fix task v2 block goto url issue (#1899)

skyvern/forge/sdk/services/task_v2_service.py

@@ -6,6 +6,7 @@ from typing import Any
 
 import httpx
 import structlog
+from playwright.async_api import Page
 from sqlalchemy.exc import OperationalError
 
 from skyvern.config import settings
@@ -411,7 +412,15 @@ async def run_task_v2_helper(
         task_history_record: dict[str, Any] = {}
         context = skyvern_context.ensure_context()
 
-        if i == 0:
+        current_url: str | None = None
+        page: Page | None = None
+        browser_state = app.BROWSER_MANAGER.get_for_workflow_run(workflow_run_id, workflow_run.parent_workflow_run_id)
+        if browser_state:
+            page = await browser_state.get_working_page()
+            if page:
+                current_url = await SkyvernFrame.get_url(page)
+
+        if i == 0 and current_url != url:
             # The first iteration is always a GOTO_URL task
             task_type = "goto_url"
             plan = f"Go to this website: {url}"
@@ -422,11 +431,12 @@ async def run_task_v2_helper(
             )
         else:
             try:
-                browser_state = await app.BROWSER_MANAGER.get_or_create_for_workflow_run(
+                if browser_state is None:
-                    workflow_run=workflow_run,
+                    browser_state = await app.BROWSER_MANAGER.get_or_create_for_workflow_run(
-                    url=url,
+                        workflow_run=workflow_run,
-                    browser_session_id=browser_session_id,
+                        url=url,
-                )
+                        browser_session_id=browser_session_id,
+                    )
                 scraped_page = await scrape_website(
                     browser_state,
                     url,
@@ -434,15 +444,14 @@ async def run_task_v2_helper(
                     scrape_exclude=app.scrape_exclude,
                 )
                 element_tree_in_prompt: str = scraped_page.build_element_tree(ElementTreeFormat.HTML)
-                page = await browser_state.get_working_page()
+                if page is None:
+                    page = await browser_state.get_working_page()
             except Exception:
                 LOG.exception(
                     "Failed to get browser state or scrape website in task v2 iteration", iteration=i, url=url
                 )
                 continue
-            current_url = str(
+            current_url = current_url if current_url else str(await SkyvernFrame.get_url(frame=page) if page else url)
-                await SkyvernFrame.evaluate(frame=page, expression="() => document.location.href") if page else url
-            )
 
             task_v2_prompt = prompt_engine.load_prompt(
                 "task_v2",

skyvern/forge/sdk/workflow/models/block.py

@@ -68,6 +68,7 @@ from skyvern.forge.sdk.workflow.models.parameter import (
     WorkflowParameter,
 )
 from skyvern.webeye.browser_factory import BrowserState
+from skyvern.webeye.utils.page import SkyvernFrame
 
 LOG = structlog.get_logger()
 
@@ -2144,6 +2145,15 @@ class TaskV2Block(Block):
         from skyvern.forge.sdk.services import task_v2_service
         from skyvern.forge.sdk.workflow.models.workflow import WorkflowRunStatus
 
+        if not self.url:
+            browser_state = app.BROWSER_MANAGER.get_for_workflow_run(workflow_run_id)
+            if browser_state:
+                page = await browser_state.get_working_page()
+                if page:
+                    current_url = await SkyvernFrame.get_url(frame=page)
+                    if current_url != "about:blank":
+                        self.url = current_url
+
         if not organization_id:
             raise ValueError("Running TaskV2Block requires organization_id")
 

skyvern/webeye/browser_manager.py

@@ -132,13 +132,11 @@ class BrowserManager:
     ) -> BrowserState:
         parent_workflow_run_id = workflow_run.parent_workflow_run_id
         workflow_run_id = workflow_run.workflow_run_id
-        browser_state = self.get_for_workflow_run(workflow_run_id=workflow_run_id)
+        browser_state = self.get_for_workflow_run(
-        if parent_workflow_run_id:
+            workflow_run_id=workflow_run_id, parent_workflow_run_id=parent_workflow_run_id
-            browser_state = self.get_for_workflow_run(workflow_run_id=parent_workflow_run_id)
+        )
-            if browser_state:
+        if browser_state:
-                self.pages[workflow_run_id] = browser_state
+            self.pages[workflow_run_id] = browser_state
-
-        if browser_state is not None:
             return browser_state
 
         if browser_session_id:
@@ -193,9 +191,15 @@ class BrowserManager:
         )
         return browser_state
 
-    def get_for_workflow_run(self, workflow_run_id: str) -> BrowserState | None:
+    def get_for_workflow_run(
+        self, workflow_run_id: str, parent_workflow_run_id: str | None = None
+    ) -> BrowserState | None:
         if workflow_run_id in self.pages:
             return self.pages[workflow_run_id]
+
+        if parent_workflow_run_id and parent_workflow_run_id in self.pages:
+            return self.pages[parent_workflow_run_id]
+
         return None
 
     def set_video_artifact_for_task(self, task: Task, artifacts: list[VideoArtifact]) -> None:

skyvern/webeye/utils/page.py

@@ -46,6 +46,10 @@ class SkyvernFrame:
             LOG.exception("Timeout to evaluate expression", expression=expression)
             raise TimeoutError("timeout to evaluate expression")
 
+    @staticmethod
+    async def get_url(frame: Page | Frame) -> str:
+        return await SkyvernFrame.evaluate(frame=frame, expression="() => document.location.href")
+
     @staticmethod
     async def take_screenshot(
         page: Page,