[
{
"stage_index": 1,
"name": "DOM State Ingestion & Tree Extraction",
"action_type": "inspection",
"description": "Extract interactive DOM elements, build indexed accessibility tree, capture viewport screenshot."
},
{
"stage_index": 2,
"name": "Action Synthesis (Click, Type, Scroll)",
"action_type": "code_modification",
"description": "Model selects element index and outputs structured action (click_element, input_text, scroll_page)."
},
{
"stage_index": 3,
"name": "Playwright Action Execution",
"action_type": "dispatch",
"description": "Execute action on browser page via CDP or Playwright locator with network idle wait."
},
{
"stage_index": 4,
"name": "URL / DOM State Change Verification",
"action_type": "output_validation",
"description": "Verify page navigation or DOM mutation occurred as a result of action."
},
{
"stage_index": 5,
"name": "Goal Completion Check",
"action_type": "stop_evaluation",
"description": "Evaluate if goal statement is satisfied or if next step is required."
}
]