Agent Skills: Usage Insights Generator

Usage Insights Generator

Generate a production usage insights system that records user activity events with SwiftData, computes personalized insights (streaks, most active day, top categories), and displays them in a dashboard with insight cards, period pickers, trend indicators, and optional shareable recap screens.

When This Skill Activates

Use this skill when the user:

• Asks to "show usage statistics" or "add usage stats"
• Wants "user insights" or "activity insights"
• Mentions "activity summary" or "weekly summary"
• Asks about a "usage dashboard" or "insights dashboard"
• Wants a "weekly recap" or "monthly recap"
• Mentions "year in review" or "year-in-review"
• Asks for "Spotify Wrapped style" or "Wrapped-style recap"
• Wants to "show the user their activity" or "personalized stats"

Pre-Generation Checks

1. Project Context Detection

• [ ] Check Swift version (requires Swift 5.9+)
• [ ] Check deployment target (iOS 17+ / macOS 14+ for @Observable and SwiftData)
• [ ] Identify source file locations
• [ ] Check for Swift Charts availability (iOS 16+ / macOS 13+, but recommend iOS 17+)

2. Conflict Detection

Search for existing usage tracking or insights code:

Glob: **/*UsageEvent*.swift, **/*Insight*.swift, **/*Recap*.swift, **/*ActivityLog*.swift
Grep: "UsageEvent" or "InsightCalculator" or "activitySummary" or "SwiftData" and "event"

If existing analytics/tracking found:

• Ask if user wants to build insights on top of existing event data
• If yes, adapt InsightCalculator to work with existing models
• If no, generate fresh event recording alongside existing code

3. Data Layer Detection

Check for SwiftData usage:

Grep: "import SwiftData" or "@Model" or "ModelContainer"

If SwiftData already in use:

• Integrate UsageEvent into existing ModelContainer
• Use existing schema migration strategy

If no SwiftData:

• Generate full setup including ModelContainer configuration

Configuration Questions

Ask user via AskUserQuestion:

1. Insight period?

   • Daily (today's activity breakdown)
   • Weekly (7-day recap with day-by-day comparison) — recommended
   • Monthly (30-day trends with weekly rollups)
   • Yearly (year-in-review with monthly highlights)
   • All of the above (period picker lets user switch)

2. Visualization style?

   • Cards only (simple stat cards with trend indicators)
   • Charts only (Swift Charts bar/line graphs)
   • Both cards and charts — recommended

3. Include shareable recap card?

   • Yes (generates a recap view that can be rendered to an image and shared)
   • No (dashboard only, no sharing)

4. Data source?

   • SwiftData (generate UsageEvent model and recorder) — recommended
   • Custom (user provides their own event data; generate calculator and views only)

Generation Process

Step 1: Read Templates

Read templates.md for production Swift code.

Step 2: Create Core Data Files

Generate these files:

1. UsageEvent.swift — SwiftData @Model for recording user activity events
2. InsightResult.swift — Model for computed insights (title, value, trend, visualization type)

Step 3: Create Calculation Engine

3. InsightCalculator.swift — Pure functions that aggregate events into insights

Step 4: Create UI Files

4. InsightsDashboardView.swift — Main dashboard with grid of insight cards and period picker
5. InsightCardView.swift — Individual insight card with icon, value, trend indicator, sparkline

Step 5: Create Optional Files

Based on configuration:

• UsageRecapView.swift — If shareable recap selected (paged summary with share card generation)
• UsageEventRecorder.swift — If SwiftData data source selected (convenience class for recording events)

Step 6: Determine File Location

Check project structure:

• If Sources/ exists -> Sources/UsageInsights/
• If App/ exists -> App/UsageInsights/
• Otherwise -> UsageInsights/

Output Format

After generation, provide:

Files Created

UsageInsights/
├── UsageEvent.swift           # SwiftData @Model for activity events
├── InsightResult.swift        # Computed insight model
├── InsightCalculator.swift    # Aggregation engine
├── InsightsDashboardView.swift # Dashboard with period picker
├── InsightCardView.swift      # Individual insight card
├── UsageRecapView.swift       # Shareable recap (optional)
└── UsageEventRecorder.swift   # Event recording helper (optional)

Integration Steps

Add ModelContainer (if not already present):

@main
struct MyApp: App {
    var body: some Scene {
        WindowGroup {
            ContentView()
        }
        .modelContainer(for: [UsageEvent.self])
    }
}

Record events from anywhere in the app:

struct TaskDetailView: View {
    @Environment(\.modelContext) private var modelContext
    @State private var recorder: UsageEventRecorder?

    var body: some View {
        Button("Complete Task") {
            completeTask()
            recorder?.record(
                .taskCompleted,
                metadata: ["category": "work", "priority": "high"]
            )
        }
        .onAppear {
            recorder = UsageEventRecorder(modelContext: modelContext)
        }
    }
}

Show the insights dashboard:

NavigationLink("My Insights") {
    InsightsDashboardView()
}

Show a weekly recap:

struct WeeklyRecapSheet: View {
    @Environment(\.modelContext) private var modelContext
    @State private var showRecap = false

    var body: some View {
        Button("View Weekly Recap") {
            showRecap = true
        }
        .sheet(isPresented: $showRecap) {
            UsageRecapView(period: .week)
        }
    }
}

Record events with duration tracking:

// Start a timed session
let sessionStart = Date()

// ... user does work ...

