Saltar al contenido
TakoAPI
Volver a skills

midscene-ios-automation

por quanru

Web & Frontend Development
853Descargas según clawskills.sh
0Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

Vision-driven iOS device automation using Midscene CLI.

Descripción

Qué hace esta skill

Vision-driven iOS automation that operates entirely from screenshots, requiring no DOM access or accessibility labels. Sends natural language commands to any iOS app via WebDriverAgent and a vision-capable LLM. Works with apps built on any technology stack.

Works on any iOS app regardless of how it was built, since it reads the screen visually instead of relying on accessibility labels or DOM structure.

Cuándo usarla

  • Automating repetitive UI tasks in iOS apps during QA testing
  • Extracting data from iOS apps that lack a web or API interface
  • Navigating iOS settings programmatically for device configuration
  • Verifying iOS app UI flows without writing XCTest or UITest code
  • Capturing screen state during multi-step iOS app walkthroughs

Flujo de trabajo de ejemplo

ENTRADA

User asks: Open Settings, go to Wi-Fi, and tell me the name of the connected network

AGENTE
  1. 1

    Connect to the iOS device via WebDriverAgent

  2. 2

    Take a screenshot to confirm the current screen state

  3. 3

    Run act with prompt "open Settings and navigate to Wi-Fi, tell me the connected network name"

  4. 4

    Take a screenshot to capture the Wi-Fi screen

  5. 5

    Disconnect from the device and report the network name found

SALIDA

Connected Wi-Fi network name extracted from the Settings > Wi-Fi screen

Requisitos

  • MIDSCENE_MODEL_API_KEY environment variable (API key for a vision-capable LLM provider)
  • MIDSCENE_MODEL_NAME, MIDSCENE_MODEL_BASE_URL, and MIDSCENE_MODEL_FAMILY environment variables
  • WebDriverAgent installed and running on the target iOS device

Instalación

clawhub install quanru/midscene-ios-automation

Ejecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.

Enlaces