Conciseness

Prose is terse and every section is SwiftData-specific: version-gated facts Claude cannot reliably know (iOS 18 unidirectional relationships, iOS 26 model inheritance, default deleteRule .nullify, ModelContext non-Sendable, @ModelActor save requirements). Nothing explains Swift/SwiftUI basics, so it matches anchor 5 ('every token earns its place') rather than anchor 4, which would require trimmable over-explanation.

Actionability

Nearly all code is executable and covers the common cases (model definition, container setup, CRUD, @Query variants, migration plan, @ModelActor), matching anchor 5 in spirit. But minor gaps exist: the custom-migration snippet references an undefined SchemaV3 and declares 'static let migrateV2toV3' outside any enum, and the #Predicate example uses a 'tags' relationship absent from the Trip model shown earlier. Anchor 4 ('minor gaps') is the best fit.

Workflow Clarity

Destructive/batch operations include explicit feedback loops — 'first run the exact predicate or version hop against a disposable copy and record affected identifiers/counts... On failure, fix the predicate/schema and restore a pristine fixture before retry', plus 'migrate a fresh copy of each old store and assert post-migration data before release' and a review checklist — so the batch-operation cap does not apply. Still, the skill is organized topically rather than as a sequenced workflow, and some processes (e.g., CloudKit schema rollout) verify only implicitly, fitting anchor 4 rather than 5.

Progressive Disclosure

The body keeps core patterns inline while bulk detail is split across five real, one-level-deep reference files (swiftdata-advanced, swiftdata-queries, core-data-coexistence, predicate-pitfalls, indexing), each verified to exist and linked with a descriptive label in the References section. This matches anchor 5 ('well-signaled one-level-deep references; content appropriately split'); it is not anchor 4, since no bulk content that belongs in a reference is inlined.

Specificity

Multiple concrete actions are named ('Implement, review, or improve data persistence using SwiftData') and elaborated across the when-clauses: defining @Model classes, querying with @Query/#Predicate/FetchDescriptor, configuring ModelContainer/ModelContext, planning migrations with VersionedSchema/SchemaMigrationPlan, setting up CloudKit sync, and migrating from Core Data. Comprehensive coverage of the domain matches the anchor-5 example rather than anchor 4, which requires minor gaps.

Completeness

Explicitly answers 'what' ('Implement, review, or improve data persistence using SwiftData') and 'when' with six concrete trigger clauses ('Use when defining... when querying... when configuring... when planning... when setting up... or when coexisting with or migrating from Core Data'). This matches the anchor-5 example pattern of actions plus an explicit 'Use when' clause; it is well above anchor 4's weaker 'when'.

Trigger Term Quality

Covers the natural terms a user would say for this domain — 'SwiftData', 'data persistence', 'CloudKit sync', 'schema migrations', 'Core Data' — plus the exact API names (@Model, @Attribute, #Predicate, FetchDescriptor) users quote verbatim when asking for help. No meaningful synonym is missing; it is not anchor 4, which expects a few natural terms absent.

Distinctiveness Conflict Risk

The description carves a clear SwiftData niche and even names the Core Data boundary ('coexisting with or migrating from Core Data'), which minimizes overlap with a sibling core-data skill rather than inviting it. Written in third-person/infinitive voice with no fluff; distinct triggers make wrong-skill triggering unlikely.