+    }
+
+    // Fall back to llama.cpp (broadest compatibility)
+    if llamaModelAvailable() {
+        return try await llamaRespond(prompt)
+    }
+
+    throw AIError.noBackendAvailable
+}
+```
+
+### Architecture Guidelines
+
+1. Create a router that checks Foundation Models availability first
+2. Fall back to MLX or llama.cpp when Foundation Models is unavailable
+3. Define model tiers based on device capabilities
+4. Serialize all model access through a coordinator actor to prevent contention
+5. Ensure tool systems work across backends (schema translation may be needed)
+
+### Coordinator Actor
+
+```swift
+actor ModelCoordinator {
+    private var activeBackend: Backend?
+
+    func withExclusiveAccess<T>(
+        _ work: () async throws -> T
+    ) async rethrows -> T {
+        try await work()
+    }
+
+    enum Backend {
+        case foundationModels
+        case mlx
+        case llamaCpp
+    }
+}
+```
+
+## Built-in Apple Frameworks
+
+Before reaching for custom models, consider built-in frameworks:
+
+### Natural Language Framework
+
+No model downloads required:
+
+- `NLLanguageRecognizer` -- Language detection
+- `NLTokenizer` -- Word, sentence, paragraph tokenization
+- `NLTagger` -- Parts of speech, named entity recognition, sentiment
+- `NLEmbedding` -- Word and sentence vectors, similarity search
+
+### Vision Framework
+
+Built-in computer vision (legacy `VN*` API; for iOS 18+ prefer modern Swift equivalents like `RecognizeTextRequest`):
+
+- `VNRecognizeTextRequest` -- OCR
+- `VNClassifyImageRequest` -- Image classification
+- `VNDetectFaceRectanglesRequest` -- Face detection
+- `VNDetectHumanBodyPoseRequest` -- Body pose estimation
+
+### Create ML
+
+Training custom classifiers directly on device or Mac:
+
+- Image classification
+- Text classification
+- Tabular data models
+- Sound classification
+
+## Performance Best Practices
+
+1. Run outside debugger for accurate benchmarks (Xcode: Cmd-Opt-R, uncheck
+   "Debug Executable")
+2. Use `session.prewarm()` for Foundation Models before user interaction
+3. Batch Vision framework requests in a single `perform()` call
+4. Use `.fast` recognition level for real-time camera processing
+5. Neural Engine (Core ML) is most energy-efficient for compatible operations
+6. For MLX Swift, monitor token generation speed and adjust model size if
+   below acceptable thresholds
+
+## Review Checklist
+
+- [ ] Model size appropriate for target device RAM
+- [ ] Memory pressure monitoring implemented
+- [ ] Models unloaded on app backgrounding
+- [ ] GPU cache limits set appropriately
+- [ ] Pre-flight memory check before loading large models
+- [ ] Fallback strategy when model unavailable
+- [ ] All model access serialized through coordinator
+- [ ] Quantization level appropriate for quality/size tradeoff
+- [ ] Energy efficiency considered (Neural Engine vs GPU)
+- [ ] Physical device testing (not simulator) for Metal-dependent code

package/skills/ios/asc-privacy-manifest/SKILL.md

@@ -0,0 +1,350 @@
+---
+name: asc-privacy-manifest
+description: Generate and validate PrivacyInfo.xcprivacy files for iOS apps — required reason APIs, collected data types, tracking declarations, and SDK aggregation.
+license: MIT
+metadata:
+  author: oneshot
+  version: "1.0.0"
+---
+
+# iOS Privacy Manifest (PrivacyInfo.xcprivacy)
+
+Since iOS 17 / Xcode 15, every app and third-party SDK must ship a `PrivacyInfo.xcprivacy` file declaring what data it collects, which privacy-sensitive APIs it calls, and whether it tracks users. App Store Connect rejects submissions that omit required entries (ITMS-91053, ITMS-91055).
