ios-performance

name	ios-performance
description	iOS 性能分析与测试技能。只在需要处理 UIKit / SwiftUI 的掉帧、启动慢、CPU / 内存压力、性能回归基线、`measure(metrics:)`、`xctrace` 或 Instruments 取证时使用；如果问题核心是 crash、异常、对象未释放根因、纯静态审查或普通单元/UI 测试补齐，不要把它当作主 skill；若任务产出修改了 Apple Xcode 项目相关内容，默认以定向测试/必要验证与独立 reviewer subAgent `code-review` 放行为收口；`ios-verification` / `ios-verification` 仅在用户显式要求或需要补强完整项目环境证据时按需使用。

iOS 性能分析与测试

Purpose

Analyze and improve iOS performance issues using the smallest credible evidence set, focusing on profiling strategy, baseline design, and before/after validation.

中文说明

该 Skill 负责性能基线设计、profiling 取证、模板选择、优化方向和前后对比验证。

• 适用于：掉帧、启动慢、CPU / 内存压力、性能回归、measure(metrics:)、xctrace、Instruments。
• 不适用于：普通业务实现、泛化 crash 排查、普通单元/UI 测试补齐。

When to Use

• UIKit / SwiftUI 列表滚动掉帧、动画卡顿、页面进入慢、启动慢。
• CPU 高、内存增长、主线程繁忙、SwiftUI 更新过多。
• 需要建立 XCTest 性能基线，如 measure {}、measure(metrics:)、XCTApplicationLaunchMetric。
• 需要用 xcrun xctrace / Instruments 录制 Time Profiler、Animation Hitches、Allocations、Leaks、App Launch。

When Not to Use

• 问题核心是 crash、异常、野指针、对象未释放根因；使用 debugging。
• 只是补业务单元测试、UI 测试或测试替身；使用 ios-feature-implementation(test-implementation)。
• 只是普通 SwiftUI/UIKit 功能实现；使用对应实现型 Skill。

Agent Rules

• Always define one symptom and one target interaction before profiling.
• Distinguish baseline design from runtime evidence collection.
• Prefer Release configuration and stable device / OS baselines for comparison.
• Output should include symptom, evidence, hypothesis, optimization direction, and validation method.
• When performance fixes edit code, document non-obvious performance invariants, caching/lifecycle side effects, concurrency boundaries, and fallback behavior in Chinese in touched code.
• Update stale comments, keep code comments Chinese by default, and avoid adding comments that merely restate optimization syntax.
• If the user asks for a formal HTML performance report, regression report, or shareable postmortem, prepare evidence summaries and route final document generation to html-docs.
• If code changes are produced, final closure still follows targeted validation / necessary verification plus independent reviewer subAgent code-review; the implementation Agent must not self-review.

Inputs

{
  "goal": "Analyze iOS performance issue",
  "symptom": "scroll hitch | slow launch | cpu spike | memory growth | unknown",
  "surface": "UIKit | SwiftUI | mixed",
  "available_evidence": [],
  "constraints": []
}

Outputs

{
  "status": "completed | partial | blocked",
  "symptom": "...",
  "evidence": [],
  "hypotheses": [],
  "recommended_tools": [],
  "optimization_directions": [],
  "validation_plan": [],
  "known_risks": [],
  "next_action": "ios-verification | code-review | debugging | ask-user | blocked"
}

Exit Conditions

• completed: symptom, evidence path, optimization direction, and validation plan are all explicit.
• partial: useful hypotheses exist but evidence or environment is incomplete.
• blocked: no stable reproduction path, no usable profiling environment, or user intent is actually a different problem class.

Escalation Rules

• Escalate to debugging when the issue is primarily crash, hang, leak root cause, or runtime behavior diagnosis.
• Escalate to ios-feature-implementation(test-implementation) when benchmark/regression test code is needed; escalate to ios-verification for low-cost targeted validation.
• Escalate to apple-docs when official API or Instruments behavior must be confirmed.
• Escalate to html-docs when the performance evidence and conclusions must become a formal HTML report or handoff document.

Token Budget

• Do not paste raw xctrace dumps or large logs.
• Prefer local scripts or command-line aggregation to produce compact top stacks, time windows, thread hotspots, allocation deltas, and artifact paths before handing evidence to the Agent.
• For .trace packages, prefer scripts/summarize-xctrace.py to export selected xctrace tables and consume trace-summary.json / trace-summary.md instead of raw XML exports.
• Prefer compact symptom summaries, selected counters, and one clear evidence thread.
• Avoid mixing unrelated hotspots into the same report.

Reference Resources

• scripts/summarize-xctrace.py: exports high-signal xctrace tables and writes compact trace-summary.json / trace-summary.md evidence, including actionable findings, thread hotspots, stack signatures, and symbolication diagnostics.

Relationship to Other Skills

• Use debugging for crash and runtime fault analysis.
• Use ios-feature-implementation(test-implementation) for test authoring and ios-verification for targeted validation execution.
• Use ios-feature-implementation with advanced-swift mode when performance work reveals deeper concurrency or abstraction implementation needs.
• Use apple-docs for official API and Instruments fact lookup.
• Use html-docs for final formal HTML performance reports; this Skill only owns the evidence, hypotheses, and validation plan.