Skip to content

Automate browser based workflows with AI

Add Claude Haiku 5.5 as an LLM_NAME variant for v1 (SKY-18088) (#8868)

Cindy Li57890df3b43c
diff --git a/skyvern/forge/sdk/copilot/enforcement.py b/skyvern/forge/sdk/copilot/enforcement.py
index 979981895..1b5a884fd 100644
--- a/skyvern/forge/sdk/copilot/enforcement.py
+++ b/skyvern/forge/sdk/copilot/enforcement.py
@@ -643,6 +643,14 @@ def _bounded_error_codes(codes: Any) -> list[str]:
return [str(c)[:64] for c in codes if isinstance(c, str)][:_PACKET_LIST_CAP]
+def _bounded_scalars(fact: dict[str, Any]) -> dict[str, bool | int | float | str]:
+ return {
+ str(key)[:64]: item if isinstance(item, (bool, int, float)) else item[:_PACKET_NOTICE_CAP]
+ for key, item in list(fact.items())[:_PACKET_LIST_CAP]
+ if isinstance(item, (bool, int, float, str))
+ }
+
+
def _retained_run_packet(packet: Any) -> dict[str, Any] | None:
"""The packet's identity and failure scalars, bounded. Lists are tails, so they yield first."""
if not isinstance(packet, dict):
@@ -656,6 +664,12 @@ def _retained_run_packet(packet: Any) -> dict[str, Any] | None:
run_id = run.get("workflow_run_id")
status = run.get("status")
bounded_run = {k: v for k, v in (("workflow_run_id", run_id), ("status", status)) if v}
+ # A later test selection reads an older run by which browser ran it and how that browser
+ # started; reduced to an id and a status, the run reads as having left its page for the next.
+ for key in ("browser", "browser_start"):
+ browser_fact = run.get(key)
+ if isinstance(browser_fact, dict) and (scalars := _bounded_scalars(browser_fact)):
+ bounded_run[key] = scalars
if bounded_run:
kept["run"] = bounded_run
challenge = packet.get("challenge")
diff --git a/skyvern/forge/sdk/copilot/tools/run_execution.py b/skyvern/forge/sdk/copilot/tools/run_execution.py
index 81c577b4a..25db2cc95 100644
--- a/skyvern/forge/sdk/copilot/tools/run_execution.py
+++ b/skyvern/forge/sdk/copilot/tools/run_execution.py
@@ -2317,6 +2317,8 @@ def _attach_run_session_facts(
*,
used_fresh_run_session: bool,
run_detached_from_chat: bool,
+ chat_browser_opened_for_run: bool,
+ continued_earlier_test: bool,
run_ok: bool,
page_evidence: Mapping[str, Any] | None,
) -> None:
@@ -2327,13 +2329,26 @@ def _attach_run_session_facts(
# A carried resume browser is also not the chat's, so detachment is the honest fact for
# "can I look at this run's page from here", not whether the session was minted.
data["run_detached_from_chat"] = run_detached_from_chat
+ # Say which browser the run started in, so code that needed a signed-in state reads as started
+ # somewhere else rather than as broken, and the next request can bring the blocks it needs.
if used_fresh_run_session:
- # Say that the page was blank, so code that needed a signed-in state reads as started
- # somewhere else rather than as broken, and the next request can bring the blocks it needs.
- data.setdefault(
- "browser_start",
- {"kind": "separate_blank_context", "restored_saved_profile": False, "inherited_browser_state": False},
- )
+ browser_start_kind = "separate_blank_context"
+ elif run_detached_from_chat:
+ browser_start_kind = "separate_context_continued_from_earlier_test"
+ elif chat_browser_opened_for_run:
+ browser_start_kind = "chat_browser_context_opened_for_this_run"
+ elif continued_earlier_test:
+ browser_start_kind = "chat_browser_context_continued_from_earlier_test"
+ else:
+ browser_start_kind = "chat_browser_context"
+ data.setdefault(
+ "browser_start",
+ {
+ "kind": browser_start_kind,
+ "restored_saved_profile": False,
+ "inherited_browser_state": not (used_fresh_run_session or chat_browser_opened_for_run),
+ },
+ )
if not isinstance(page_evidence, Mapping):
return
data["challenge_stalled_fresh_session"] = (
@@ -3464,6 +3479,8 @@ async def _attach_post_run_browser_enrichment(
run_ok: bool,
used_fresh_run_session: bool,
run_detached_from_chat: bool,
+ chat_browser_opened_for_run: bool,
+ continued_earlier_test: bool,
origin_registry: OriginRunRedactionRegistry | None = None,
) -> tuple[str, dict[str, Any] | None]:
"""Probe the browser for post-run facts and stamp them onto an already-recorded run result.