+
+## File structure
+
+`PrivacyInfo.xcprivacy` is a property list (XML plist) with four top-level keys:
+
+| Key | Type | Purpose |
+|-----|------|---------|
+| `NSPrivacyTracking` | Boolean | `true` if the app tracks users per ATT definition |
+| `NSPrivacyTrackingDomains` | Array of String | Domains contacted for tracking (only when tracking is true) |
+| `NSPrivacyCollectedDataTypes` | Array of Dict | Each data category collected, with purpose and linkage |
+| `NSPrivacyAccessedAPITypes` | Array of Dict | Each required-reason API category used, with reason codes |
+
+## 1. Required reason APIs (NSPrivacyAccessedAPITypes)
+
+Five API categories require declared reasons. Each entry is a dict with `NSPrivacyAccessedAPIType` (category string) and `NSPrivacyAccessedAPITypeReasons` (array of reason codes).
+
+### File timestamp APIs — `NSPrivacyAccessedAPICategoryFileTimestamp`
+
+Triggered by: `FileManager` attribute reads (`modificationDate`, `creationDate`), `stat()`, `fstat()`, `getattrlist()`, `NSFileModificationDate`, `NSURLContentModificationDateKey`, `NSURLCreationDateKey`.
+
+| Code | Meaning |
+|------|---------|
+| `DDA9.1` | Display file timestamps to the user on-device only (no off-device transmission) |
+| `C617.1` | Access file metadata inside app container, app group container, or CloudKit container |
+| `3B52.1` | Access metadata for files the user explicitly selected (document picker, drag-and-drop) |
+| `0A2A.1` | Third-party SDK wrapper — only when the host app invokes the wrapper API |
+
+### System boot time APIs — `NSPrivacyAccessedAPICategorySystemBootTime`
+
+Triggered by: `ProcessInfo.processInfo.systemUptime`, `mach_absolute_time()`, `clock_gettime(CLOCK_MONOTONIC)`.
+
+| Code | Meaning |
+|------|---------|
+| `35F9.1` | Measure elapsed time between events within the app |
+| `8FFB.1` | Calculate absolute timestamps for events (e.g. UIKit or AVFAudio timestamps) |
+| `3D61.1` | Include boot time in an optional, user-initiated bug report |
+
+### Disk space APIs — `NSPrivacyAccessedAPICategoryDiskSpace`
+
+Triggered by: `FileManager` `attributesOfFileSystem(forPath:)`, `volumeAvailableCapacityKey`, `volumeTotalCapacityKey`, `statfs()`, `statvfs()`.
+
+| Code | Meaning |
+|------|---------|
+| `85F4.1` | Display disk space information to the user on-device |
+| `E174.1` | Check available space before writing files or delete files on low disk |
+| `7D9E.1` | Include disk space in an optional, user-initiated bug report |
+| `B728.1` | Health research apps detecting low disk for data integrity |
+
+### Active keyboards APIs — `NSPrivacyAccessedAPICategoryActiveKeyboards`
+
+Triggered by: `UITextInputMode.activeInputModes`.
+
+| Code | Meaning |
+|------|---------|
+| `3EC4.1` | Custom keyboard app checking which keyboards are active |
+| `54BD.1` | Customize UI or input behavior based on active keyboards |
+
+### User defaults APIs — `NSPrivacyAccessedAPICategoryUserDefaults`
+
+Triggered by: `UserDefaults` standard or suite access. Almost every app hits this.
