handle search bar for custom selection

skyvern/forge/prompts/skyvern/parse-input-or-select-context.j2

@@ -7,6 +7,7 @@ Reply in the following JSON format:
     "thought": str, // A string to describe how you double-check the information to ensure the accuracy.
     "field": str, // Which field is this action intended to fill out?
     "is_required": bool, // True if this is a required field, otherwise false.
+    "is_search_bar": bool, // True if the element to take the action is a search bar, otherwise false.
 }
 
 Existing reasoning context:

skyvern/webeye/actions/actions.py

@@ -46,9 +46,10 @@ def __repr__(self) -> str:
 class InputOrSelectContext(BaseModel):
     field: str | None = None
     is_required: bool | None = None
+    is_search_bar: bool | None = None  # don't trigger custom-selection logic when it's a search bar
 
     def __repr__(self) -> str:
-        return f"InputOrSelectContext(field={self.field}, is_required={self.is_required})"
+        return f"InputOrSelectContext(field={self.field}, is_required={self.is_required}, is_search_bar={self.is_search_bar})"
 
 
 class Action(BaseModel):

skyvern/webeye/actions/handler.py

@@ -466,7 +466,7 @@ async def handle_input_text_action(
                 if result is not None:
                     return [result]
                 LOG.info(
-                    "No dropdown menu detected, indicating it couldn't be a selectable auto-completion input",
+                    "It might not be a selectable auto-completion input, exit the custom selection mode",
                     task_id=task.task_id,
                     step_id=step.step_id,
                     element_id=skyvern_element.get_id(),
@@ -508,16 +508,16 @@ async def handle_input_text_action(
             return [ActionSuccess()]
 
         if await skyvern_element.is_auto_completion_input():
-            result = await input_or_auto_complete_input(
+            if result := await input_or_auto_complete_input(
                 action=action,
                 page=page,
                 dom=dom,
                 text=text,
                 skyvern_element=skyvern_element,
                 step=step,
                 task=task,
-            )
-            return [result]
+            ):
+                return [result]
 
         await skyvern_element.input_sequentially(text=text)
         return [ActionSuccess()]
@@ -1198,7 +1198,7 @@ async def input_or_auto_complete_input(
     skyvern_element: SkyvernElement,
     step: Step,
     task: Task,
-) -> ActionResult:
+) -> ActionResult | None:
     LOG.info(
         "Trigger auto completion",
         task_id=task.task_id,
@@ -1257,6 +1257,15 @@ async def input_or_auto_complete_input(
         if isinstance(result.action_result, ActionSuccess):
             return ActionSuccess()
 
+        if input_or_select_context.is_search_bar:
+            LOG.info(
+                "Stop generating potential values for the auto-completion since it's a search bar",
+                context=input_or_select_context,
+                step_id=step.step_id,
+                task_id=task.task_id,
+            )
+            return None
+
         tried_values.append(current_value)
         whole_new_elements.extend(result.incremental_elements)
 
@@ -1373,6 +1382,15 @@ async def sequentially_select_from_dropdown(
         step_id=step.step_id,
     )
 
+    if not force_select and input_or_select_context.is_search_bar:
+        LOG.info(
+            "Exit custom selection mode since it's a non-force search bar",
+            context=input_or_select_context,
+            task_id=task.task_id,
+            step_id=step.step_id,
+        )
+        return None, None
+
     # TODO: only suport the third-level dropdown selection now
     MAX_SELECT_DEPTH = 3
     values: list[str | None] = []