SKILL.md
Maestro Mobile E2E Testing
Overview
Maestro is a declarative YAML-based mobile E2E testing framework. It provides automatic waiting, built-in retry logic, and fast execution without boilerplate. It's more stable than Detox or Appium for React Native apps.
Key Features
- Declarative YAML — no imperative test code, just steps
- Automatic waiting — no manual
sleep()or flaky waits
- Built-in retry — reduces test flakiness
- Fast execution — runs quickly without setup overhead
- Maestro Studio — interactive test builder (
maestro studio)
- Sub-flows — reusable YAML sequences for DRY tests
- JavaScript scripting — GraalJS runtime for HTTP calls and data manipulation
- Maestro Cloud — real device testing in CI without local simulators
Quick Start
Install
curl -Ls "https://get.maestro.mobile.dev" | bash
brew install openjdk@17
export JAVA_HOME=/opt/homebrew/opt/openjdk@17/libexec/openjdk.jdk/Contents/Home
Minimal test
appId: com.myapp
---
- launchApp
- tapOn:
id: "my-button"
- assertVisible: "Expected Text"
Run
maestro test .maestro/smoke-test.yaml
maestro test --debug .maestro/smoke-test.yaml # step through
maestro studio # interactive builder
Core Patterns
1. Selector Strategy: testID vs Text
Choose your selector approach based on project context. Both are valid — the right choice depends on whether your app is localized and your team's testing philosophy.
Context
Recommended Selector
Rationale
Multi-language / i18n
id: (testID)
Stable across translations
Single language
Text labels
Human-readable, self-documenting tests
Agent-maintained tests
Either — ask the developer
Readability matters less for AI-maintained flows
System dialogs
Text (always)
No testID possible on native alerts
# testID selector — stable across translations
- tapOn:
id: "submit-button"
# Text selector — human-readable, self-documenting
- tapOn: "Submit"
When to prefer testIDs:
- App supports multiple languages or will be translated
- UI text is dynamic or frequently changes
- Multiple elements share the same visible text
When to prefer text selectors:
- Single-language app with stable copy
- Readability and self-documentation are a priority
- Testing user-visible behavior exactly as it appears
In React Native, add testID props when using ID-based selectors:
<TouchableOpacity testID="submit-button" onPress={handleSubmit}>
<Text>{t('submit')}</Text>
</TouchableOpacity>
testID Naming Convention
When using ID-based selectors:
{component}-{action/type}[-{variant}]
Examples:
- auth-prompt-login-button
- product-card-{id}
- otp-input-0
- tab-home
- dashboard-loading
2. Auth Pre-Flight Pattern
Prevent race conditions where Maestro interacts with the UI before auth state resolves. Add a zero-size auth-loaded marker that only renders when auth loading completes:
// In your tab bar or root layout
{!isLoading && <View testID="auth-loaded" style={{ width: 0, height: 0 }} />}
Then in every test:
- launchApp
# Prevent XCTest crash on cold boot (iOS)
- swipe:
direction: DOWN
duration: 100
# Wait for auth state to resolve
- extendedWaitUntil:
visible:
id: "auth-loaded"
timeout: 15000
# Now safe to interact
- tapOn:
id: "tab-home"
3. Adaptive Tests (Handle Both Auth States)
Tests should work regardless of whether the user is authenticated:
# Auth flow — only runs if login prompt is visible
- runFlow:
when:
visible: "Sign In"
file: flows/auth-flow.yaml
# Already authenticated — proceed directly
- runFlow:
when:
visible:
id: "tab-home"
file: flows/authenticated-action.yaml
4. Testing Optimistic Updates
Use short timeouts to verify UI changes happen before server response:
# Trigger mutation
- tapOn:
id: "action-button"
# OPTIMISTIC: UI must change within 3s (not waiting for server)
- extendedWaitUntil:
visible:
id: "undo-button"
timeout: 3000
# Verify derived UI state
- extendedWaitUntil:
visible:
id: "user-indicator"
timeout: 5000
Action
Expected Change
Timeout
Mutation trigger
Button state flips
< 3s
List update
Item appears/disappears
< 5s
Re-do action
Proves persistence
< 3s
5. Dismissing Native Alerts
React Native Alert.alert() creates native dialogs that block the UI:
- tapOn:
id: "action-button"
# Wait for expected state change first
- extendedWaitUntil:
visible:
id: "new-state-element"
timeout: 5000
# Dismiss alert (optional in case it already closed)
- tapOn:
text: "OK"
optional: true
# Brief delay for alert animation
- swipe:
direction: DOWN
duration: 300
6. Sub-Flows for Reusability
Break repeated sequences into sub-flow files:
.maestro/
├── flows/
│ ├── auth-and-return.yaml
│ ├── complete-purchase.yaml
│ └── verify-result.yaml
├── smoke-test.yaml
└── feature-test.yaml
# In main test
- runFlow:
file: flows/auth-and-return.yaml
7. Deep Links (Expo)
Use the Expo scheme from app.json, not the bundle ID:
# WRONG
- openLink: "com.myapp://profile/settings"
# CORRECT
- openLink: "myapp://profile/settings"
Deep links must be registered in your app's deep link handler. Unregistered routes silently fail.