+
+| Code | Meaning |
+|------|---------|
+| `CA92.1` | Read/write data accessible only to the app itself |
+| `1C8F.1` | Read/write data shared via App Group (app + widgets/extensions) |
+| `C56D.1` | Third-party SDK wrapper providing access to user defaults |
+| `AC6B.1` | Read managed app configuration or write feedback for MDM |
+
+## 2. Collected data types (NSPrivacyCollectedDataTypes)
+
+Each dict in the array declares one data category with four keys:
+
+| Key | Type | Description |
+|-----|------|-------------|
+| `NSPrivacyCollectedDataType` | String | Identifier from Apple's predefined list |
+| `NSPrivacyCollectedDataTypeLinked` | Boolean | `true` if this data is linked to the user's identity |
+| `NSPrivacyCollectedDataTypeTracking` | Boolean | `true` if this data is used for tracking |
+| `NSPrivacyCollectedDataTypePurposes` | Array of String | Why this data is collected |
+
+### Common data type identifiers
+
+| Identifier | Category |
+|------------|----------|
+| `NSPrivacyCollectedDataTypeName` | Contact info — name |
+| `NSPrivacyCollectedDataTypeEmailAddress` | Contact info — email |
+| `NSPrivacyCollectedDataTypePhoneNumber` | Contact info — phone |
+| `NSPrivacyCollectedDataTypeUserID` | Identifiers — user ID |
+| `NSPrivacyCollectedDataTypeDeviceID` | Identifiers — device ID (IDFA, IDFV) |
+| `NSPrivacyCollectedDataTypePreciseLocation` | Location — precise |
+| `NSPrivacyCollectedDataTypeCoarseLocation` | Location — coarse |
+| `NSPrivacyCollectedDataTypeProductInteraction` | Usage data — product interaction |
+| `NSPrivacyCollectedDataTypeOtherUsageData` | Usage data — other |
+| `NSPrivacyCollectedDataTypeCrashData` | Diagnostics — crash data |
+| `NSPrivacyCollectedDataTypePerformanceData` | Diagnostics — performance |
+| `NSPrivacyCollectedDataTypeOtherDiagnosticData` | Diagnostics — other |
+| `NSPrivacyCollectedDataTypePaymentInfo` | Financial — payment info |
+| `NSPrivacyCollectedDataTypePhotosorVideos` | User content — photos/videos |
+| `NSPrivacyCollectedDataTypeContacts` | User content — contacts |
+| `NSPrivacyCollectedDataTypeHealth` | Health & fitness |
+| `NSPrivacyCollectedDataTypeFitness` | Health & fitness |
+| `NSPrivacyCollectedDataTypeOtherDataTypes` | Catch-all |
+
+### Purpose identifiers
+
+| Identifier | Meaning |
+|------------|---------|
+| `NSPrivacyCollectedDataTypePurposeAppFunctionality` | Core app features |
+| `NSPrivacyCollectedDataTypePurposeAnalytics` | Analytics |
+| `NSPrivacyCollectedDataTypePurposeProductPersonalization` | Personalization |
+| `NSPrivacyCollectedDataTypePurposeDeveloperAdvertising` | Developer's own ads/marketing |
+| `NSPrivacyCollectedDataTypePurposeThirdPartyAdvertising` | Third-party ad networks |
+| `NSPrivacyCollectedDataTypePurposeOther` | Other purposes |
+
+## 3. Tracking and ATT (NSPrivacyTracking)
+
+Set `NSPrivacyTracking` to `true` only if the app links collected data with third-party data for advertising or ad measurement. This aligns with Apple's App Tracking Transparency (ATT) definition.
+
+When `NSPrivacyTracking` is `true`:
+- The app **must** call `ATTrackingManager.requestTrackingAuthorization()` before tracking.
+- `NSPrivacyTrackingDomains` must list every domain contacted for tracking purposes.
+- iOS blocks network requests to listed domains when the user denies tracking permission.
+- `Info.plist` must include `NSUserTrackingUsageDescription` with a user-facing explanation.
+
+When `NSPrivacyTracking` is `false`, leave `NSPrivacyTrackingDomains` as an empty array.
+
+## 4. Third-party SDK manifests and aggregation
+
+Each SDK must bundle its own `PrivacyInfo.xcprivacy` inside its framework or Swift package bundle. During archival, Xcode aggregates all manifests into a unified **Privacy Report** (accessible via Product > Generate Privacy Report in Xcode).
+
+Rules:
+- The app's manifest covers the app target only — not its dependencies.
+- Each framework/xcframework/Swift package provides its own manifest.
+- Xcode merges all manifests at archive time; duplicates are union-merged.
+- If an SDK lacks a manifest, App Store Connect flags the submission (ITMS-91061).
