some chanegs in the system prompt for checking on the scrollable elements

Author: Jeomon

src/agent/web/__init__.py

@@ -130,6 +130,7 @@ async def action(self,state:AgentState):
             'informative_elements':browser_state.dom_state.informative_elements_to_string(),
             'scrollable_elements':browser_state.dom_state.scrollable_elements_to_string()
         })
+        print(browser_state.dom_state.scrollable_nodes)
         messages=[AIMessage(action_prompt),ImageMessage(text=observation_prompt,image_obj=image_obj) if self.use_vision and image_obj is not None else HumanMessage(observation_prompt)]
         return {**state,'messages':messages,'prev_observation':observation}
 

src/agent/web/dom/views.py

@@ -80,7 +80,7 @@ def interactive_elements_to_string(self)->str:
         return '\n'.join([f'{index} - Tag: {node.tag} Role: {node.role} Name: {node.name} Attributes: {node.attributes} Cordinates: {node.center.to_string()}' for index,(node) in enumerate(self.interactive_nodes)])
 
     def informative_elements_to_string(self)->str:
-        return  '\n'.join([f'Tag: {node.tag} Role: {node.role} Content: {node.content} Cordinates: {node.center.to_string()}' for node in self.informative_nodes])
+        return  '\n'.join([f'Tag: {node.tag} Role: {node.role} Content: {node.content}' for node in self.informative_nodes])
 
     def scrollable_elements_to_string(self)->str:
         n=len(self.interactive_nodes)

src/agent/web/prompt/system.md

@@ -55,6 +55,14 @@ You are a highly advanced, expert-level Web Agent with human-like browsing capab
    ```
    <element_index> - Tag: <element_tag> Role: <element_role> Name: <element_name> Attributes: <element_attributes> Coordinates: <element_coordinate>
    ```
+6. **Scrollable Elements:** Available webpage elements in format:
+   ```
+   <element_index> - Tag: <element_tag> Role: <element_role> Name: <element_name> Attributes: <element_attributes>
+   ```
+7. **Informative Elements:** Available webpage elements in format:
+   ```
+   Tag: <element_tag> Role: <element_role> Content: <element_content>
+   ```
 
 ## Execution Framework:
 
@@ -63,6 +71,7 @@ You are a highly advanced, expert-level Web Agent with human-like browsing capab
 - Thoroughly analyze element properties before interaction
 - Reference elements exclusively by their numeric index
 - Consider element position and visibility when planning interactions
+- Use scrollable elements to scroll through specific sections of the webpage
 
 ### Visual Analysis Protocol:
 
@@ -85,7 +94,7 @@ You are a highly advanced, expert-level Web Agent with human-like browsing capab
 - Wait for complete page loading before proceeding with interactions
 
 ### Human in Loop:
-- Human in the loop enabled: {human_in_loop}
+- Human in the loop is enabled: {human_in_loop}
 - Talk to user when needed using `Human Tool` if its available
 - Use it for clarification, credentials, for performing a new task within the main task and more
 - When unsolvable security challenges comes in the way during navigation
@@ -133,4 +142,4 @@ Respond exclusively in this XML format:
   <Action-Name>[Selected tool name]</Action-Name>
   <Action-Input>{{'param1':'value1','param2':'value2'}}</Action-Input>
 </Option>
-```
+```