Skip to content

Automate browser based workflows with AI

27.2 KB
Raw Blame
from __future__ import annotations

import asyncio
import base64
from dataclasses import dataclass, field
from datetime import UTC, datetime
from typing import Any
from unittest.mock import AsyncMock, MagicMock
from zoneinfo import ZoneInfo

import pytest

from skyvern.exceptions import FailedToReloadPage, NoTOTPVerificationCodeFound, ScrapingFailedBlankPage
from skyvern.forge import app
from skyvern.forge.agent import ForgeAgent, StepPromptResult
from skyvern.forge.sdk.api.llm.exceptions import LLMResponseMissingActionsError
from skyvern.forge.sdk.core import skyvern_context
from skyvern.forge.sdk.core.skyvern_context import SkyvernContext
from skyvern.forge.sdk.models import Step, StepStatus
from skyvern.forge.sdk.schemas.organizations import Organization
from skyvern.forge.sdk.schemas.tasks import Task
from skyvern.forge.sdk.workflow.models.block import BaseTaskBlock, FileDownloadBlock
from skyvern.schemas.steps import AgentStepOutput
from skyvern.webeye.actions.action_types import ActionType
from skyvern.webeye.actions.actions import (
    Action,
    ClickAction,
    CompleteAction,
    DownloadFileAction,
    ExtractAction,
    TerminateAction,
    WaitAction,
)
from skyvern.webeye.actions.models import DetailedAgentStepOutput
from skyvern.webeye.actions.responses import ActionFailure, ActionResult, ActionSuccess
from skyvern.webeye.scraper.scraped_page import ScrapedPage
from tests.unit.helpers import make_browser_state, make_organization, make_step, make_task
from tests.unit.scoped_asyncio import ScopedAsyncio


def _click(element_id: str = "node-1") -> ClickAction:
    return ClickAction(
        element_id=element_id,
        organization_id="org-123",
        workflow_run_id="workflow-1",
        task_id="task-123",
        step_id="step-char",
        step_order=0,
        action_order=0,
    )


@dataclass
class AgentStepRig:
    agent: ForgeAgent
    organization: Organization
    task: Task
    step: Step
    browser_state: MagicMock
    scraped_page: ScrapedPage
    context: SkyvernContext
    llm_handler: AsyncMock
    action_handler: AsyncMock
    update_statuses: list[StepStatus | None] = field(default_factory=list)

    async def run(self, task_block: BaseTaskBlock | None = None) -> tuple[Step, DetailedAgentStepOutput]:
        skyvern_context.set(self.context)
        try:
            return await self.agent.agent_step(
                task=self.task,
                step=self.step,
                browser_state=self.browser_state,
                organization=self.organization,
                task_block=task_block,
            )
        finally:
            skyvern_context.reset()


def make_agent_step_rig(
    monkeypatch: pytest.MonkeyPatch,
    *,
    parsed_actions: list[Action] | None = None,
    action_handler: AsyncMock | None = None,
    injected_actions: list[Action] | None = None,
    task_overrides: dict[str, Any] | None = None,
    disable_user_goal_check: bool = True,
) -> AgentStepRig:
    agent = ForgeAgent()
    now = datetime.now(UTC)
    organization = make_organization(now)
    overrides: dict[str, Any] = {"navigation_goal": "Reach confirmation page", "workflow_run_id": "workflow-1"}
    overrides.update(task_overrides or {})
    task = make_task(now, organization, **overrides)
    step = make_step(now, task, step_id="step-char", status=StepStatus.created, order=0, output=None)

    browser_state, _, page = make_browser_state()
    browser_state.must_get_working_page = AsyncMock(return_value=page)
    browser_state.get_working_page = AsyncMock(return_value=page)
    browser_state.reload_page = AsyncMock()

    async def _dummy_cleanup(*_args, **_kwargs) -> list[dict]:
        return []

    scraped_page = ScrapedPage(
        elements=[],
        element_tree=[],
        element_tree_trimmed=[],
        _browser_state=browser_state,
        _clean_up_func=_dummy_cleanup,
        _scrape_exclude=None,
    )
    scraped_page.screenshots = [b"image"]

    agent.build_and_record_step_prompt = AsyncMock(
        return_value=StepPromptResult(
            scraped_page=scraped_page,
            extract_action_prompt="prompt",
            use_caching=False,
            prompt_name="extract-actions",
            without_page_information=False,
        )
    )
    json_response: dict[str, object] = {"actions": [{"action_type": "CLICK", "element_id": "node-1"}]}
    agent.handle_potential_OTP_actions = AsyncMock(return_value=(json_response, []))