+- Update third-party SDKs regularly — most major SDKs (Alamofire, Firebase, Sentry, etc.) ship manifests since mid-2024.
+
+For Swift packages: place `PrivacyInfo.xcprivacy` in the package's resource bundle and declare it in `Package.swift` via `.process("PrivacyInfo.xcprivacy")` in the target's `resources` array.
+
+## 5. Xcode integration
+
+### Where to place the file
+- **App target**: project root, alongside `Info.plist`. Xcode copies it into the app bundle at build time.
+- **Framework target**: inside the framework's bundle resources.
+- **Swift Package**: in `Sources/<Target>/` and declared in `Package.swift` resources.
+
+### How to create it in Xcode
+1. File > New > File (or Cmd-N).
+2. Search for "App Privacy" and select the template.
+3. Choose the correct target in the file inspector.
+4. Xcode creates `PrivacyInfo.xcprivacy` with a visual editor for the four top-level keys.
+
+The file **must** be named exactly `PrivacyInfo.xcprivacy` — Apple's tooling scans for this specific filename.
+
+### Generate a privacy report
+1. Archive the app (Product > Archive).
+2. Right-click the archive in the Organizer > Generate Privacy Report.
+3. Review the aggregated report for missing entries before submission.
+
+## 6. Common App Store rejection reasons
+
+| Error code | Cause | Fix |
+|------------|-------|-----|
+| ITMS-91053 | App uses a required-reason API without declaring it in the manifest | Add the missing `NSPrivacyAccessedAPIType` entry with valid reason codes |
+| ITMS-91055 | Invalid reason code for a declared API category | Replace with a valid reason code from Apple's predefined list |
+| ITMS-91056 | Malformed or invalid privacy manifest file | Validate plist XML structure; re-create via Xcode template |
+| ITMS-91061 | A third-party SDK is missing its privacy manifest | Update the SDK or add a manifest to its bundle |
+
+**TestFlight catches these errors** — always run a TestFlight build before production submission.
+
+Other common pitfalls:
+- Declaring `NSPrivacyTracking = true` but omitting `NSUserTrackingUsageDescription` in `Info.plist`.
+- Using reason code `CA92.1` (app-only) when the app actually shares `UserDefaults` via App Group — use `1C8F.1` instead.
+- Forgetting to declare `NSPrivacyAccessedAPICategoryFileTimestamp` when using `SDWebImage`, `Kingfisher`, or other caching libraries that call `stat()`.
+- Declaring collected data types without matching entries in the App Store Connect privacy questionnaire.
+
+## 7. Complete example
+
+A typical app using `UserDefaults`, file timestamps (caching), and analytics with crash reporting:
+
+```xml
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
+  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+
+  <!-- Tracking -->
+  <key>NSPrivacyTracking</key>
+  <false/>
+  <key>NSPrivacyTrackingDomains</key>
+  <array/>
+
+  <!-- Collected data types -->
+  <key>NSPrivacyCollectedDataTypes</key>
+  <array>
+    <!-- Crash data for stability monitoring -->
+    <dict>
+      <key>NSPrivacyCollectedDataType</key>
+      <string>NSPrivacyCollectedDataTypeCrashData</string>
+      <key>NSPrivacyCollectedDataTypeLinked</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypeTracking</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypePurposes</key>
+      <array>
+        <string>NSPrivacyCollectedDataTypePurposeAppFunctionality</string>
+      </array>
+    </dict>
+    <!-- Performance metrics -->
+    <dict>
+      <key>NSPrivacyCollectedDataType</key>
+      <string>NSPrivacyCollectedDataTypePerformanceData</string>
+      <key>NSPrivacyCollectedDataTypeLinked</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypeTracking</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypePurposes</key>
+      <array>
+        <string>NSPrivacyCollectedDataTypePurposeAnalytics</string>
