mobile-testing

name	mobile-testing
description	Mobile testing capability pack. Covers E2E testing (Detox/Maestro), unit testing (Jest/RNTL), device compatibility matrices, performance budgets, VoiceOver accessibility, human-AI pair testing, and mobile test strategy — iOS/React Native first. Use for any mobile app testing, mobile test strategy, or mobile QA task.
version	0.1.0
type	reference-based
keywords	["mobile testing","移动端测试","E2E","端到端测试","Detox","Maestro","Appium","React Native","Jest","RNTL","unit test","单元测试","device compatibility","设备兼容性","simulator","模拟器","performance","性能测试","VoiceOver","accessibility","无障碍","可访问性","pair testing","人机协作测试","test strategy","测试策略","iOS","flaky","片状"]

Mobile Testing Capability Pack

Full-stack mobile testing judgment — E2E, unit, device compatibility, performance, accessibility, human-AI pair testing, and test strategy. Scope (v0.1, inherited from source v1.0): iOS + React Native 为主线；Android（TalkBack、adb、emulator）与原生 Swift/Kotlin 单元测试为扩展方向。核心差异 vs Web：模拟器/真机、设备碎片化、VoiceOver。 UNIVERSAL RULE: Fabricated test results, pass rates, coverage numbers, performance numbers, compatibility results, findings, or screenshots = FAIL. Run actual measurements or explicitly report "not run".

---

Step 1: Context Detection

User Signal / Task Type	Load Reference
E2E, user flow test, Detox, Maestro, Appium, 端到端, simulator flow	references/e2e-testing.md
unit test, Jest, RNTL, component test, hook test, coverage, 单元测试, mock	references/unit-testing.md
device matrix, screen sizes, OS versions, iPhone SE, iPad, 设备兼容, 碎片化	references/device-compatibility.md
startup time, FPS, memory, bundle size, 冷启动, 性能, 帧率, 内存	references/performance-testing.md
accessibility, VoiceOver, TalkBack, a11y, WCAG, 无障碍, 可访问性	references/accessibility.md
pair testing, 4D protocol, human feel, 手感, 协作测试, exploratory	references/pair-testing.md
test strategy, pyramid, CI/CD testing, 测试策略, 金字塔, test plan	references/test-strategy.md
before Gate review / reviewing mobile test work / acceptance check	references/review-checklist.md

Multi-signal: load all matched references.

---

Step 2: Decision Entry Point

Q1 — What kind of testing task?

• New app / no tests yet → START with test-strategy.md (pyramid + CI design), then per-layer references
• Writing/fixing user-flow tests → e2e-testing.md
• Writing/fixing component or logic tests → unit-testing.md
• "Does it work on device X / iOS Y?" → device-compatibility.md
• "App feels slow / big" → performance-testing.md
• Compliance / screen reader → accessibility.md
• Human wants to test alongside AI → pair-testing.md

Q2 — Which E2E framework? (details + comparison table in e2e-testing.md)

• React Native project → Detox (灰盒, <2% 片状率, 最快)
• 原生/跨框架 → Maestro (黑盒 YAML, 低学习曲线)
• 企业级 WebDriver 需求 → Appium (accept 10-15% 片状率 + 最慢)
• Selection MUST include a comparison table and a project-match rationale.

Q3 — Reviewing or accepting work? → ALSO load review-checklist.md (personas + Gate 2/Gate 4 checklists).

---

Step 3: Core Judgment Rules

E2E (references/e2e-testing.md)

• 元素选择 MUST 用 testID；never text matchers（动态内容/语言切换会碎）
• never sleep() — MUST 用显式等待 waitFor().toBeVisible().withTimeout(5000)
• ≥5 个核心流程；每个测试独立（不依赖执行顺序）；失败自动截图
• 10-20 个核心流程是上限区间 — 超过 20 个 = 维护成本爆炸
• Maestro: 禁用 JavaScript 表达式（GraalJS 限制）；clearState 不清 Keychain
• 片状率 >5% = 必须修复（策略层定为 P0）

Unit (references/unit-testing.md)

• 查询优先级 MUST 遵守: getByRole > getByLabelText > getByText > getByTestId
• 禁止 getByType 和直接访问组件实例（测行为不测实现）
• 覆盖率: ≥80% 关键模块, ≥60% 整体；100% 通过率；测试 co-located
• 快照测试仅用于稳定 UI 组件

Device Compatibility (references/device-compatibility.md)

• 最小矩阵 ≥4 台（最小屏/主流/最大屏/平板），每台有"为什么测"理由
• OS 覆盖 = 当前版本 + 前一个大版本；每台设备 MUST 有截图证据
• 布局溢出/截断问题 = 0 才能通过