8. Platform-Specific Logic
- runFlow:
when:
platform: ios
file: flows/ios-specific.yaml
- runFlow:
when:
platform: android
file: flows/android-specific.yaml
9. Environment Variables
appId: com.myapp
env:
TEST_EMAIL: maestro-test@example.com
API_BASE_URL: http://localhost:3000
---
- inputText: ${TEST_EMAIL}
10. Selector State Properties
Use enabled, selected, checked, and focused to target elements by their current state. This is useful for validating interactive element states before or after actions.
# Only tap the submit button if it's enabled
- tapOn:
id: "submit-button"
enabled: true
# Assert a checkbox is checked
- assertVisible:
id: "terms-checkbox"
checked: true
# Wait for an input to be focused
- extendedWaitUntil:
visible:
id: "email-input"
focused: true
timeout: 3000
Property
Values
Use Case
enabled
true / false
Buttons that disable during submission or until form is valid
checked
true / false
Checkboxes, toggle switches
selected
true / false
Tab items, segmented controls
focused
true / false
Input fields with auto-focus
11. Relative Position Selectors
Distinguish between similar elements by their spatial relationship to other elements. This is more idiomatic and resilient than index-based selection.
# BAD — fragile, breaks if order changes
- tapOn:
text: "Add to Basket"
index: 1
# GOOD — contextual, self-documenting
- tapOn:
text: "Add to Basket"
below:
text: "Awesome Shoes"
Available relative selectors:
# Target element below another
- tapOn:
text: "Buy Now"
below: "Product Title"
# Target element that is a child of a parent
- tapOn:
text: "Delete"
childOf:
id: "item-card-42"
# Target a parent that contains a specific child
- tapOn:
containsChild: "Urgent"
# Target by multiple descendants
- tapOn:
containsDescendants:
- id: title_id
text: "Specific Title"
- "Another descendant text"
# Horizontal positioning
- tapOn:
text: "Edit"
rightOf: "Username"
Selector
Meaning
below:
Element is positioned below the referenced element
above:
Element is positioned above the referenced element
leftOf:
Element is to the left of the referenced element
rightOf:
Element is to the right of the referenced element
childOf:
Element is a direct child of the referenced parent
containsChild:
Element contains a direct child matching the reference
containsDescendants:
Element contains all specified descendant elements
Authentication Testing
Architecture
Testing OTP or magic-link authentication in E2E requires capturing emails programmatically. The general pattern:
┌─────────────┐ ┌──────────────┐ ┌─────────────────┐
│ Maestro │────▶│ Auth │────▶│ Email Capture │
│ Test │ │ Provider │ │ Service │
└─────────────┘ └──────────────┘ └─────────────────┘
│ │
│ ┌──────────────────────────────┘
│ ▼
│ ┌─────────────┐
└──▶│ REST API │ ─── GET /api/v1/messages
│ (email) │ ─── Extract OTP code
└─────────────┘
Common email capture services: Mailpit, MailHog, Ethereal.
OTP Fetch Script
Fetch OTP codes from your email capture service using Maestro's GraalJS runtime:
// CRITICAL: Maestro uses GraalJS — NO async/await, NO fetch()
var email = typeof EMAIL !== "undefined" ? EMAIL : "test@example.com";
var emailServiceUrl = typeof EMAIL_SERVICE_URL !== "undefined"
? EMAIL_SERVICE_URL : "http://localhost:8025";
var response = http.get(emailServiceUrl + "/api/v1/messages");
if (!response.ok) {
throw new Error("Failed to fetch emails: " + response.status);
}
var data = json(response.body);
// Find the latest email and extract OTP code
var body = data.messages[0].Content.Body;
var match = body.match(/(\d{6})/);
output.OTP_CODE = match[1];
OTP Input Strategy
OTP components with auto-focus need individual digit entry. Tap each input before typing:
# Split OTP into digits via helper script
- runScript:
file: scripts/split-otp.js
env:
OTP_CODE: ${output.OTP_CODE}
# Enter each digit by tapping its input
- tapOn:
id: "otp-input-0"
- inputText: ${output.OTP_0}
- tapOn:
id: "otp-input-1"
- inputText: ${output.OTP_1}
# ... repeat for all digits
For provider-specific implementations (Supabase + Mailpit, Firebase Auth, Auth0), create a project-level skill that extends this one.
