Writes and reviews Swift App Intents code that exposes app actions and data to Siri, Shortcuts, Spotlight, widgets, Control Center, and Apple Intelligence. Use when adding AppIntent, AppEntity, OpenIntent, AppShortcutsProvider, EntityQuery, Focus Filters, AssistantEntity/AssistantIntent schemas, on-screen awareness, LongRunningIntent/CancellableIntent, EntityCollection, SyncableEntity, AppIntentsTesting, or when wiring SwiftData/networked data into intents.
Instrucciones de origen · Vista previa de solo lectura
name
app-intents
description
Writes and reviews Swift App Intents code that exposes app actions and data to Siri, Shortcuts, Spotlight, widgets, Control Center, and Apple Intelligence. Use when adding AppIntent, AppEntity, OpenIntent, AppShortcutsProvider, EntityQuery, Focus Filters, AssistantEntity/AssistantIntent schemas, on-screen awareness, LongRunningIntent/CancellableIntent, EntityCollection, SyncableEntity, AppIntentsTesting, or when wiring SwiftData/networked data into intents.
license
MIT
metadata
{"author":"Anton Novoselov","version":"1.2.0"}
Write and review Swift code that exposes app functionality through the App Intents framework, ensuring correct protocol conformance, safe data flow, and idiomatic discoverability wiring.
Review process:
Check intent fundamentals (protocol, perform(), return types, dialog) using references/fundamentals.md.
Validate parameters and parameter options using references/parameters.md.
Validate entity types, display representations, and queries using references/entities.md.
Check the AppShortcutsProvider registration and discoverability UI using references/shortcuts-and-siri.md.
Validate OpenIntent navigation and snippet view return types using references/open-and-snippet-intents.md.
Check dependency injection and the data-controller pattern using references/dependencies.md.
Validate Spotlight indexing via IndexedEntity, IndexedEntityQuery, and attribute sets using references/spotlight.md.
Check @AssistantEntity / @AssistantIntent schema adoption using references/assistant-schemas.md.
Validate the Siri / Apple Intelligence integration (finding content, custom responses, donations, confirmation, ownership) using references/siri-intelligence.md.
Check on-screen awareness and content transfer (view annotations, IntentValueRepresentation, system-integration annotations) using references/onscreen-awareness.md.
For intents that run past 30 seconds, handle cancellation, or target a specific process, use references/long-running-and-execution.md.
If writing or reviewing tests for intents, use references/testing-intents.md (covers the new AppIntentsTesting framework).
Catch common mistakes using references/anti-patterns.md.
If doing partial work, load only the relevant reference files.
Task-based routing
Match the user's goal to the read order. Load only what you need.
references/assistant-schemas.md - schema macros, domains, Use Model action, onscreen content
references/entities.md - @Property, Transferable (required for several schemas)
"Let Siri understand what's on screen"
references/onscreen-awareness.md - the 4 annotation APIs, content transfer, displayRepresentations fast path, annotations on notifications/Now Playing/AlarmKit
references/siri-intelligence.md - how onscreen context fits the larger Siri story
"Make my intent run longer than 30 seconds or target a process"
references/testing-intents.md - the AppIntentsTesting framework (out-of-process integration), direct struct unit tests, mocking @Dependency, test-only intents, what you can't test
"I'm hitting build errors or runtime bugs"
references/anti-patterns.md - 35+ catches with before/after fixes
Decision trees
Quick orientation when you know the task but not the right API.
Which intent protocol should I conform to?
Generic action?
→ AppIntent
Opens the app to a specific entity?
→ OpenIntent (+ TargetContentProvidingIntent on iOS)
Renders an interactive view only, no business logic?
→ SnippetIntent
Needs to bring the app forward conditionally?
→ AppIntent with supportedModes (iOS 26+)
→ or ForegroundContinuableIntent (iOS 17-18)
Runs longer than 30 seconds (upload, sync, inference)?
→ LongRunningIntent + performBackgroundTask (iOS 27+)
Needs to clean up gracefully when cancelled (with the reason)?
→ CancellableIntent + withIntentCancellationHandler (iOS 26.4+)
Deletes entities with standard confirmation?