// Record when session ends
recorder?.record(
    .sessionCompleted,
    metadata: ["screen": "editor"],
    duration: Date().timeIntervalSince(sessionStart)
)

Testing

@Test
func calculatesWeeklySummaryCorrectly() async throws {
    let calendar = Calendar.current
    let now = Date()
    let events: [UsageEvent] = (0..<7).flatMap { dayOffset in
        let date = calendar.date(byAdding: .day, value: -dayOffset, to: now)!
        return (0..<(dayOffset == 2 ? 5 : 2)).map { _ in
            UsageEvent(eventType: "taskCompleted", timestamp: date)
        }
    }

    let calculator = InsightCalculator()
    let insights = calculator.weeklySummary(from: events, referenceDate: now)

    let totalInsight = insights.first { $0.title == "Total Events" }
    #expect(totalInsight != nil)
    #expect(totalInsight?.value == "19") // 5 + (6 * 2)
}

@Test
func identifiesMostActiveDay() async throws {
    let calendar = Calendar.current
    let now = Date()

    // Create 5 events on Wednesday, 2 on other days
    var events: [UsageEvent] = []
    for dayOffset in 0..<7 {
        let date = calendar.date(byAdding: .day, value: -dayOffset, to: now)!
        let count = calendar.component(.weekday, from: date) == 4 ? 5 : 1
        for _ in 0..<count {
            events.append(UsageEvent(eventType: "action", timestamp: date))
        }
    }

    let calculator = InsightCalculator()
    let insights = calculator.weeklySummary(from: events, referenceDate: now)
    let mostActive = insights.first { $0.title == "Most Active Day" }
    #expect(mostActive?.value == "Wednesday")
}

@Test
func handlesEmptyEventList() async throws {
    let calculator = InsightCalculator()
    let insights = calculator.weeklySummary(from: [], referenceDate: Date())

    #expect(!insights.isEmpty) // Should still return cards with zero values
    let totalInsight = insights.first { $0.title == "Total Events" }
    #expect(totalInsight?.value == "0")
}

@Test
func recorderBatchesWritesForPerformance() async throws {
    let config = ModelConfiguration(isStoredInMemoryOnly: true)
    let container = try ModelContainer(for: UsageEvent.self, configurations: config)
    let context = ModelContext(container)
    let recorder = UsageEventRecorder(modelContext: context)

    // Record 100 events rapidly
    for i in 0..<100 {
        recorder.record(.custom("event_\(i)"))
    }

    // Flush pending writes
    recorder.flush()

    let descriptor = FetchDescriptor<UsageEvent>()
    let count = try context.fetchCount(descriptor)
    #expect(count == 100)
}

Common Patterns

Record an Event

recorder?.record(.featureUsed, metadata: ["feature": "darkMode"])

Calculate a Weekly Summary

let calculator = InsightCalculator()
let events = try modelContext.fetch(FetchDescriptor<UsageEvent>())
let insights = calculator.weeklySummary(from: events, referenceDate: Date())

Render Insight Cards in a Grid

LazyVGrid(columns: [GridItem(.flexible()), GridItem(.flexible())], spacing: 16) {
    ForEach(insights) { insight in
        InsightCardView(insight: insight)
    }
}

Generate and Share a Recap Image

let renderer = ImageRenderer(content: UsageRecapView(period: .week))
renderer.scale = 2.0
if let image = renderer.uiImage {
    // Use with ShareLink or UIActivityViewController
}

Gotchas

Privacy: All Data Stays On-Device

• Usage events are stored in SwiftData on the user's device only
• Never transmit usage data to a server (that's what analytics-setup is for)
• Clearly communicate to users that insights are computed locally
• Consider adding a "Delete My Data" option in settings

Performance with Large Datasets

• Fetch only the events needed for the current period (use #Predicate with date range)
• For yearly summaries, pre-aggregate monthly totals instead of scanning all events
• Batch event recording to avoid excessive SwiftData writes (use UsageEventRecorder)
• Set a retention policy — auto-delete events older than 1-2 years

Calendar-Based Period Calculations

• Always use Calendar.current for date math — never assume 7 days = 1 week
• Respect the user's first-day-of-week setting (calendar.firstWeekday)
• Use calendar.dateInterval(of: .weekOfYear, for: date) for accurate week boundaries
• Handle timezone changes gracefully — store timestamps in UTC, display in local time

Swift Charts Considerations

• Swift Charts is iOS 16+ / macOS 13+, but works best with iOS 17+ for latest features
• Keep chart data points reasonable (7 for weekly, 30 for monthly, 12 for yearly)
• Use .chartYScale(domain:) to prevent axis from starting at a misleading value
• Provide VoiceOver descriptions with .accessibilityLabel on chart marks

Trend Comparison Edge Cases

• First week of use has no "previous period" — show "New!" instead of a trend arrow
• Handle periods with zero events gracefully (don't divide by zero for averages)
• Percentage changes can be misleading for small numbers (1 -> 2 = +100%)

References

• templates.md — All production Swift templates
• Related: generators/share-card — Render recap views as shareable images
• Related: generators/analytics-setup — Backend analytics (complements on-device insights)
• Related: generators/streak-tracker — Streak tracking pairs well with usage insights
• Related: generators/milestone-celebration — Celebrate milestones surfaced by insights