    actions = parsed_actions if parsed_actions is not None else [_click()]
    monkeypatch.setattr("skyvern.forge.agent.parse_actions", lambda *_, **__: actions)

    if action_handler is None:
        action_handler = AsyncMock(return_value=[ActionSuccess()])
    monkeypatch.setattr("skyvern.forge.agent.ActionHandler.handle_action", action_handler)
    agent.record_artifacts_after_action = AsyncMock()
    agent.check_user_goal_complete = AsyncMock()

    llm_handler = AsyncMock(return_value=json_response)
    monkeypatch.setattr(
        "skyvern.forge.agent.LLMAPIHandlerFactory.get_override_llm_api_handler",
        lambda *_args, **_kwargs: llm_handler,
    )
    monkeypatch.setattr(
        "skyvern.forge.agent.app.AGENT_FUNCTION.prepare_step_execution",
        AsyncMock(return_value=injected_actions),
    )
    monkeypatch.setattr("skyvern.forge.agent.app.AGENT_FUNCTION.post_action_execution", AsyncMock())
    monkeypatch.setattr("skyvern.forge.agent.asyncio", ScopedAsyncio(sleep=AsyncMock(return_value=None)))
    monkeypatch.setattr("skyvern.forge.agent.random.uniform", lambda *_args, **_kwargs: 0)
    monkeypatch.setattr("skyvern.forge.agent.app.DATABASE.workflow_params.create_action", AsyncMock())
    # Wait-time optimization is a cloud experiment (OSS/killswitch-off returns None).
    # Pin that here so the rig never routes into the half-mocked experiment provider,
    # which would cache a malformed WaitConfig in a module-global keyed by task_id and
    # leak "coroutine never awaited" warnings / cross-test state.
    monkeypatch.setattr("skyvern.forge.agent.get_or_create_wait_config", AsyncMock(return_value=None))

    async def _flag(flag_name: str, *_args, **_kwargs) -> bool:
        if flag_name == "DISABLE_USER_GOAL_CHECK":
            return disable_user_goal_check
        return False

    monkeypatch.setattr(
        "skyvern.forge.agent.app.EXPERIMENTATION_PROVIDER.is_feature_enabled_cached",
        AsyncMock(side_effect=_flag),
    )

    update_statuses: list[StepStatus | None] = []

    async def fake_update_step(
        step: Step,
        status: StepStatus | None = None,
        output=None,
        is_last: bool | None = None,
        retry_index: int | None = None,
        **_kwargs,
    ) -> Step:
        update_statuses.append(status)
        if status is not None:
            step.status = status
        if output is not None:
            step.output = output
        return step

    agent.update_step = AsyncMock(side_effect=fake_update_step)

    context = SkyvernContext(
        task_id=task.task_id,
        step_id=None,
        organization_id=task.organization_id,
        workflow_run_id=task.workflow_run_id,
        tz_info=ZoneInfo("UTC"),
    )
    return AgentStepRig(
        agent=agent,
        organization=organization,
        task=task,
        step=step,
        browser_state=browser_state,
        scraped_page=scraped_page,
        context=context,
        llm_handler=llm_handler,
        action_handler=action_handler,
        update_statuses=update_statuses,
    )


@pytest.mark.asyncio
async def test_injected_actions_from_prepare_step_execution_skip_llm(monkeypatch: pytest.MonkeyPatch) -> None:
    injected = _click()
    rig = make_agent_step_rig(monkeypatch, injected_actions=[injected])

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    assert rig.llm_handler.await_count == 0
    assert rig.action_handler.await_count == 1
    assert rig.action_handler.await_args.kwargs["action"] is injected
    assert output.actions == [injected]


@pytest.mark.asyncio
async def test_agent_step_wires_file_download_false_click_eligibility(monkeypatch: pytest.MonkeyPatch) -> None:
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[_click()])

    await rig.run(task_block=FileDownloadBlock.model_construct(label="test", complete_on_download=False))

    assert rig.action_handler.await_args.kwargs["file_download_false_click_eligible"] is True


@pytest.mark.asyncio
async def test_no_generated_actions_marks_step_failed(monkeypatch: pytest.MonkeyPatch) -> None:
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[])

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert rig.action_handler.await_count == 0
    assert output.actions == []