→ DeleteIntent
Routes a search query into the app?
→ ShowInAppSearchResultsIntent
Backs widget configuration?
→ WidgetConfigurationIntent (empty conformance, no perform)
Backs a Control Center control?
→ ControlConfigurationIntent
Has a universal-link URL representation?
→ URLRepresentableIntent + OpenIntent (no perform needed)
Matches an Apple Intelligence domain?
→ AppIntent + @AppIntent(schema: .domain.action)
Which entity type should I use?
Fixed set known at compile time?
→ AppEnum
Dynamic data with a persistent id?
→ AppEntity
Entity must appear in Spotlight?
→ AppEntity + IndexedEntity
Entity IS a file (scan, voice memo, exported image)?
→ FileEntity
Computed / aggregated data with no stable id?
→ TransientAppEntity
Needs Apple Intelligence schema awareness?
→ AppEntity + @AppEntity(schema: .domain.type)
Supports cross-app sharing?
→ AppEntity + Transferable
Which query should my entity use?
Small fixed set, enumerable?
→ EnumerableEntityQuery (also gets a basic Find intent)
Large dataset, searchable by name?
→ EntityQuery + EntityStringQuery
Many queryable properties, user should build predicates?
→ EntityPropertyQuery (auto-generates Find intent with comparators + sort)
Simple id-only lookup, no search?
→ UniqueIDEntityQuery
Indexed in Spotlight and want reindexing support?
→ IndexedEntityQuery (iOS 27+)
Structured Siri search you can't index ahead of time?
→ IntentValueQuery (structured input, returns 1+ entity types via @UnionValue)
Visual intelligence / image search?
→ IntentValueQuery + SemanticContentDescriptor
How should Siri find my content?
Content is local and indexable?
→ IndexedEntity + indexAppEntities (semantic index) - the primary path
→ add IndexedEntityQuery for reindexing support (iOS 27+)
Too large / server-side / too volatile to index?
→ IntentValueQuery (structured search input)
User wants to search inside your own UI?
→ @AppIntent(schema: .system.searchInApp) + ShowInAppSearchResultsIntent
Content nobody has found or used yet, but relevant right now?
→ RelevantEntities.updateEntities(_:for:) (iOS 27+)
Teach the system how people use the app (patterns)?
→ IntentDonationManager (donate UI interactions)
Which property wrapper should I use on entity fields?
Stored on the entity struct, should appear in Shortcuts/Find/summary?
→ @Property
Derived from an underlying model object (cheap to compute)?
→ @ComputedProperty (preferred over @Property for wrappers around models)
Derived AND should be indexed in Spotlight?
→ @ComputedProperty(indexingKey: \.keyName)
Expensive to compute (network, ML inference, heavy query)?
→ @DeferredProperty (async getter, only runs when system asks)
Internal to the entity, not shown anywhere?
→ plain stored property (no wrapper)
Core Instructions
Target iOS 16+ / macOS 13+ minimum for App Intents. IndexedEntity, OpenIntent, focus filters, and control widgets require iOS 16+; @AssistantEntity / @AssistantIntent schemas require iOS 18.2+; anchored relative date styles and many assistant schemas require iOS 18.4+. The 27 releases (WWDC 2026, iOS 27 / macOS 27) add LongRunningIntent, EntityCollection, SyncableEntity, RelevantEntities, IndexedEntityQuery, OwnershipProvidingEntity, allowedExecutionTargets, the AppIntentsTesting framework, and native Duration/PersonNameComponents parameters; CancellableIntent and IntentValueRepresentation are iOS 26.4+; supportedModes is iOS 26+. Apple's own naming for the WWDC 2026 cohort is "the 27 releases" - use that, not "iOS 19".
Never make a SwiftData @Model class or other reference-type data model conform to AppEntity. AppEntity requires Sendable; @Model classes are not sendable. Create a separate struct entity that shadows the fields you want to expose.
Never pass ModelContext across actor boundaries. ModelContext is not sendable. Pass ModelContainer (which is sendable) and create a local context inside the actor that needs one.