@@ -3586,6 +3603,8 @@ async def _attach_post_run_browser_enrichment(
result_data,
used_fresh_run_session=used_fresh_run_session,
run_detached_from_chat=run_detached_from_chat,
+ chat_browser_opened_for_run=chat_browser_opened_for_run,
+ continued_earlier_test=continued_earlier_test,
run_ok=run_ok,
page_evidence=post_run_page_evidence,
)
@@ -3646,6 +3665,7 @@ async def _run_blocks_and_collect_debug(
use_ephemeral_inputs: bool = True,
new_exit: bool = False,
verified_new_exit: FreshExitReceipt | None = None,
+ chat_browser_opened_before_retry: bool = False,
) -> dict[str, Any]:
"""With ``new_exit``, run in a browser whose network exit is verified to differ from the most recent build
test's; the model cannot choose or verify an exit, so the platform supplies it or says why it cannot."""
@@ -4081,6 +4101,8 @@ async def _run_blocks_and_collect_debug(
# Whether the run executed outside the chat's browser at all, by either route. This is what
# gates the post-run rebind and the pane association, neither of which cares which route.
run_detached_from_chat = False
+ # The chat held no usable browser, so acquiring one opened it blank for this run.
+ chat_browser_opened_for_run = False
debug_session_id: str | None = None
run_session_id: str | None
new_exit_receipt: FreshExitReceipt | None = None
@@ -4164,10 +4186,14 @@ async def _run_blocks_and_collect_debug(
else:
# This id is dispatched into a workflow run without ever being attached here, so an
# unverified session cannot be discovered later and must fail now instead.
+ chat_session_id_before_acquisition = ctx.browser_session_id
session_err = await acquire_build_test_browser_session(ctx, fresh=False)
if session_err is not None:
return _with_build_test_acquisition_context(session_err, requested_block_labels=block_labels)
run_session_id = ctx.browser_session_id
+ chat_browser_opened_for_run = (
+ chat_browser_opened_before_retry or run_session_id != chat_session_id_before_acquisition
+ )
dispatch_draft_workflow_id: str | None = None
# run_task is the in-process inline execution task, only ever set on the dev-only inline path.
@@ -4321,6 +4347,7 @@ async def _run_blocks_and_collect_debug(
use_ephemeral_inputs=use_ephemeral_inputs,
new_exit=new_exit,
verified_new_exit=new_exit_receipt,
+ chat_browser_opened_before_retry=chat_browser_opened_for_run,
)
execution.dispatched_input_values = copy.deepcopy(data)
@@ -4792,6 +4819,8 @@ async def _run_blocks_and_collect_debug(
result["data"],
used_fresh_run_session=used_fresh_run_session,
run_detached_from_chat=run_detached_from_chat,
+ chat_browser_opened_for_run=chat_browser_opened_for_run,
+ continued_earlier_test=resume_session_id is not None,
run_ok=False,
page_evidence=_same_run_page_evidence_for_result(ctx, workflow_run.workflow_run_id),
)
@@ -5063,6 +5092,8 @@ async def _run_blocks_and_collect_debug(
run_ok=run_ok,
used_fresh_run_session=used_fresh_run_session,
run_detached_from_chat=run_detached_from_chat,
+ chat_browser_opened_for_run=chat_browser_opened_for_run,
+ continued_earlier_test=resume_session_id is not None,
origin_registry=origin_registry,
)
settle_terminal_challenge_after_enrichment(ctx, response)
diff --git a/skyvern/forge/sdk/workflow/code_block_safety.py b/skyvern/forge/sdk/workflow/code_block_safety.py
index 97acdb5ae..152c0fe19 100644
--- a/skyvern/forge/sdk/workflow/code_block_safety.py
+++ b/skyvern/forge/sdk/workflow/code_block_safety.py
@@ -213,6 +213,14 @@ SANDBOX_ONLY_BUILTINS: frozenset[str] = frozenset(
)
+def json_safe_default(item: object) -> str:
+ # isinstance falls back to reading item.__class__, which runs a block-defined __getattribute__.
+ # A datetime is a date subclass and keeps the placeholder rather than losing its time.
