performance-profiling

Guide performance profiling with Instruments, diagnose hangs, memory issues, slow launches, and energy drain. Use when reviewing app performance or …

不碰外部（只输出文字）无严重或高危命中rshankras/claude-code-apple-skills

它会碰到什么

扫了多少6 个文本文件，96 KB

它会碰到什么不碰外部（只输出文字）

命中总数5 处

命中统计严重 0 · 高 0 · 中 0 · 低 0

这一栏是扫描器报的事实，不是结论。命中多不等于有毒（安全工具、规则库、示例脚本本来就会包含危险写法），命中少也不等于干净。它和你手上的凭据、文件、网络有什么关系，需要你自己看。

技能内容

Performance Profiling

Systematic guide for profiling Apple platform apps using Instruments, Xcode diagnostics, and MetricKit. Covers CPU, memory, launch time, and energy analysis with actionable fix patterns.

When This Skill Activates

Use this skill when the user:

Reports app hangs, stutters, or dropped frames
Needs to profile CPU usage or find hot code paths
Has memory leaks, high memory usage, or OOM crashes
Wants to optimize app launch time
Needs to reduce battery/energy impact
Asks about Instruments, Time Profiler, Allocations, or Leaks
Wants to add os_signpost or performance measurement to code
Is preparing for App Store review and needs performance validation

Decision Tree

What performance problem are you investigating?
│
├─ App hangs / unresponsive to taps / slow UI
│  └─ Read time-profiler.md
│
├─ Scroll or animation stutter / dropped frames / hitches
│  └─ Read hitches.md
│
├─ High memory / leaks / OOM crashes / growing footprint
│  └─ Read memory-profiling.md
│
├─ Slow app launch / time to first frame
│  └─ Read launch-optimization.md
│
├─ Battery drain / thermal throttling / background energy
│  └─ Read energy-diagnostics.md
│
├─ General "app feels slow" (unknown cause)
│  └─ Start with time-profiler.md, then memory-profiling.md
│
└─ Pre-release performance audit
   └─ Read ALL reference files, use Review Checklist below

Quick Reference

|---------|-------------------|------------|-----------|

The 8 Key Metrics (WWDC21 10181)

Apple enumerates exactly eight things to track for app performance, all coverable with five tools (Xcode Organizer, MetricKit, Instruments, XCTest, App Store Connect API):

|---|--------|--------------------|------|

| 1 | Battery usage | Energy Gauge flags CPU use > 20% as High CPU, plus CPU Wake Overhead regions; top subsystems to watch: CPU, Networking, Location | Energy Gauge → Time Profiler |

The Organizer aggregates all of these from consented user devices across the last 16 app versions; the Regressions pane (Xcode 13) isolates every metric that increased significantly in the most recent version, in one place. The same data is available as JSON via the App Store Connect API (WWDC21 10181).

Process

1. Identify the Problem Category

Ask the user or inspect their description to classify the issue:

Responsiveness: Hangs, stutters, animation drops
Memory: Leaks, growth, OOM crashes
Launch: Slow cold/warm start
Energy: Battery drain, thermal throttling

2. Read the Appropriate Reference File

Each file contains:

Which Instruments template to use
Step-by-step profiling workflow
How to interpret results
Common fix patterns with code examples

3. Profile on Real Hardware

Always remind users:

Profile on device, not Simulator (Simulator uses host CPU/memory)
Use Release build configuration (optimizations change behavior)
Profile with representative data (empty databases hide real perf)
Close other apps to reduce noise

4. Apply Fixes and Verify

After identifying bottlenecks:

Apply targeted fix from the reference file
Re-profile to confirm improvement
Add os_signpost markers for ongoing monitoring

Xcode Diagnostic Settings

Recommend enabling these in Scheme > Run > Diagnostics:

| Setting | What It Catches |

|---------|-----------------|

| Main Thread Checker | UI work off main thread |

| Thread Sanitizer | Data races |

| Address Sanitizer | Buffer overflows, use-after-free |

| Malloc Stack Logging | Memory allocation call stacks |

| Zombie Objects | Messages to deallocated objects |

MetricKit Integration

For production monitoring, recommend MetricKit (WWDC20 10081, WWDC21 10181):

import MetricKit

final class PerformanceReporter: NSObject, MXMetricManagerSubscriber {
    func startCollecting() {
        MXMetricManager.shared.add(self)   // subscribe once, early in launch
    }
    // best practice: remove(self) in deinit (WWDC21 10181)

    func didReceive(_ payloads: [MXMetricPayload]) {
        for payload in payloads {
            // Launch time
            if let launch = payload.applicationLaunchMetrics {
                log("Resume time: \(launch.histogrammedResumeTime)")
            }
            // Hang rate
            if let responsiveness = payload.applicationResponsivenessMetrics {
                log("Hang time: \(responsiveness.histogrammedApplicationHangTime)")
            }
            // Memory
            if let memory = payload.memoryMetrics {
                log("Peak memory: \(memory.peakMemoryUsage)")
            }
            // Scroll hitches (ratio of time hitching to time scrolling)
            if let animation = payload.animationMetrics {
                log("Scroll hitch ratio: \(animation.scrollHitchTimeRatio)")
            }
            // Exit reasons — daily counts per termination cause, fg + bg
            if let exits = payload.applicationExitMetrics {
                log("Exits: \(exits.backgroundExitData)")
            }
        }
    }

    func didReceive(_ payloads: [MXDiagnosticPayload]) {
        for payload in payloads {
            if let hangs = payload.hangDiagnostics {
                for hang in hangs {
                    log("Hang: \(hang.callStackTree)")
                }
            }
        }
    }
}

Key facts (WWDC20 10081, WWDC21 10181):

❌ Subscribing lazily (e.g. in a settings screen) silently misses delivered payloads — ✅ subscribe in App/AppDelegate init; the previous day's payloads are handed to you on launch.
Metrics arrive as 24-hour payloads, at most once per day, in three aggregation forms: cumulative, averaged, and bucketized (MXHistogram). On iOS 15+/macOS 12+, all diagnostics are delivered immediately after the issue occurs instead of daily.
Termination and memory telemetry arrive in the daily metric payload by default — no extra instrumentation.
Diagnostic types: MXHangDiagnostic (main-thread unresponsive time + backtraces), MXCrashDiagnostic (exception info, termination reason, VM region info), MXCPUExceptionDiagnostic (threads burning CPU — the programmatic form of Organizer energy logs), MXDiskWriteExceptionDiagnostic (fires on the 1 GB/day write threshold).
MXCallStackTree backtraces are unsymbolicated by design — symbolicate off-device with atos-class tools + your dSYMs; ❌ don't try on-device.
MXSignpost marks critical code sections for field telemetry; the animation variant captures hitch-rate telemetry for the interval:

let handle = MXMetricManager.makeLogHandle(category: "animation_telemetry")
mxSignpostAnimationIntervalBegin(log: handle, name: "custom_animation")
// ... animation ...
mxSignpost(OSSignpostType.end, log: handle, name: "custom_animation")

Review Checklist

Responsiveness

[ ] No synchronous work on main thread > 100ms
[ ] No file I/O or network calls on main thread
[ ] Core Data / SwiftData fetches use background contexts for large queries
[ ] Images decoded off main thread (use .preparingThumbnail or async decoding)
[ ] @MainActor only on code that truly needs UI access