@pytest.mark.asyncio
async def test_credited_dead_blank_recovery_persists_completed_not_failed(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    """Full path: a dead-blank scrape whose credited grace returns [] must NOT persist the step as
    failed at the zero-action seam. execute_step's single complete-on-download seam owns the finalize
    and task completion, so agent_step returns a completed step (no failed terminal step/metrics)."""
    rig = make_agent_step_rig(monkeypatch)
    rig.agent.build_and_record_step_prompt = AsyncMock(side_effect=ScrapingFailedBlankPage())
    rig.agent._empty_page_recovery_plan = AsyncMock(return_value=[])  # credited complete-on-download

    step, output = await rig.run(task_block=FileDownloadBlock.model_construct(label="dl", complete_on_download=True))

    assert step.status == StepStatus.completed
    assert StepStatus.failed not in rig.update_statuses
    assert rig.action_handler.await_count == 0
    assert output.actions == []


@pytest.mark.asyncio
async def test_totp_polling_timeout_produces_terminate_action(monkeypatch: pytest.MonkeyPatch) -> None:
    rig = make_agent_step_rig(monkeypatch, task_overrides={"totp_identifier": "[email protected]"})
    rig.agent.handle_potential_OTP_actions = AsyncMock(side_effect=NoTOTPVerificationCodeFound(task_id="task-123"))

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    assert output.actions is not None
    assert output.actions[0].action_type == ActionType.TERMINATE
    assert "[email protected]" in output.actions[0].reasoning


@pytest.mark.asyncio
async def test_pdf_viewer_embed_generates_download_action(monkeypatch: pytest.MonkeyPatch) -> None:
    pdf_bytes = b"%PDF-1.4 characterization"
    pdf_src = "data:application/pdf;base64," + base64.b64encode(pdf_bytes).decode()
    rig = make_agent_step_rig(monkeypatch)
    monkeypatch.setattr(ScrapedPage, "check_pdf_viewer_embed", lambda self: pdf_src)

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    assert output.actions is not None
    action = output.actions[0]
    assert isinstance(action, DownloadFileAction)
    assert action.byte == pdf_bytes
    assert action.download is True
    assert len(rig.context.downloaded_pdf_sources) == 1


@pytest.mark.asyncio
async def test_wait_actions_skipped_when_batched_with_other_actions(monkeypatch: pytest.MonkeyPatch) -> None:
    wait = WaitAction(seconds=3)
    click = _click()
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[wait, click])

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    assert rig.action_handler.await_count == 1
    assert output.actions_and_results is not None
    assert [action for action, _ in output.actions_and_results] == [click]


@pytest.mark.asyncio
async def test_failed_action_marks_step_failed_and_skips_remaining(monkeypatch: pytest.MonkeyPatch) -> None:
    first, second = _click("node-1"), _click("node-2")
    # The stop-the-batch decision is driven by the RESULT's stop_execution_on_failure
    # (default True), not by the action. Set it explicitly to pin the flag-driven path.
    handler = AsyncMock(return_value=[ActionFailure(Exception("element vanished"), stop_execution_on_failure=True)])
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[first, second], action_handler=handler)

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert handler.await_count == 1
    # get_clean_detailed_output strips the (second, []) placeholder: only executed actions survive.
    assert output.actions_and_results is not None
    assert len(output.actions_and_results) == 1
    assert output.actions_and_results[0][0] is first
    assert output.actions_and_results[0][1][0].success is False


@pytest.mark.asyncio
async def test_failed_action_with_continue_flag_executes_remaining(monkeypatch: pytest.MonkeyPatch) -> None:
    # Mirror of the skip case: a failure result that opts out of stopping the batch
    # (stop_execution_on_failure=False) lets the loop run every action.
    first, second = _click("node-1"), _click("node-2")
    handler = AsyncMock(return_value=[ActionFailure(Exception("transient"), stop_execution_on_failure=False)])
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[first, second], action_handler=handler)

    step, output = await rig.run()

    # A tolerated failure (stop_execution_on_failure=False) does not fail the step —
    # every action runs and the step still completes.
    assert step.status == StepStatus.completed
    assert handler.await_count == 2
    assert output.actions_and_results is not None
    assert [action for action, _ in output.actions_and_results] == [first, second]


@pytest.mark.asyncio
@pytest.mark.parametrize(("skip", "expected_calls"), [(True, 1), (False, 2)])
async def test_failed_action_skip_remaining_controls_duplicate_element_retry(
    monkeypatch: pytest.MonkeyPatch, skip: bool, expected_calls: int
) -> None:
    first, duplicate = _click("node-1"), _click("node-1")
    failure = ActionFailure(Exception("unverified click"))
    failure.skip_remaining_actions = skip
    action_handler = AsyncMock(return_value=[failure])
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[first, duplicate], action_handler=action_handler)