+ if issubclass(type(item), datetime.date) and not issubclass(type(item), datetime.datetime):
+ return datetime.date.isoformat(item) # type: ignore[arg-type]
+ return f"Object '{type(item)}' is not JSON serializable"
+
+
def safe_builtins() -> dict[str, Any]:
"""Builtins a CodeBlock may use while sharing the worker process; dunders drop out via the underscore filter."""
denied = ALWAYS_DENIED_BUILTINS | SANDBOX_ONLY_BUILTINS
diff --git a/skyvern/forge/sdk/workflow/models/block.py b/skyvern/forge/sdk/workflow/models/block.py
index ed39076ef..d310bf8cc 100644
--- a/skyvern/forge/sdk/workflow/models/block.py
+++ b/skyvern/forge/sdk/workflow/models/block.py
@@ -218,7 +218,7 @@ from skyvern.forge.sdk.workflow.code_block_authorized_files import (
)
from skyvern.forge.sdk.workflow.code_block_safety import BLOCKED_ATTRS as CODE_BLOCK_BLOCKED_ATTRS
from skyvern.forge.sdk.workflow.code_block_safety import is_safe_code as _shared_is_safe_code
-from skyvern.forge.sdk.workflow.code_block_safety import module_shims, safe_builtins
+from skyvern.forge.sdk.workflow.code_block_safety import json_safe_default, module_shims, safe_builtins
from skyvern.forge.sdk.workflow.constants import OUTPUT_PARAMETER_MAX_VALUE_BYTES
from skyvern.forge.sdk.workflow.context_manager import (
NON_SECRET_CREDENTIAL_FIELDS,
@@ -10374,9 +10374,7 @@ async def wrapper({default_args}):
# would either crash registration or, via the default= fallback, replace the value with a
# useless placeholder — normalize the wrapper family to its selector/marker first.
result = json_safe_recorder_output(result)
- result = json.loads(
- json.dumps(result, default=lambda value: f"Object '{type(value)}' is not JSON serializable")
- )
+ result = json.loads(json.dumps(result, default=json_safe_default))
# Mask resolved secrets (OTP codes, passwords) a user assigned to a local before they
# reach captured locals, the persisted output, or the logged value. Mirrors
# HttpRequestBlock and is stronger than the name-based excluded_parameter_keys filter.
diff --git a/tests/unit/test_copilot_diagnosis_repair_contract.py b/tests/unit/test_copilot_diagnosis_repair_contract.py
index 1104ded2e..ddf47904a 100644
--- a/tests/unit/test_copilot_diagnosis_repair_contract.py
+++ b/tests/unit/test_copilot_diagnosis_repair_contract.py
@@ -3629,6 +3629,8 @@ def test_fresh_session_run_envelope_carries_typed_session_facts() -> None:
data,
used_fresh_run_session=True,
run_detached_from_chat=False,
+ chat_browser_opened_for_run=False,
+ continued_earlier_test=False,
run_ok=False,
page_evidence=_challenge_wall_page_evidence(ChallengeKind.CAPTCHA.value),
)
@@ -3644,6 +3646,8 @@ def test_run_envelope_omits_the_challenge_stall_fact_without_a_structured_packet
data,
used_fresh_run_session=True,
run_detached_from_chat=False,
+ chat_browser_opened_for_run=False,
+ continued_earlier_test=False,
run_ok=False,
page_evidence=None,
)
@@ -3659,6 +3663,8 @@ def test_passing_fresh_session_run_did_not_stall_on_the_challenge() -> None:
data,
used_fresh_run_session=True,
run_detached_from_chat=False,
+ chat_browser_opened_for_run=False,
+ continued_earlier_test=False,
run_ok=True,
page_evidence=_challenge_wall_page_evidence(ChallengeKind.CAPTCHA.value),
)
diff --git a/tests/unit/test_copilot_generated_browser_page_state_repair.py b/tests/unit/test_copilot_generated_browser_page_state_repair.py
index 76aed4168..ea0fa4978 100644
--- a/tests/unit/test_copilot_generated_browser_page_state_repair.py
+++ b/tests/unit/test_copilot_generated_browser_page_state_repair.py
@@ -26,11 +26,13 @@ from skyvern.forge.sdk.copilot.build_test_outcome import (
from skyvern.forge.sdk.copilot.composition_evidence import model_visible_composition_evidence, parse_composition_html
