Executes a shell command via ADB.

Check if current model is an Anthropic model.

Check if current model is a Meta/Llama model.

Format action history for LLM context.

Build request body based on model type.

Navigate back.

Mark task as complete and exit.

Press the Enter key.

Navigate to home screen.

Execute a swipe action in the specified direction.

Execute a tap action at specified coordinates.

Execute a type action to input text.

Wait for UI to load.

Extract text response based on model type.

Parse JSON from LLM response, handling extra text.

Executes the action decided by the LLM.

Get the next action decision from the LLM.

Factory function to get the appropriate LLM provider.

Get the model name based on the selected provider.

Dumps the current UI XML and returns the sanitized JSON string.

Entry point for the Android Action Kernel.

Main agent loop: Perceive -> Reason -> Act Args: goal: The task to accomplish max_steps: Maximum steps before stopping (defa

Validate that required configuration is present.

Parses Android Accessibility XML and returns a lean list of interactive elements. Calculates center coordinates (x, y) for every clickable el