SDK: Support AI act feature (#3888)

skyvern/forge/sdk/routes/sdk.py

@@ -174,6 +174,11 @@ async def run_sdk_action(
                 data=action.data,
                 timeout=action.timeout,
             )
+        elif action.type == "ai_act":
+            await page_ai.ai_act(
+                prompt=action.intention,
+            )
+            result = None
         elif action.type == "extract":
             extract_result = await page_ai.ai_extract(
                 prompt=action.prompt,

skyvern/forge/sdk/schemas/sdk_actions.py

@@ -12,6 +12,7 @@ class SdkActionType(str, Enum):
     AI_CLICK = "ai_click"
     AI_INPUT_TEXT = "ai_input_text"
     AI_SELECT_OPTION = "ai_select_option"
+    AI_ACT = "ai_act"
     EXTRACT = "extract"
 
 
@@ -57,6 +58,14 @@ class SelectOptionAction(SdkActionBase):
     timeout: float = Field(default=settings.BROWSER_ACTION_TIMEOUT_MS, description="Timeout in milliseconds")
 
 
+class ActAction(SdkActionBase):
+    """AI act action parameters."""
+
+    type: Literal["ai_act"] = "ai_act"
+    intention: str = Field(default="", description="Natural language prompt for the action")
+    data: str | dict[str, Any] | None = Field(None, description="Additional context data")
+
+
 class ExtractAction(SdkActionBase):
     """Extract data action parameters."""
 
@@ -70,7 +79,7 @@ class ExtractAction(SdkActionBase):
 
 # Discriminated union of all action types
 SdkAction = Annotated[
-    Union[ClickAction, InputTextAction, SelectOptionAction, ExtractAction],
+    Union[ClickAction, InputTextAction, SelectOptionAction, ActAction, ExtractAction],
     Field(discriminator="type"),
 ]