swift-testing（Swift 测试）

Swift Testing (XCUITest)

This skill teaches the agent how to turn plain-English iOS UI scenarios into production-grade XCUITest code that is stable, readable, scalable, and CI-friendly.

When to use this skill

Use this skill when the user asks for any of the following:

• Write or refactor XCUITest UI tests in Swift
• Convert a scenario / user story into XCUITest
• Stabilize flaky UI tests (waits, scrolling, selectors, alerts)
• Create screen/page objects for XCUITest suites

Do not use this skill for:

• Unit tests, integration tests, snapshot tests
• Appium/Detox/Espresso (non-XCUITest)
• Performance tests or Xcode Instruments workflows

Inputs to ask for (only if missing)

Ask ONLY for what’s needed to generate correct tests; do not over-question.

1. Scenario

• Steps the user wants to automate
• Expected outcomes / assertions

1. Selector availability

• Do accessibility identifiers exist already?
• If not, generate the list of identifiers needed

1. App launch & environment

• Is login required? Are there feature flags?
• Should tests start from a deep link or home screen?
• Any special environment setup (staging, mock server)?

1. System dialogs

• Are permission prompts expected (notifications, location, camera, etc.)?

If the user cannot provide identifiers, proceed anyway:

• Generate test code using *assumed* identifiers (clearly labeled)
• Output the “Identifiers to add” list as part of the deliverable

Output contract (STRICT)

When you generate a test, you MUST always output:

A) Test plan

• numbered steps
• assertions after critical steps

B) Swift XCUITest code

• clean, compiling, copy/paste ready

C) Screen Objects

• created or updated (with selectors centralized)

D) Accessibility identifiers required

• exact strings and where they’re used

E) Stability notes

• waits, scroll strategy, alert handling, anchors

F) App-side implementation code (REQUIRED)

• Swift code to add accessibility identifiers to the app
• Must be copy/paste ready for the ViewController
• Include viewDidLoad() setup method
• Include helper extensions if needed (e.g., UIView.allSubviews())

App-Side Implementation (CRITICAL)

Tests WILL FAIL if accessibility identifiers are not implemented in the app. Always generate:

1. ViewController Setup Code

override func viewDidLoad() {
    super.viewDidLoad()
    setupAccessibilityIdentifiers()
}

private func setupAccessibilityIdentifiers() {
    myLabel.accessibilityIdentifier = "screen.label"
    myButton.accessibilityIdentifier = "screen.button"
}

2. For elements without IBOutlets

private func findAndSetAccessibilityIdentifiers() {
    for subview in view.allSubviews() {
        guard let button = subview as? UIButton,
              let title = button.configuration?.title ?? button.title(for: .normal) else {
            continue
        }
        switch title {
        case "Submit":
            button.accessibilityIdentifier = "screen.button.submit"
        default:
            break
        }
    }
}

3. Required UIView Extension

extension UIView {
    func allSubviews() -> [UIView] {
        var result = subviews
        for subview in subviews {
            result.append(contentsOf: subview.allSubviews())
        }
        return result
    }
}

Pre-Flight Verification (REQUIRED)

Before tests can run successfully, verify:

Common Failure: "Element not found"

Cause: Accessibility identifiers not set in app code

Fix:

1. Add setupAccessibilityIdentifiers() to viewDidLoad()
2. Rebuild app (⌘+B)
3. Re-run tests (⌘+U)

Non-negotiable engineering rules

Selector rules

1. Prefer accessibilityIdentifier selectors:

- app.buttons["id"] - app.textFields["id"] - app.staticTexts["id"]

1. Avoid queries by localized labels, titles, or dynamic text.
2. Avoid XPath-like approaches (not applicable here) and brittle hierarchy traversal.
3. If identifiers are missing, output the required list and proceed with assumed IDs.

Waiting/synchronization rules

• Never use sleep().
• Use explicit waits:

- waitForExistence - XCTNSPredicateExpectation for exists / hittable

• Wait on navigation anchors (a stable element that proves the screen is loaded).

Assertions

• Assert after each meaningful navigation:

- screen loaded anchor visible - error state visible - success state visible

• Prefer XCTAssertTrue(anchor.exists) only after waitForVisible.

Scrolling

• Use bounded scrolling with max swipes.
• If not found after max swipes, fail with diagnostics.

Failure diagnostics (always)

• Screenshot on failure (keepAlways)
• Include meaningful failure messages (“Expected Home title to appear…”)

Project conventions the agent must follow

Naming

• Test classes: FeatureFlowTests (e.g., LoginFlowTests)
• Test methods: test_<action>_<expectedOutcome>()
• Screen objects: LoginScreen, HomeScreen, SettingsScreen
• Identifier style: screen.element (e.g., login.email, home.title)

Structure (recommended)

• UITests/BaseUITestCase.swift
• UITests/Helpers/*.swift
• UITests/Screens/*.swift
• UITests/Tests/*.swift

Required helper set

When writing any test suite, prefer reusing the helper patterns in templates/:

• templates/BaseUITestCase.swift
• templates/XCUIElement+Waits.swift
• templates/XCUIApplication+Scroll.swift
• templates/SystemAlerts.swift

What to generate (decision rules)

If user gives a scenario only

Generate:

• screen objects required for that scenario
• the test case using those screen objects
• identifier list to add
• app-side implementation code for identifiers

If user provides existing test code

Refactor into:

• screen objects
• helpers
• explicit waits
• stable selectors …and return the improved code.

If user asks to “make tests easier”

Introduce:

• screen objects
• small helper APIs (tapWhenHittable, typeAndDismissKeyboard, etc.) but keep it simple and conventional (no heavy DSL unless asked).

References in this skill folder

• references/authoring-contract.md — how to interpret user input and output format
• references/locator-strategy.md — detailed selector rules, tables, cells, dynamic content
• references/flake-playbook.md — flake patterns and fixes
• references/examples.md — complete end-to-end examples

Minimal example (pattern)

User scenario:

“Open app, login, verify home title.”

Agent output:

1. Test plan steps + assertions
2. LoginScreen, HomeScreen
3. LoginFlowTests
4. identifier list: login.email, login.password, login.submit, home.title

Enforcement checklist (before final answer)

• No sleep()
• Accessibility IDs first
• Anchors used for screen load
• Assertions after critical steps
• Scrolling bounded
• Failure diagnostics included
• Output includes A–E sections (contract)
• App-side identifier implementation code included
• Pre-flight verification steps communicated to user

Dependency Verification

Before marking test generation complete:

Failure Prevention

Root Cause of Most Failures: Gap between "identifiers listed" and "identifiers implemented"

Prevention:

1. Always generate app-side implementation code (Section F)
2. Always update the ViewController file directly
3. Always verify app builds before claiming tests are ready