GraalJS Script Rules
Maestro uses the GraalJS runtime. These constraints are non-negotiable:
Feature
Status
async/await
NOT supported
fetch()
NOT supported
http.get(), http.post()
Use these instead
json()
Use to parse response bodies
output.VAR
Set variables for use in YAML flow
var declarations
Required (use var, not const/let for safety)
// Script template
var response = http.get("http://localhost:8025/api/endpoint");
if (!response.ok) {
throw new Error("Request failed: " + response.status);
}
var data = json(response.body);
output.RESULT = data.value;
Critical Gotchas
clearState Does NOT Clear iOS Keychain
clearState: true clears the app sandbox (UserDefaults, files, caches) but does NOT clear the iOS Keychain. Auth tokens stored via expo-secure-store (or any Keychain-based storage) persist across clearState resets and even app reinstalls.
# WRONG — user may still be authenticated
- launchApp:
clearState: true
- assertVisible: "Welcome" # Fails if Keychain has tokens
# CORRECT — wait for auth resolution, then adapt
- launchApp
- extendedWaitUntil:
visible:
id: "auth-loaded"
timeout: 15000
Rules:
- Never rely on
clearStateto produce guest state on iOS
- For auth tests: skip
clearState, useauth-loadedpre-flight
- For guest tests: use adaptive flows that handle both states
- Never assert guest-only UI after
clearState
Note: On Android, clearState: true fully resets app data including credentials. This is an iOS-only gotcha.
XCTest kAXErrorInvalidUIElement Crash (iOS)
The XCTest driver may crash if Maestro interacts with the accessibility tree before the first render cycle completes on cold boot.
Fix: Add a no-op swipe immediately after launchApp:
- launchApp
- swipe:
direction: DOWN
duration: 100
API Server Dependency
Mobile apps calling backend APIs on localhost need either the full server or a mock server running. Without it, all API-dependent screens show loading spinners or empty states (queries fail silently).
Fix: Start a mock API server before running Maestro tests:
# Start mock server (serves canned responses on your API port)
npx tsx scripts/mock-api-server.ts &
# Then run tests
maestro test .maestro/my-test.yaml
Create a lightweight mock that returns canned JSON for each endpoint your app calls. This is faster and more deterministic than running your full backend.
Auth-Aware Tab Bars
Tab bars that show different tabs for guest vs authenticated users will cause selector failures:
State
Typical Tabs
Guest
home, search, cart, profile
Auth
home, feed, create, messages, profile
Only assert tabs that exist in both states, or use adaptive when: conditions.
Test File Template
# {Feature} {Action} Test
#
# Tests: {what this validates}
# Prerequisites:
# - Simulator/emulator running with app installed
# - Backend or mock server running (if API-dependent)
appId: com.myapp
env:
TEST_EMAIL: maestro-{feature}@example.com
EMAIL_SERVICE_URL: http://localhost:8025
---
# ==========================================
# STEP 1: LAUNCH + AUTH PRE-FLIGHT
# ==========================================
- launchApp
- swipe:
direction: DOWN
duration: 100
- extendedWaitUntil:
visible:
id: "auth-loaded"
timeout: 15000
- takeScreenshot: 01-initial-state
# ==========================================
# STEP 2: {ACTION}
# ==========================================
- tapOn:
id: "target-element"
# ==========================================
# STEP 3: VERIFY
# ==========================================
- extendedWaitUntil:
visible:
id: "expected-result"
timeout: 5000
- takeScreenshot: 02-final-state
Folder Structure
.maestro/
├── README.md # Quick reference + testID inventory
├── config.yaml # Shared configuration
├── flows/ # Reusable sub-flows
│ ├── auth-and-return.yaml
│ ├── complete-action.yaml
│ └── verify-result.yaml
├── scripts/ # GraalJS helpers
│ ├── fetch-otp.js
│ └── split-otp.js
├── smoke-test.yaml # Guest navigation
├── auth-signin.yaml # OTP sign-in flow
├── feature-screenshots.yaml # Screenshot capture flows
└── feature-action.yaml # Feature-specific tests
scripts/
├── mock-api-server.ts # Lightweight mock for E2E
└── run-e2e.sh # Orchestration script
Naming Conventions
Type
Pattern
Example
Main test
{feature}-{action}.yaml
checkout-purchase.yaml
Sub-flow
{action}-{context}.yaml
auth-and-return-to-dashboard.yaml
Script
{verb}-{noun}.js
fetch-otp.js
Infrastructure
Mock Server for API-Dependent Tests
Create a lightweight mock server that serves canned responses for your API layer. This is faster and more deterministic than running your full backend during E2E tests.