Performance (references/performance-testing.md)

• 阈值: 冷启动 <2s；JS Bundle <1.5MB；帧率 >55 FPS；内存 <250MB 活跃/<50MB 后台；二进制 <50MB
• MUST 在 Release 模式测量（Debug 结果完全不同）
• 帧率看 p99 不只平均值；内存看趋势不只单次快照（泄漏需要时间暴露）

Accessibility (references/accessibility.md)

• 自动化只覆盖 30-40% — MUST 加手动 VoiceOver 层才能宣告达标
• WCAG 2.2 AA: 对比度 ≥4.5:1；触控目标 ≥44×44pt；可交互元素全有 accessibilityLabel
• Label 描述操作（"保存到收藏"）而非元素类型（"按钮"）

Pair Testing (references/pair-testing.md)

• 严重度由人类决定 — AI never 单独判断（手感/动画卡顿属于人域）
• MUST 覆盖移动端特有 round: 手势/弱网/状态转换/键盘
• 每轮有截图证据；Fix Now 决策有即时 handoff

Strategy (references/test-strategy.md)

• 70/20/10 金字塔只是起点 — MUST 按 App 类型调整并给出依据
• 覆盖率目标按模块设定，never 全局平均
• 设备测试分层: 模拟器(CI) + 云测试(release) + 真机(关键版本)

---

Anti-Patterns

E2E

• ❌ sleep(3000) 替代 waitFor = 不稳定 + 慢
• ❌ by.text('动态内容') = 语言切换/数据变化后必 FAIL
• ❌ 测试间共享状态 = 顺序依赖 = 片状
• ❌ 只测 happy path 不测错误/边界场景
• ❌ E2E 测试超过 20 个流程 = 维护成本爆炸（10-20 个核心流程足够）

Unit

• ❌ getByType(Button) = 测试实现细节而非行为
• ❌ 快照测试覆盖所有组件 = 每次 UI 微调都要更新快照
• ❌ 测试间共享可变状态
• ❌ mock 过度 = 测试只验证 mock 行为不验证真实逻辑

Device Compatibility

• ❌ 只在一台设备上测试 = 小屏幕/大屏幕布局问题遗漏
• ❌ 只测最新 OS = 老版本用户遇到崩溃
• ❌ 不测横屏（如果 App 支持旋转）
• ❌ 兼容性报告没有截图 = 无法验证

Performance

• ❌ 只在 Debug 模式测性能（Release 模式结果完全不同）
• ❌ 只看整体 bundle 大小不分析依赖拆分
• ❌ 帧率只看平均值不看 p99（偶尔卡顿用户也能感知）
• ❌ 内存测试只看单次快照不看趋势（泄漏需要时间暴露）

Accessibility

• ❌ 只跑自动化工具就宣告 a11y 达标（自动化只覆盖 30-40%）
• ❌ accessibilityLabel 写'按钮'而非操作描述（'保存到收藏'）
• ❌ 图片装饰性标注为'图片'而非设为 accessible={false}
• ❌ VoiceOver 焦点顺序和视觉顺序不一致

Pair Testing

• ❌ AI 单独决定严重度（人类的'感觉'无法量化但很重要）
• ❌ 只看静态截图不测动态交互（动画/手势需要人类判断）
• ❌ 跳过弱网测试（移动端最常见的体验问题）

Strategy

• ❌ 通用 70/20/10 不根据项目调整
• ❌ 没有片状率监控 = 测试套件慢慢腐烂
• ❌ 只用模拟器不用真机 = 漏掉真机特有问题
• ❌ 性能只测一次不做回归 = 慢慢变慢没人发现

---

Anti-Skip Table

Shortcut Attempt	Required Action
"I'll just pick Detox/Maestro, it's obvious"	MUST produce the comparison table + project-match rationale in e2e-testing.md — framework depends on project stack + CI needs
"A sleep() here is fine, it's one test"	MUST use explicit waits — zero-sleep is a pass/fail criterion, not a preference
"Tests pass on my simulator, ship it"	MUST run the ≥4-device matrix in device-compatibility.md with per-device screenshots
"Performance looks fine in dev"	MUST measure in Release mode against the thresholds in performance-testing.md
"eslint a11y passed, we're accessible"	MUST complete the manual VoiceOver checklist in accessibility.md — automation covers only 30-40%
"AI can rate this jank severity"	MUST hand severity judgment to the human per pair-testing.md — feel is a human-domain call
"70/20/10 pyramid, done"	MUST justify the ratio for THIS project type per test-strategy.md
"I'll report the numbers we expect"	Fabricated results/coverage/screenshots = FAIL across every capability. Run it or report "not run"