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert action_handler.await_count == expected_calls
    assert output.actions_and_results is not None
    assert [action for action, _ in output.actions_and_results] == ([first] if skip else [first, duplicate])


@pytest.mark.asyncio
async def test_freetext_mismatch_failure_stops_duplicate_element_submit(monkeypatch: pytest.MonkeyPatch) -> None:
    # Regression (SKY-13631 review): a free-text mismatch failure from the heal seam must terminally stop the
    # batch even when a queued Submit targets the SAME element id (the duplicate-element-id branch). The heal
    # builds its failures through _freetext_mismatch_failure, which sets skip_remaining_actions=True, so the
    # loop marks the step failed and returns instead of continuing into the duplicate Submit.
    from skyvern.exceptions import FreeTextInputMismatch
    from skyvern.webeye.actions.handler import _freetext_mismatch_failure

    input_action, submit = _click("node-1"), _click("node-1")  # same element id -> duplicate linked node
    failure = _freetext_mismatch_failure(FreeTextInputMismatch(element_id="node-1", intended_length=30))
    handler = AsyncMock(return_value=[failure])
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[input_action, submit], action_handler=handler)

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert handler.await_count == 1  # the duplicate Submit was NOT dispatched
    assert output.actions_and_results is not None
    assert [action for action, _ in output.actions_and_results] == [input_action]


@pytest.mark.asyncio
async def test_skip_remaining_actions_stops_batch_but_step_completes(monkeypatch: pytest.MonkeyPatch) -> None:
    first, second = _click("node-1"), _click("node-2")
    handler = AsyncMock(return_value=[ActionResult(success=True, skip_remaining_actions=True)])
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[first, second], action_handler=handler)

    step, _output = await rig.run()

    assert step.status == StepStatus.completed
    assert handler.await_count == 1


@pytest.mark.asyncio
async def test_refresh_working_page_signal_reloads_and_skips_batch(monkeypatch: pytest.MonkeyPatch) -> None:
    rig = make_agent_step_rig(monkeypatch)
    rig.context.refresh_working_page = True

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    rig.browser_state.reload_page.assert_awaited_once()
    assert rig.action_handler.await_count == 0
    assert output.actions_and_results is not None
    assert output.actions_and_results[0][0].action_type == ActionType.RELOAD_PAGE
    assert rig.context.refresh_working_page is False


@pytest.mark.asyncio
async def test_reload_action_window_encloses_the_reload(monkeypatch: pytest.MonkeyPatch) -> None:
    """started_at is captured BEFORE the reload, so the recorded window encloses it instead of
    collapsing to a zero-duration stamp taken after the reload already finished."""
    rig = make_agent_step_rig(monkeypatch)
    rig.context.refresh_working_page = True
    during_reload: list[datetime] = []

    async def observed_reload() -> None:
        await asyncio.sleep(0.01)
        during_reload.append(datetime.now(UTC).replace(tzinfo=None))

    rig.browser_state.reload_page = AsyncMock(side_effect=observed_reload)

    _step, output = await rig.run()

    assert during_reload, "reload side effect never ran; the test is not exercising the reload path"
    assert output.actions_and_results is not None
    reload_action = output.actions_and_results[0][0]
    assert reload_action.started_at is not None and reload_action.finished_at is not None
    assert reload_action.started_at <= during_reload[0] <= reload_action.finished_at


@pytest.mark.asyncio
async def test_unexpected_exception_returns_failed_step_instead_of_raising(monkeypatch: pytest.MonkeyPatch) -> None:
    handler = AsyncMock(side_effect=RuntimeError("browser exploded"))
    rig = make_agent_step_rig(monkeypatch, action_handler=handler)

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert output.step_exception == "RuntimeError"


@pytest.mark.asyncio
@pytest.mark.parametrize(
    ("raised", "expected_log_method"),
    [
        (FailedToReloadPage("https://example.test", "Page.reload: Timeout 60000ms exceeded."), "warning"),
        (LLMResponseMissingActionsError(["page_info"]), "warning"),
        (RuntimeError("browser exploded"), "exception"),
    ],
)
async def test_expected_step_failures_log_at_warning_and_other_exceptions_at_error(
    monkeypatch: pytest.MonkeyPatch, raised: Exception, expected_log_method: str
) -> None:
    log = MagicMock()
    monkeypatch.setattr("skyvern.forge.agent.LOG", log)
    rig = make_agent_step_rig(monkeypatch, action_handler=AsyncMock(side_effect=raised))

    step, output = await rig.run()

    assert step.status == StepStatus.failed
    assert output.step_exception == type(raised).__name__
    leveled = [call[0] for call in log.method_calls if call[0] in ("warning", "error", "exception")]
    assert leveled == [expected_log_method]