# Start mock server before running Maestro tests
npx tsx scripts/mock-api-server.ts &
Orchestration Script
Automate the full E2E setup with a shell script that:
- Starts backend services (database, auth)
- Seeds test data
- Starts mock API server
- Runs Maestro tests
- Cleans up all services
bash scripts/run-e2e.sh
Seed Data
Tests that depend on specific data require seeded databases. Keep seed scripts alongside your test infrastructure and run them before each test suite.
Android-Specific Patterns
Emulator Setup
Android tests require an emulator or a USB-connected physical device. Maestro auto-detects connected devices.
# List available system images
sdkmanager --list | grep system-images
# Create emulator
avdmanager create avd -n maestro_test \
-k "system-images;android-34;google_apis;arm64-v8a"
# Start emulator
emulator -avd maestro_test
iOS vs Android Differences
Aspect
iOS
Android
Device type
Simulator only (no physical)
Emulator + physical via ADB
clearState
Does NOT clear Keychain
Fully resets app data
Cold boot crash
XCTest kAXError (add swipe delay)
No equivalent issue
Performance
Runs natively (fast)
ARM emulation (slower on x86)
Permission dialogs
System alerts
System dialogs with different text
ADB Debugging
adb devices # List connected devices
adb shell am start -n com.myapp/.MainActivity # Launch app
adb logcat | grep Maestro # Filter Maestro logs
adb shell input keyevent 82 # Unlock screen
Android Permission Handling
Android permissions appear as system dialogs. Dismiss with optional taps:
- tapOn:
text: "Allow"
optional: true
- tapOn:
text: "While using the app"
optional: true
CI/CD Integration
GitHub Actions with Maestro Cloud
Maestro Cloud provides real devices in CI without local simulators. Use the official action:
name: Mobile E2E Tests
on:
push:
branches: [main]
pull_request:
branches: [main]
jobs:
maestro-e2e:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
# Build Android APK
- uses: actions/setup-java@v4
with:
java-version: '17'
distribution: 'temurin'
- name: Build Android APK
run: |
cd apps/mobile
npx expo prebuild --platform android --no-install
cd android && ./gradlew assembleRelease
- name: Run Maestro Cloud Tests
id: maestro
uses: mobile-dev-inc/action-maestro-cloud@v2
with:
api-key: ${{ secrets.MAESTRO_API_KEY }}
app-file: apps/mobile/android/app/build/outputs/apk/release/app-release.apk
workspace: .maestro
include-tags: ci
# Access results
# ${{ steps.maestro.outputs.MAESTRO_CLOUD_CONSOLE_URL }}
# ${{ steps.maestro.outputs.MAESTRO_CLOUD_UPLOAD_STATUS }}
# ${{ steps.maestro.outputs.MAESTRO_CLOUD_FLOW_RESULTS }}
Tag-Based Flow Filtering
Use tags to control which tests run in CI vs locally:
# In your flow file header
appId: com.myapp
tags:
- ci
- smoke
---
- launchApp
# ... test steps
# Run only CI-tagged flows locally
maestro test --include-tags ci .maestro/
# Exclude work-in-progress flows
maestro test --exclude-tags wip .maestro/
Local CI with Docker (Android Only)
FROM openjdk:17-slim
RUN curl -Ls "https://get.maestro.mobile.dev" | bash
ENV PATH="/root/.maestro/bin:${PATH}"
COPY .maestro/ /app/.maestro/
WORKDIR /app
CMD ["maestro", "test", ".maestro/"]
Note: iOS tests cannot run in Docker (requires macOS). Use Maestro Cloud for iOS in CI.
Maestro Cloud
Maestro Cloud runs tests on real devices without local simulator setup.