from skyvern.forge.sdk.copilot.config import BlockAuthoringPolicy
from skyvern.forge.sdk.copilot.context import CodeAuthoringRepairContext, CopilotContext
+from skyvern.forge.sdk.copilot.enforcement import _summarize_tool_output
from skyvern.forge.sdk.copilot.output_contracts import code_block_available_contracts_by_label
from skyvern.forge.sdk.copilot.output_utils import (
_compact_packet_for_aggregate_limit,
project_build_test_packet_for_llm,
project_direct_test_handoff_packet_for_llm,
+ sanitize_tool_result_for_llm,
)
from skyvern.forge.sdk.copilot.runtime_authoring_repair import (
REPAIR_INSTRUCTION_MAX_CHARS,
@@ -43,6 +45,7 @@ from skyvern.forge.sdk.copilot.runtime_authoring_repair import (
)
from skyvern.forge.sdk.copilot.tools.run_execution import (
CopilotExecutionSnapshot,
+ _attach_run_session_facts,
_build_recorded_build_test_outcome,
_ExecutionResult,
_failure_action_trace_summary,
@@ -50,6 +53,7 @@ from skyvern.forge.sdk.copilot.tools.run_execution import (
_record_run_blocks_result,
_RunExecution,
build_test_evidence_packet,
+ finalize_build_test_result,
)
from skyvern.forge.sdk.workflow.models.code_block_recorder import CODE_BLOCK_FILENAME, user_code_line_from_exception
from skyvern.forge.sdk.workflow.models.workflow import Workflow
@@ -1045,3 +1049,74 @@ def test_hidden_control_crowded_page_keeps_text_labeled_summaries_first() -> Non
assert summaries[0] == "Export CSV"
for _, selector in _HIDDEN_CONTROL_EXPANDERS.values():
assert any(summary.startswith(selector) for summary in summaries[1:]), summaries
+
+
+_SIGNED_OUT_READ_LABEL = "read_visitors"
+_SIGNED_OUT_WRITE_LABEL = "write_report_cell"
+_SIGNED_OUT_TITLE = "Sign in"
+
+
+def _signed_out_focused_result() -> dict[str, Any]:
+ """A read and a native write that completed on the sign-in page, in the chat's browser."""
+ data: dict[str, Any] = {
+ "workflow_run_id": "wr_signed_out_focused",
+ "browser_session_id": "pbs_chat",
+ "overall_status": "completed",
+ "requested_block_labels": [_SIGNED_OUT_READ_LABEL, _SIGNED_OUT_WRITE_LABEL],
+ "executed_block_labels": [_SIGNED_OUT_READ_LABEL, _SIGNED_OUT_WRITE_LABEL],
+ "current_url": "https://metrics.fixture.test/dashboard",
+ "page_title": _SIGNED_OUT_TITLE,
+ "blocks": [
+ {
+ "workflow_run_block_id": "wrb_read_visitors",
+ "label": _SIGNED_OUT_READ_LABEL,
+ "block_type": "EXTRACTION",
+ "status": "completed",
+ "output": {"status": "completed", "extracted_information": {"visitors": None}},
+ },
+ {
+ "workflow_run_block_id": "wrb_write_report_cell",
+ "label": _SIGNED_OUT_WRITE_LABEL,
+ "block_type": "HTTP_REQUEST",
+ "status": "completed",
+ "output": {"status_code": 200, "request_body": {"cell": "A1", "value": "None"}},
+ },
+ ],
+ }
+ _attach_run_session_facts(
+ data,
+ used_fresh_run_session=False,
+ run_detached_from_chat=False,
+ chat_browser_opened_for_run=False,
+ continued_earlier_test=False,
+ run_ok=True,
+ page_evidence=None,
+ )
+ return {"ok": True, "data": data}
+
+
+def test_a_signed_out_focused_run_reaches_the_model_as_facts_and_keeps_its_browser_as_an_older_result() -> None:
+ ctx = _copilot_context()
+ result = _signed_out_focused_result()
+ _record_run_blocks_result(ctx, copy.deepcopy(result))
+
+ model_facing = sanitize_tool_result_for_llm(
+ "run_blocks_and_collect_debug",
+ finalize_build_test_result(ctx, source_tool="run_blocks_and_collect_debug", result=result),
+ )
+
+ assert model_facing["ok"] is True and "error" not in model_facing
+ packet = model_facing["data"]["build_test_packet"]
+ assert packet["page_state"]["title"] == _SIGNED_OUT_TITLE
+ outputs = {entry["label"]: entry["output"] for entry in packet["registered_outputs"]}
+ assert outputs[_SIGNED_OUT_READ_LABEL]["extracted_information"] == {"visitors": None}
+ assert outputs[_SIGNED_OUT_WRITE_LABEL]["request_body"] == {"cell": "A1", "value": "None"}