@pytest.mark.asyncio
async def test_successful_complete_action_with_extraction_goal_appends_extract(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    complete = CompleteAction(reasoning="goal reached")
    rig = make_agent_step_rig(monkeypatch, parsed_actions=[complete])
    monkeypatch.setattr("skyvern.forge.agent.app.DATABASE.tasks.get_task", AsyncMock(return_value=rig.task))
    extract = ExtractAction(
        reasoning="collect",
        data_extraction_goal=rig.task.data_extraction_goal,
        data_extraction_schema=None,
    )
    rig.agent.create_extract_action = AsyncMock(return_value=extract)

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    rig.agent.create_extract_action.assert_awaited_once()
    assert rig.action_handler.await_count == 2
    assert output.actions_and_results is not None
    assert output.actions_and_results[-1][0] is extract


@pytest.mark.asyncio
async def test_parallel_verification_marks_speculative_original_status(monkeypatch: pytest.MonkeyPatch) -> None:
    rig = make_agent_step_rig(monkeypatch, disable_user_goal_check=False)

    step, _output = await rig.run()

    assert step.status == StepStatus.completed
    assert step.speculative_original_status == StepStatus.completed


# SKY-15xxx: TaskStatus.terminated requires a non-empty failure_reason (Task.validate_update).
# get_failure_reason_for_task is the only source of that reason for a terminated task; either
# empty path below used to return None, which handle_completed_step then handed straight to
# update_task, tripping the invariant and crashing the step as an "unexpected exception".
def _prime_blank_recovery(rig: AgentStepRig, *, survivors: list[str] | None = None, dead_url: str = ":") -> MagicMock:
    """Make the step-body scrape raise a dead-blank error and shape the browser so recovery can
    inspect a dead working page plus any survivors. Returns the dead working page mock."""
    from skyvern.exceptions import ScrapingFailedBlankPage

    dead_page = MagicMock()
    dead_page.url = dead_url
    dead_page.main_frame.child_frames = []
    dead_page.is_closed.return_value = False
    rig.browser_state.get_working_page = AsyncMock(return_value=dead_page)

    pages = [dead_page]
    for url in survivors or []:
        survivor = MagicMock()
        survivor.url = url
        survivor.is_closed.return_value = False
        pages.append(survivor)
    rig.browser_state.list_valid_pages = AsyncMock(return_value=pages)

    rig.agent.build_and_record_step_prompt = AsyncMock(side_effect=ScrapingFailedBlankPage())
    return dead_page


@pytest.mark.asyncio
async def test_dead_blank_working_page_recovers_by_injecting_internal_close(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    # A ":" working page with one http survivor: the step's plan is exactly one internal-recovery
    # ClosePageAction, synthesized with no action-plan LLM call, and the step completes.
    rig = make_agent_step_rig(monkeypatch)
    _prime_blank_recovery(rig, survivors=["https://survivor.test/app"])

    step, output = await rig.run()

    assert step.status == StepStatus.completed
    assert rig.llm_handler.await_count == 0
    assert output.actions is not None
    assert len(output.actions) == 1
    close = output.actions[0]
    assert close.action_type == ActionType.CLOSE_PAGE
    assert close.is_internal_recovery is True
    assert rig.action_handler.await_args.kwargs["action"] is close
    assert rig.context.empty_page_recovery_step_id == step.step_id
    assert rig.context.empty_page_recovery_attempts[rig.task.task_id] == 1


@pytest.mark.asyncio
async def test_dead_blank_without_http_survivor_reraises_to_terminal(monkeypatch: pytest.MonkeyPatch) -> None:
    from skyvern.exceptions import ScrapingFailedBlankPage

    # No usable survivor -> ineligible -> the original blank exception propagates to the terminal
    # handler, and nothing (no LLM, no action) is executed.
    rig = make_agent_step_rig(monkeypatch)
    _prime_blank_recovery(rig, survivors=[])

    with pytest.raises(ScrapingFailedBlankPage):
        await rig.run()
    assert rig.llm_handler.await_count == 0
    assert rig.action_handler.await_count == 0
    assert rig.task.task_id not in rig.context.empty_page_recovery_attempts


@pytest.mark.asyncio
async def test_dead_blank_recovery_cap_reraises_after_three_attempts(monkeypatch: pytest.MonkeyPatch) -> None:
    from skyvern.exceptions import ScrapingFailedBlankPage

    rig = make_agent_step_rig(monkeypatch)
    _prime_blank_recovery(rig, survivors=["https://survivor.test/app"])
    rig.context.empty_page_recovery_attempts[rig.task.task_id] = 3

    with pytest.raises(ScrapingFailedBlankPage):
        await rig.run()
    # The cap was already consumed; a 4th detection declines rather than injecting again.
    assert rig.action_handler.await_count == 0
    assert rig.context.empty_page_recovery_attempts[rig.task.task_id] == 3


@pytest.mark.asyncio
async def test_speculative_plan_consumption_resets_recovery_attempt_counter(monkeypatch: pytest.MonkeyPatch) -> None:
    from skyvern.forge.agent import SpeculativePlan

    # A step that consumes a successfully-scraped speculative plan (parallel-verification path) is a
    # successful scrape too, so it must reset the per-task recovery counter — otherwise the
    # "consecutive" cap silently becomes cumulative across non-adjacent blank incidents.
    rig = make_agent_step_rig(monkeypatch)
    rig.agent._persist_scrape_artifacts = AsyncMock()
    rig.context.empty_page_recovery_attempts[rig.task.task_id] = 2
    rig.context.speculative_plans[rig.step.step_id] = SpeculativePlan(
        scraped_page=rig.scraped_page,
        extract_action_prompt="prompt",
        use_caching=False,
        llm_json_response={"actions": [{"action_type": "CLICK", "element_id": "node-1"}]},
    )

    step, _output = await rig.run()

    assert step.status == StepStatus.completed
    assert rig.task.task_id not in rig.context.empty_page_recovery_attempts


@pytest.mark.asyncio
async def test_non_blank_scrape_failure_is_not_recovered(monkeypatch: pytest.MonkeyPatch) -> None:
    from skyvern.exceptions import ScrapingFailed

    # A generic scrape failure (not a dead blank) must propagate untouched: recovery only ever adds a
    # path for the blank signature, never intercepts other terminal scrape errors.
    rig = make_agent_step_rig(monkeypatch)
    rig.agent.build_and_record_step_prompt = AsyncMock(side_effect=ScrapingFailed())

    with pytest.raises(ScrapingFailed):
        await rig.run()
    assert rig.action_handler.await_count == 0
    assert rig.task.task_id not in rig.context.empty_page_recovery_attempts


@pytest.mark.asyncio
async def test_successful_scrape_resets_recovery_attempt_counter(monkeypatch: pytest.MonkeyPatch) -> None:
    # A normal (non-blank) step-body scrape clears a prior task's recovery counter so the next dead
    # blank starts from a fresh budget.
    rig = make_agent_step_rig(monkeypatch)
    rig.context.empty_page_recovery_attempts[rig.task.task_id] = 2

    step, _output = await rig.run()

    assert step.status == StepStatus.completed
    assert rig.task.task_id not in rig.context.empty_page_recovery_attempts


@pytest.mark.asyncio
async def test_get_failure_reason_for_task_falls_back_when_terminate_reasoning_is_empty(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    now = datetime.now(UTC)
    organization = make_organization(now)
    task = make_task(now, organization)
    terminate = TerminateAction(reasoning=None, organization_id=organization.organization_id, task_id=task.task_id)
    output = AgentStepOutput(actions_and_results=[(terminate, [])])
    step = make_step(now, task, step_id="step-0", status=StepStatus.completed, order=0, output=output)
    monkeypatch.setattr(app.DATABASE.tasks, "get_task_steps", AsyncMock(return_value=[step]))

    reason = await ForgeAgent().get_failure_reason_for_task(task)

    assert reason


@pytest.mark.asyncio
async def test_get_failure_reason_for_task_falls_back_when_no_terminate_action_found(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    now = datetime.now(UTC)
    organization = make_organization(now)
    task = make_task(now, organization)
    step = make_step(now, task, step_id="step-0", status=StepStatus.completed, order=0, output=None)
    monkeypatch.setattr(app.DATABASE.tasks, "get_task_steps", AsyncMock(return_value=[step]))

    reason = await ForgeAgent().get_failure_reason_for_task(task)

    assert reason