agent.yml

agent: >
  You are an helpful assistant with access to an Android phone and using the tools to perform tasks with it.
  
  ## Guidance

  - Use the provided tools to see the device screen and interact with it.
  - Interact with the device by performing any action necessary to perform the task.
  - For every action, make sure it's been completed successfully, for instance if the request if sending a message, make sure the message has been sent successfully.

using:
  - reasoning
  - task

tools:
  - name: get_view_state
    description: "Get a description of what's visible on the screen and how to interact with each element."
    # uses uiautomator dump to get an XML with all the elements on the screen
    tool: "adb shell uiautomator dump --compressed /sdcard/window_dump.xml > /dev/null 2>&1 && 
      adb pull /sdcard/window_dump.xml /tmp/window_dump.xml > /dev/null 2>&1 && 
      cat /tmp/window_dump.xml"

  - name: take_screenshot
    description: "Take a screenshot of the device screen."
    # uses adb to take a screenshot of the device screen
    mime: image/png
    tool: adb exec-out screencap -p

  - name: adb_shell
    description: "Execute an adb shell command on the Android device and return the output."
    arguments:
      - name: command
        description: "The command to execute on the Android device."
        example: "input tap 900 2140"
    # uses adb to execute a shell command on the device
    tool: adb shell {{ command }}