vodorod/Dorod-Sky
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

suchintan.vibe code user goal check (#2349)

Browse Source 
Co-authored-by: lawyzheng <lawyzheng1106@gmail.com>
This commit is contained in:
Shuchang Zheng
2025-05-15 08:18:24 -07:00
committed by GitHub
parent 847ddacebd
commit ed4280153f
30 changed files with 251 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
skyvern/forge/agent.py
View File

@@ -172,6 +172,7 @@ class ForgeAgent:
retry=task_retry,
max_steps_per_run=task_block.max_steps_per_run,
error_code_mapping=task_block.error_code_mapping,
include_action_history_in_verification=task_block.include_action_history_in_verification,
)
LOG.info(
"Created a new task for workflow run",
@@ -226,6 +227,7 @@ class ForgeAgent:
extracted_information_schema=task_request.extracted_information_schema,
error_code_mapping=task_request.error_code_mapping,
application=task_request.application,
include_action_history_in_verification=task_request.include_action_history_in_verification,
)
LOG.info(
"Created new task",
@@ -1033,6 +1035,7 @@ class ForgeAgent:
for result in results:
result.step_retry_number = step.retry_index
result.step_order = step.order
step.output = detailed_agent_step_output.to_agent_step_output()
action_results.extend(results)
# Check the last result for this action. If that succeeded, assume the entire action is successful
if results and results[-1].success:
@@ -1462,8 +1465,9 @@ class ForgeAgent:
)
return actions
@staticmethod
async def complete_verify(page: Page, scraped_page: ScrapedPage, task: Task, step: Step) -> CompleteVerifyResult:
async def complete_verify(
self, page: Page, scraped_page: ScrapedPage, task: Task, step: Step
) -> CompleteVerifyResult:
LOG.info(
"Checking if user goal is achieved after re-scraping the page",
task_id=task.task_id,
@@ -1477,6 +1481,10 @@ class ForgeAgent:
scraped_page_refreshed = await scraped_page.refresh(draw_boxes=False, scroll=scroll)
actions_and_results_str = ""
if task.include_action_history_in_verification:
actions_and_results_str = await self._get_action_results(task, current_step=step)
verification_prompt = load_prompt_with_elements(
scraped_page=scraped_page_refreshed,
prompt_engine=prompt_engine,
@@ -1484,6 +1492,7 @@ class ForgeAgent:
navigation_goal=task.navigation_goal,
navigation_payload=task.navigation_payload,
complete_criterion=task.complete_criterion,
action_history=actions_and_results_str,
local_datetime=datetime.now(skyvern_context.ensure_context().tz_info).isoformat(),
)
@@ -1496,12 +1505,11 @@ class ForgeAgent:
)
return CompleteVerifyResult.model_validate(verification_result)
@staticmethod
async def check_user_goal_complete(
page: Page, scraped_page: ScrapedPage, task: Task, step: Step
self, page: Page, scraped_page: ScrapedPage, task: Task, step: Step
) -> CompleteAction | None:
try:
verification_result = await app.agent.complete_verify(
verification_result = await self.complete_verify(
page=page,
scraped_page=scraped_page,
task=task,
@@ -1878,11 +1886,20 @@ class ForgeAgent:
current_context.totp_codes.pop(task.task_id)
return final_navigation_payload
async def _get_action_results(self, task: Task) -> str:
async def _get_action_results(self, task: Task, current_step: Step | None = None) -> str:
"""
Get the action results from the last app.SETTINGS.PROMPT_ACTION_HISTORY_WINDOW steps.
If current_step is provided, the current executing step will be included in the action history.
Default is excluding the current executing step from the action history.
"""
# Get action results from the last app.SETTINGS.PROMPT_ACTION_HISTORY_WINDOW steps
steps = await app.DATABASE.get_task_steps(task_id=task.task_id, organization_id=task.organization_id)
# the last step is always the newly created one and it should be excluded from the history window
window_steps = steps[-1 - settings.PROMPT_ACTION_HISTORY_WINDOW : -1]
if current_step:
window_steps.append(current_step)
actions_and_results: list[tuple[Action, list[ActionResult]]] = []
for window_step in window_steps:
if window_step.output and window_step.output.actions_and_results: