midscene-android-automation
от quanru
Web & Frontend DevelopmentКратко
Vision-driven Android device automation using Midscene.
Описание
Что делает этот навык
Controls Android devices using screenshot-based vision, with no DOM access or accessibility labels required. Accepts natural language commands to tap, swipe, type, launch apps, and capture screenshots. Works across any Android app regardless of the underlying framework.
Unlike Appium or UI Automator, it requires no accessibility labels or view hierarchy access, so it works on any app including games and fully custom-rendered UIs.
Когда его использовать
- Automating repetitive tasks in mobile apps that have no scriptable API
- Extracting data from Android apps without a web or API equivalent
- Testing a mobile UI without writing platform-specific test code
- Running multi-step workflows in messaging or social apps
- Capturing screenshots of specific app states for documentation or bug reports
Пример сценария
User asks: 'Open Settings, go to Wi-Fi, and tell me the connected network name'
- 1
Connect to the Android device via ADB
- 2
Launch the Settings app using adb shell am start and take a screenshot to confirm it is in the foreground
- 3
Run act with the prompt 'navigate to Wi-Fi settings and read the connected network name'
- 4
Take a screenshot to verify the current screen state
- 5
Disconnect and report the network name extracted from the screen
Connected Wi-Fi network name read from the device screen and reported to the user
Требования
- MIDSCENE_MODEL_API_KEY environment variable from a supported AI vision model provider
- MIDSCENE_MODEL_NAME, MIDSCENE_MODEL_BASE_URL, and MIDSCENE_MODEL_FAMILY environment variables
- Android SDK Platform Tools (provides adb)
- Android device with USB debugging enabled in Developer Options
Установка
clawhub install quanru/midscene-android-automationВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.