Never expose an intent to Siri/Spotlight only by writing its type, always register it through an AppShortcutsProvider. Types not registered there will not appear in Shortcuts, Siri suggestions, or the action button picker.
Never write a Siri activation phrase without interpolating \(.applicationName). The App Intents macro rejects phrases without it at compile time, because phrases without the app name would collide with other apps' commands.
Never reach back into a SwiftUI @Query from inside an intent. @Query only works inside a View. Run a one-shot FetchDescriptor through a ModelContext instead, or route through a centralized data controller.
Output Format
If the user asks for a review, organize findings by file. For each issue:
State the file and relevant line(s).
Name the anti-pattern being replaced.
Show a brief before/after code fix.
Skip files with no issues. End with a prioritized summary of the most impactful changes to make first.
If the user asks you to write or fix intent code, make the changes directly instead of returning a findings report.
Example output:
RecentItemsIntent.swift
Line 18: SwiftData @Model cannot conform to AppEntity (not Sendable).
references/siri-intelligence.md - the Siri / Apple Intelligence integration model: finding content (semantic index / structured search / in-app search), custom responses, interaction donations, confirmation + OwnershipProvidingEntity, validation flow.
Never instantiate services, data stores, or authentication managers inside perform(). Inject them through @Dependency and register them once in App.init() with AppDependencyManager.shared.add(dependency:).
Never use String(format:) or manual concatenation for localized intent dialog. Use LocalizedStringResource, and use Foundation's grammar-agreement markdown (^[\(count) item](inflect: true)) inside an AttributedString for pluralization.
Prefer OpenIntent for "take me to this thing" actions, AppIntent & ShowsSnippetView for self-contained one-shot summaries, and AppIntent & ShowsSnippetIntent + a paired SnippetIntent when the snippet contains Button(intent:) and needs to re-render after buttons fire.
Always set static let isDiscoverable: Bool = false on helper intents that only back a widget button, snippet button, or other intent - otherwise they pollute the user's Shortcuts library.
When an intent mutates data that widgets or control widgets display, call WidgetCenter.shared.reloadAllTimelines() inside perform() before returning.
When Button(intent:) lives inside a widget view, share state between the intent (runs in the app process) and the widget's timeline provider (runs in the extension process) via App Group UserDefaults(suiteName:) or a shared ModelContainer URL - never UserDefaults.standard or in-memory @Dependency state.
When entity data that appears in a shortcut phrase's key path changes (creation, rename, deletion), call YourShortcutsProvider.updateAppShortcutParameters() to invalidate the cached candidate list.
Prefer EnumerableEntityQuery when the whole set is small and cheap to load; implement EntityQuery + EntityStringQuery when the dataset is large or searchable; add EntityPropertyQuery to get a system-generated Find intent for free. entities(for identifiers:) is mandatory on every query; without it, parameter resolution breaks.
Mark entity fields that users might filter, sort, or reference in parameter summaries with @Property (stored) or @ComputedProperty (derived). Plain stored properties are invisible to the App Intents framework.
Use TransientAppEntity for return data that's computed on the fly (summaries, aggregates). Don't try to shoehorn it into AppEntity with a fake id.
When the entity already has a universal-link URL, conform it to URLRepresentableEntity and conform the open intent to URLRepresentableIntent - the system routes opens through your existing link handler without a perform().
For entities visible onscreen that Siri should be able to understand, combine .userActivity(_:element:) on the view with Transferable conformance on the entity. Identification alone is not enough; Siri needs exportable content.
SnippetIntent.perform() is called multiple times per user interaction (initial show, after each button tap, on appearance changes, on reload()). Keep it pure: read state, assemble the view, return. Never mutate app state or kick off slow work inside a snippet intent's perform().
Respect hard limits: 10 AppShortcuts per app, 1000 total trigger phrases (including parameter expansions). The first phrase in each shortcut's array is the primary one - it's shown on the Shortcuts home tile and as Siri's "what can I do with X?" answer.
Include at least one non-parameterized phrase per App Shortcut so it's discoverable before first launch; parameterized phrases don't appear in Spotlight until the app has run once and populated the parameter cache.