+ assert packet["run"]["browser_start"]["kind"] == "chat_browser_context"
+ outcome = ctx.latest_recorded_build_test_outcome
+ assert outcome is not None and outcome.verdict == "not_authoritative"
+
+ older = _summarize_tool_output(json.dumps(model_facing))
+ kept = json.loads(older)["build_test_packet"]["run"]
+ for key in ("workflow_run_id", "status", "browser", "browser_start"):
+ assert kept[key] == packet["run"][key], key
+ assert _summarize_tool_output(older) == older
diff --git a/tests/unit/test_copilot_watchdog.py b/tests/unit/test_copilot_watchdog.py
index a6096a8f8..3b4461f71 100644
--- a/tests/unit/test_copilot_watchdog.py
+++ b/tests/unit/test_copilot_watchdog.py
@@ -21,6 +21,8 @@ from skyvern.forge.sdk.copilot.context import CopilotContext
from skyvern.forge.sdk.copilot.repair_origin_run import (
OriginBlockOutput,
OriginExecutionSettings,
+ OriginOutputRefusal,
+ OriginOutputRefusalDetail,
OriginOutputSnapshot,
SelectedOutputSource,
)
@@ -602,6 +604,7 @@ async def test_non_paused_watchdog_exit_still_cancels_and_clears(monkeypatch: py
result = await _run_blocks_and_collect_debug({"block_labels": ["extract_heading"], "parameters": {}}, ctx)
assert result["data"]["control_signal"]["kind"] == "watchdog_ceiling"
+ assert result["data"]["browser_start"]["kind"] == "separate_context_continued_from_earlier_test"
harness["cancel_run_task"].assert_awaited_once()
harness["clear"].assert_awaited_once()
assert _adopted_detached_tasks(before) == []
@@ -617,6 +620,96 @@ async def test_non_paused_watchdog_exit_still_cancels_and_clears(monkeypatch: py
assert "Run ID:" in result["error"]
+ ("chat_session_id", "resume_session_id", "acquired_session_id", "kind", "inherited_browser_state"),
+ [
+ ("pbs_chat", None, "pbs_chat", "chat_browser_context", True),
+ ("pbs_chat", "pbs_chat", "pbs_chat", "chat_browser_context_continued_from_earlier_test", True),
+ (None, None, "pbs_chat", "chat_browser_context_opened_for_this_run", False),
+ ("pbs_chat", "pbs_chat", "pbs_chat_replacement", "chat_browser_context_opened_for_this_run", False),
+ ],
+)
+async def test_a_chat_browser_run_at_the_ceiling_says_which_browser_it_started_in(
+ monkeypatch: pytest.MonkeyPatch,
+ chat_session_id: str | None,
+ resume_session_id: str | None,
+ acquired_session_id: str,
+ kind: str,
+ inherited_browser_state: bool,
+) -> None:
+ harness = await _install_run_harness(
+ monkeypatch, workflow_yaml=SEARCH_THEN_SELECT_WORKFLOW_YAML, polled_status="running"
+ )
+ _install_advancing_clock(monkeypatch)
+
+ async def _acquire(ctx: CopilotContext, **_kwargs: object) -> None:
+ ctx.browser_session_id = acquired_session_id
+
+ monkeypatch.setattr(run_execution, "acquire_build_test_browser_session", _acquire)
+ ctx = make_copilot_ctx(browser_session_id=chat_session_id)
+ ctx.staged_workflow = harness["workflow"]
+ ctx.frontier_resume_session_id = resume_session_id
+
+ result = await _run_blocks_and_collect_debug({"block_labels": ["select_first_result"], "parameters": {}}, ctx)
+
+ assert result["data"]["control_signal"]["kind"] == "watchdog_ceiling"
+ assert result["data"]["browser_start"] == {
+ "kind": kind,
+ "restored_saved_profile": False,
+ "inherited_browser_state": inherited_browser_state,
+ }
+
+
+async def test_a_retry_after_a_refused_earlier_output_keeps_the_chat_browser_it_opened_as_blank(
+ monkeypatch: pytest.MonkeyPatch,
+) -> None:
+ harness = await _install_run_harness(
+ monkeypatch, workflow_yaml=SEARCH_THEN_SELECT_WORKFLOW_YAML, polled_status="running"
+ )
+ _install_advancing_clock(monkeypatch)
+
+ async def _acquire(ctx: CopilotContext, **_kwargs: object) -> None:
+ ctx.browser_session_id = ctx.browser_session_id or "pbs_chat"
+
+ refusals = [
+ OriginOutputRefusalDetail(