+      </array>
+    </dict>
+    <!-- User ID for account features -->
+    <dict>
+      <key>NSPrivacyCollectedDataType</key>
+      <string>NSPrivacyCollectedDataTypeUserID</string>
+      <key>NSPrivacyCollectedDataTypeLinked</key>
+      <true/>
+      <key>NSPrivacyCollectedDataTypeTracking</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypePurposes</key>
+      <array>
+        <string>NSPrivacyCollectedDataTypePurposeAppFunctionality</string>
+      </array>
+    </dict>
+  </array>
+
+  <!-- Required reason APIs -->
+  <key>NSPrivacyAccessedAPITypes</key>
+  <array>
+    <!-- UserDefaults: app-only storage -->
+    <dict>
+      <key>NSPrivacyAccessedAPIType</key>
+      <string>NSPrivacyAccessedAPICategoryUserDefaults</string>
+      <key>NSPrivacyAccessedAPITypeReasons</key>
+      <array>
+        <string>CA92.1</string>
+      </array>
+    </dict>
+    <!-- File timestamps: cache management -->
+    <dict>
+      <key>NSPrivacyAccessedAPIType</key>
+      <string>NSPrivacyAccessedAPICategoryFileTimestamp</string>
+      <key>NSPrivacyAccessedAPITypeReasons</key>
+      <array>
+        <string>C617.1</string>
+      </array>
+    </dict>
+    <!-- System boot time: elapsed time measurement -->
+    <dict>
+      <key>NSPrivacyAccessedAPIType</key>
+      <string>NSPrivacyAccessedAPICategorySystemBootTime</string>
+      <key>NSPrivacyAccessedAPITypeReasons</key>
+      <array>
+        <string>35F9.1</string>
+      </array>
+    </dict>
+  </array>
+
+</dict>
+</plist>
+```
+
+### Variant: app with App Group sharing and tracking
+
+If the app shares data with a widget via App Group and uses an ad SDK:
+
+```xml
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
+  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+
+  <key>NSPrivacyTracking</key>
+  <true/>
+  <key>NSPrivacyTrackingDomains</key>
+  <array>
+    <string>analytics.example.com</string>
+    <string>ads.example.com</string>
+  </array>
+
+  <key>NSPrivacyCollectedDataTypes</key>
+  <array>
+    <dict>
+      <key>NSPrivacyCollectedDataType</key>
+      <string>NSPrivacyCollectedDataTypeDeviceID</string>
+      <key>NSPrivacyCollectedDataTypeLinked</key>
+      <false/>
+      <key>NSPrivacyCollectedDataTypeTracking</key>
+      <true/>
+      <key>NSPrivacyCollectedDataTypePurposes</key>
+      <array>
+        <string>NSPrivacyCollectedDataTypePurposeThirdPartyAdvertising</string>
+      </array>
+    </dict>
+  </array>
+
+  <key>NSPrivacyAccessedAPITypes</key>
+  <array>
+    <dict>
+      <key>NSPrivacyAccessedAPIType</key>
+      <string>NSPrivacyAccessedAPICategoryUserDefaults</string>
+      <key>NSPrivacyAccessedAPITypeReasons</key>
+      <array>
+        <string>CA92.1</string>
+        <string>1C8F.1</string>
+      </array>
+    </dict>
+  </array>
+
+</dict>
+</plist>
+```
+
+Note: when `NSPrivacyTracking` is `true`, the app must also declare `NSUserTrackingUsageDescription` in `Info.plist` and call `ATTrackingManager.requestTrackingAuthorization()` before any tracking occurs.
+
+## Decision checklist
+
+Before generating the manifest, answer these:
+
+1. Does the app use `UserDefaults`? (almost always yes) --> `CA92.1` or `1C8F.1`
+2. Does the app or any dependency read file timestamps? (caching libraries do) --> `C617.1`
+3. Does the app measure elapsed time or call `mach_absolute_time()`? --> `35F9.1`
+4. Does the app check disk space before writes? --> `E174.1`
+5. Does the app inspect active keyboards? (rare) --> `54BD.1`
+6. Does the app collect crash/performance data? --> add collected data type entries
+7. Does the app link data to user identity? --> set `Linked` to `true` on those types
+8. Does the app track users for advertising? --> `NSPrivacyTracking = true` + ATT + tracking domains