Prefer supportedModes + continueInForeground (iOS 26+) over ForegroundContinuableIntent / needsToContinueInForegroundError for new code that conditionally foregrounds. The newer form lets one intent declare it may run background or foreground based on runtime state.
Snippet views have a 340-point height ceiling; beyond this, scrolling breaks the glance overlay model. Link to the full app for deep content, and keep snippet text larger than system defaults for legibility at reading distance.
For text parameters that may receive input from Apple Intelligence's Use Model action, declare the type as AttributedString rather than String so rich formatting (bold, italic, lists, tables) is preserved losslessly.
The app's App struct initializer is executed when an intent runs, even if the UI never appears. Do all intent-relevant setup (ModelContainer creation, AppDependencyManager.shared.add(...), log plumbing) inside init(), not inside view modifiers like .task or .onAppear.
LocalizedStringResource is the standard string type everywhere in App Intents (titles, dialog, parameter prompts). It shares string catalogs with SwiftUI, so localization works out of the box.
Grammar agreement (inflect: true) works in English, French, German, Italian, Spanish, and Portuguese (both variants). For other locales it falls back to the unmodified form.
Intents triggered from a system surface get ~30 seconds (no hard limit on macOS). For uploads, sync, large file ops, or on-device inference, conform to LongRunningIntent and do the work inside performBackgroundTask { ... } - and report progress regularly, because going quiet makes the system revoke the runtime extension. Pair with CancellableIntent to clean up with the cancellation reason. Don't try to beat the limit with a detached Task; unmanaged tasks aren't covered by the extension.
Use EntityCollection<Entity> instead of [Entity] for parameters that may hold many entities when perform() mostly needs ids - it skips full entity resolution during parameter handling. Call resolvedEntities() only when you need the full objects.
In schema-based update intents, distinguish "don't change" from "clear" with $parameter.valueState (.set(value) / .set(nil) / .unset), not a plain nil check - otherwise voice commands can never clear a field.
Conform entities to SyncableEntity when Siri might reference them in a cross-device conversation. If the id is already consistent everywhere (server UUID, CloudKit record id), just adopt the protocol; otherwise pair local and stable ids with SyncableEntityIdentifier.
Conform shareable entities to OwnershipProvidingEntity and keep their ownership (.shared / .public) current, so Siri confirms actions on content the user shared or published. By default Siri assumes entities are private and may skip confirmation.
To carry structured types other apps understand (a coordinate, a contact, a name), add an IntentValueRepresentation via the ValueRepresentation builder to the entity's transferRepresentation - FileRepresentation/DataRepresentation only cover file formats. Use the key-path form (ValueRepresentation(exporting: \.place)) when the entity already stores the system type.
For on-screen awareness, pick the annotation by shape: .userActivity for one primary item, .appEntityIdentifier(_:) for one among a few, .appEntityIdentifier(forSelectionType:) on a List/collection for many (lazy), .appEntityUIElements(_:) for custom Canvas/CALayer drawing. Annotations on Notifications, Now Playing, and AlarmKit require a real AppEntity - TransientAppEntity has no persistent id and can't be used there. Add a displayRepresentations query method so Siri can resolve onscreen entities without a full DB fetch.
Donate UI interactions with IntentDonationManager (the system already donates Siri/Shortcuts runs); donate after the action completes, don't over-donate (the system will start ignoring you), and delete stale donations.
Test with the AppIntentsTesting framework (out-of-process, in an XCUITest bundle) for the integration surface - intents, queries, spotlightQuery(), viewAnnotations(); use test-only intents (isDiscoverable = false + #if DEBUG) to seed data and reach internal state. Keep direct perform() struct tests for pure logic.
references/onscreen-awareness.md - the 4 on-screen annotation APIs, content transfer (IntentValueRepresentation, resolve vs import), displayRepresentations fast path, entity annotations on Notifications / Now Playing / AlarmKit.
references/testing-intents.md - the AppIntentsTesting framework (out-of-process integration), direct perform() unit tests, mocking @Dependency via AppDependencyManager, test-only intents, what you can't unit-test.
references/anti-patterns.md - common mistakes LLMs make when generating App Intents code.