+ reason=OriginOutputRefusal.CHANGED_PRODUCER, block_label="run_search", origin_workflow_run_id="wr_origin"
+ )
+ ]
+
+ def _refuse_once_after_persistence(*_args: object, check_settings: bool = True, **_kwargs: object) -> object:
+ return refusals.pop() if check_settings and refusals else None
+
+ monkeypatch.setattr(run_execution, "acquire_build_test_browser_session", _acquire)
+ monkeypatch.setattr(run_execution, "selected_output_run_refusal", _refuse_once_after_persistence)
+ ctx = make_copilot_ctx(browser_session_id=None)
+ ctx.staged_workflow = harness["workflow"]
+ origin = OriginOutputSnapshot(
+ definition=harness["workflow"].workflow_definition,
+ outputs={
+ "run_search": OriginBlockOutput(
+ status=BlockStatus.completed, has_value=True, created_at=datetime.now(UTC), value={"results": []}
+ )
+ },
+ settings=OriginExecutionSettings.of(harness["workflow"]),
+ )
+ ctx.repair_origin_outputs = origin
+ ctx.frontier_selected_output_sources = {
+ "run_search": SelectedOutputSource("run_search", "wr_origin", "origin", origin)
+ }
+
+ result = await _run_blocks_and_collect_debug({"block_labels": ["select_first_result"], "parameters": {}}, ctx)
+
+ assert refusals == []
+ assert result["data"]["browser_start"] == {
+ "kind": "chat_browser_context_opened_for_this_run",
+ "restored_saved_profile": False,
+ "inherited_browser_state": False,
+ }
+
+
@pytest.mark.asyncio
async def test_non_success_watchdog_result_types_selected_failed_block_locators(
monkeypatch: pytest.MonkeyPatch,
diff --git a/tests/unit/test_workflow_service_cancel.py b/tests/unit/test_workflow_service_cancel.py
index 256f7d273..da425b17c 100644
--- a/tests/unit/test_workflow_service_cancel.py
+++ b/tests/unit/test_workflow_service_cancel.py
@@ -57,6 +57,16 @@ async def test_mark_canceled_if_not_final_returns_conditional_result(
)
+def _patch_service_log(monkeypatch: pytest.MonkeyPatch, **methods: object) -> None:
+ """Replace the service module's LOG rather than a method on it. Undoing a patched method on
+ structlog's lazy proxy leaves a bound logger stuck to the proxy, and later tests in the same
+ process then miss that module's lines in structlog.testing.capture_logs."""
+ log = MagicMock(wraps=service_module.LOG)
+ for name, method in methods.items():
+ setattr(log, name, method)
+ monkeypatch.setattr(service_module, "LOG", log)
+
+
def _make_updated_row(now: datetime | None = None) -> MagicMock:
now = now or datetime.now(UTC)
row = MagicMock()
@@ -202,7 +212,7 @@ async def test_mark_canceled_if_not_final_logs_duration_metrics(
def fake_info(event: str, **kwargs: object) -> None:
info_calls.append((event, dict(kwargs)))
- monkeypatch.setattr(service_module.LOG, "info", fake_info)
+ _patch_service_log(monkeypatch, info=fake_info)
svc = WorkflowService()
result = await svc.mark_workflow_run_as_canceled_if_not_final(workflow_run_id="wr_live")
@@ -276,7 +286,7 @@ async def test_mark_canceled_if_not_final_skips_side_effects_on_terminal_row(
def fake_info(event: str, **kwargs: object) -> None:
info_calls.append(event)
- monkeypatch.setattr(service_module.LOG, "info", fake_info)
+ _patch_service_log(monkeypatch, info=fake_info)
svc = WorkflowService()
result = await svc.mark_workflow_run_as_canceled_if_not_final(workflow_run_id="wr_already_done")
@@ -537,7 +547,7 @@ async def test_cancel_policy_run_retries_failed_terminal_release_without_raising
monkeypatch.setattr(svc, "run_terminal_side_effects", fail_release)
warning = MagicMock()
- monkeypatch.setattr(service_module.LOG, "warning", warning)
+ _patch_service_log(monkeypatch, warning=warning)
# The cancel operation itself succeeded; a failed best-effort release is retried and left
# for the durable recovery owner instead of being raised through the API path.