Setup
- Create account at cloud.maestro.dev
- Generate API key from dashboard
- Store as
MAESTRO_API_KEYsecret in your CI provider
Running from CLI
# Upload and run on Maestro Cloud
maestro cloud --api-key $MAESTRO_API_KEY \
--app-file ./app-release.apk \
.maestro/
# With tag filtering
maestro cloud --api-key $MAESTRO_API_KEY \
--app-file ./app-release.apk \
--include-tags smoke \
.maestro/
Key Points
- iOS testing: Supported on Maestro Cloud (not on local physical devices)
- Android testing: Both local physical devices and Maestro Cloud
- Results: Dashboard with video recordings, logs, and screenshots
- CI outputs:
MAESTRO_CLOUD_CONSOLE_URL,MAESTRO_CLOUD_FLOW_RESULTS
Maestro MCP Server
The Maestro MCP server exposes Maestro's full command set as Model Context Protocol tools, letting AI agents execute tests and interact with devices directly — not just write YAML.
How It Complements This Skill
This Skill
Maestro MCP
Role
Teaches correct patterns
Provides runtime execution
Layer
Authoring (write good YAML)
Execution (run, tap, assert, screenshot)
Output
Better test files
Live device interaction
Use both together: this skill ensures the AI writes correct tests; the MCP lets it run them immediately, see failures, and iterate.
Setup
The MCP ships with the Maestro CLI — no extra install needed:
# Verify it's available
maestro mcp
Claude Code — add to project .mcp.json or global settings:
{
"mcpServers": {
"maestro": {
"command": "maestro",
"args": ["mcp"]
}
}
}
Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"maestro": {
"command": "maestro",
"args": ["mcp"]
}
}
}
Also supported on Cursor, Windsurf, VS Code, and JetBrains IDEs. See the Maestro MCP docs for IDE-specific setup.
Key Capabilities
The MCP server exposes 47 tools organized by category:
Category
Tools
Examples
UI Interaction
tap, swipe, scroll, long press
tapOn, scrollUntilVisible
Text Input
type, erase, paste, copy
inputText, eraseText
Assertions
visibility, AI-powered
assertVisible, assertWithAI
App Lifecycle
launch, stop, clear state
launchApp, clearState
Device Control
location, orientation, airplane
setLocation, hideKeyboard
Flow Control
run flows, repeat, eval scripts
runFlow, evalScript
Media
screenshots, recording
takeScreenshot, startRecording
AI-Powered
visual assertions, defect detection
assertWithAI, assertNoDefectsWithAI
Write-Run-Fix Loop
With both this skill and the MCP active, the AI can:
- Write a test YAML using patterns from this skill
- Run it via MCP's
runFlow/launchApp+ interaction tools
- See failures via screenshots and assertions
- Fix the YAML and re-run — all in one conversation
Debugging
maestro test --debug .maestro/test.yaml # Step through interactively
maestro record .maestro/test.yaml # Record as video
maestro studio # Interactive UI builder
maestro hierarchy # View element tree
Screenshots saved to ~/.maestro/tests/{timestamp}/.
Checklist for New Tests
[ ] Unique test email (maestro-{feature}@example.com)
[ ] Selector strategy chosen (testID for i18n apps, text for single-language — see Pattern 1)
[ ] Selectors use state properties where relevant (enabled, checked — see Pattern 10)
[ ] Similar elements distinguished with relative selectors, not index (see Pattern 11)
[ ] Auth pre-flight pattern used (auth-loaded)
[ ] Post-launch swipe added (iOS crash prevention)
[ ] Both auth states handled (adaptive flows)
[ ] Native alerts dismissed after mutations
[ ] Short timeouts for optimistic updates (3-5s)
[ ] Sub-flows created for reusable sequences
[ ] Descriptive screenshots at key points
[ ] Header comment with prerequisites
[ ] Added to README.md test table
[ ] Mock API server started if backend-dependent
[ ] Tags added for CI filtering (ci, smoke, wip)
Common Errors
Error
Cause
Fix
"Unable to locate Java Runtime"
Java not in PATH
export JAVA_HOME=/opt/homebrew/opt/openjdk@17/...
"Element not found" after tap
Native alert blocking
Add tapOn: text: "OK" optional: true
OTP digits not entering
Auto-focus interference
Use individual otp-input-N testIDs
Test passes but nothing happened
optional: true misused
Only use optional for truly optional actions
"Assertion is false" on visibility
Element not rendered yet
Increase timeout or verify testID exists
Script output empty
Wrong JS API
Use http.get() not fetch()
Auth state inconsistent after clearState
iOS Keychain not cleared
Don't use clearState, use adaptive flows
kAXErrorInvalidUIElement crash
Cold boot race (iOS)
Add post-launch swipe delay
Loading spinners / empty screens
No API server running
Start mock API server before tests
Permission dialog blocking (Android)
System dialog not dismissed
Add tapOn: text: "Allow" optional: true