fix(eval): plumb App through LocalEvalService to fix App.plugins bypass

Closes the loop on https://github.com/google/adk-python/issues/<TBD>:
when a project wraps its root agent in `App(root_agent=..., plugins=[...])`
and runs `adk eval`, the registered plugins (e.g.,
`BigQueryAgentAnalyticsPlugin`) now fire on every invocation just like
they do for `adk web` / `adk run`. Same applies to `App.context_cache_config`
and `App.resumability_config`, which now ride along automatically.

Changes:
* `LocalEvalService.__init__` accepts an optional `app` keyword argument
  and forwards it to `_generate_inferences_from_root_agent` for each
  eval case.
* `cli_tools_click.cli_eval` resolves the `App` via `get_app_or_root_agent`
  and passes it to `LocalEvalService`.
* `cli_optimize` (GEPA prompt optimization) also routes through
  `LocalEvalService` but currently constructs it inside `LocalEvalSampler`
  with no `app` argument; bringing the optimize path under App-plugin
  coverage is a separate, narrower follow-up and is intentionally not
  included here.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: saifer82

src/google/adk/cli/cli_tools_click.py

@@ -823,8 +823,8 @@ def cli_eval(
     from ..evaluation.simulation.user_simulator_provider import UserSimulatorProvider
     from .cli_eval import _collect_eval_results
     from .cli_eval import _collect_inferences
+    from .cli_eval import get_app_or_root_agent
     from .cli_eval import get_default_metric_info
-    from .cli_eval import get_root_agent
     from .cli_eval import parse_and_get_evals_to_run
     from .cli_eval import pretty_print_eval_result
   except ModuleNotFoundError as mnf:
@@ -834,7 +834,7 @@ def cli_eval(
   print(f"Using evaluation criteria: {eval_config}")
   eval_metrics = get_eval_metrics_from_config(eval_config)
 
-  root_agent = get_root_agent(agent_module_file_path)
+  app, root_agent = get_app_or_root_agent(agent_module_file_path)
   app_name = os.path.basename(agent_module_file_path)
   agents_dir = os.path.dirname(agent_module_file_path)
   eval_sets_manager = None
@@ -940,6 +940,7 @@ def cli_eval(
         eval_set_results_manager=eval_set_results_manager,
         user_simulator_provider=user_simulator_provider,
         metric_evaluator_registry=metric_evaluator_registry,
+        app=app,
     )
 
     inference_results = asyncio.run(

src/google/adk/evaluation/local_eval_service.py

@@ -25,6 +25,7 @@
 from typing_extensions import override
 
 from ..agents.base_agent import BaseAgent
+from ..apps.app import App
 from ..artifacts.base_artifact_service import BaseArtifactService
 from ..artifacts.in_memory_artifact_service import InMemoryArtifactService
 from ..errors.not_found_error import NotFoundError
@@ -123,8 +124,20 @@ def __init__(
       session_id_supplier: Callable[[], str] = _get_session_id,
       user_simulator_provider: UserSimulatorProvider = UserSimulatorProvider(),
       memory_service: Optional[BaseMemoryService] = None,
+      *,
+      app: Optional[App] = None,
   ):
+    """Initializes a LocalEvalService.
+
+    Args:
+      app: Optional `App` that wraps `root_agent`. When provided, eval runs
+        are executed through a Runner built from the App, so `app.plugins`,
+        `app.context_cache_config`, and `app.resumability_config` are
+        honored during inference. When None, the legacy bare-agent path is
+        used.
+    """
     self._root_agent = root_agent
+    self._app = app
     self._eval_sets_manager = eval_sets_manager
     metric_evaluator_registry = (
         metric_evaluator_registry or DEFAULT_METRIC_EVALUATOR_REGISTRY
@@ -491,6 +504,7 @@ async def _perform_inference_single_eval_item(
                 session_service=self._session_service,
                 artifact_service=self._artifact_service,
                 memory_service=self._memory_service,
+                app=self._app,
             )
         )
 

tests/unittests/evaluation/test_local_eval_service.py

@@ -19,6 +19,7 @@
 from typing import Optional
 
 from google.adk.agents.llm_agent import LlmAgent
+from google.adk.apps.app import App
 from google.adk.errors.not_found_error import NotFoundError
 from google.adk.evaluation.base_eval_service import EvaluateConfig
 from google.adk.evaluation.base_eval_service import EvaluateRequest
@@ -791,3 +792,74 @@ def test_copy_invocation_rubrics_to_actual_invocations():
   _copy_invocation_rubrics_to_actual_invocations(expected, actual)
   assert actual[0].rubrics == [rubric1]
   assert actual[1].rubrics == [rubric2]
+
+
+@pytest.mark.asyncio
+async def test_perform_inference_forwards_app_to_evaluation_generator(
+    dummy_agent, mock_eval_sets_manager, mocker
+):
+  """LocalEvalService passes its `app` through to _generate_inferences_from_root_agent."""
+  app = App(name="test_app", root_agent=dummy_agent)
+
+  eval_case = EvalCase(eval_id="case-1", conversation=[])
+  mock_eval_sets_manager.get_eval_set.return_value = EvalSet(
+      eval_set_id="set-1",
+      eval_cases=[eval_case],
+  )
+
+  mock_generate = mocker.patch(
+      "google.adk.evaluation.local_eval_service.EvaluationGenerator._generate_inferences_from_root_agent",
+      new=mocker.AsyncMock(return_value=[]),
+  )
+
+  service = LocalEvalService(
+      root_agent=dummy_agent,
+      eval_sets_manager=mock_eval_sets_manager,
+      app=app,
+  )
+
+  request = InferenceRequest(
+      app_name="test_app",
+      eval_set_id="set-1",
+      eval_case_ids=["case-1"],
+      inference_config=InferenceConfig(),
+  )
+  async for _ in service.perform_inference(inference_request=request):
+    pass
+
+  mock_generate.assert_awaited_once()
+  assert mock_generate.await_args.kwargs["app"] is app
+
+
+@pytest.mark.asyncio
+async def test_perform_inference_passes_none_when_no_app(
+    dummy_agent, mock_eval_sets_manager, mocker
+):
+  """When LocalEvalService has no `app`, it forwards None (legacy behavior)."""
+  eval_case = EvalCase(eval_id="case-1", conversation=[])
+  mock_eval_sets_manager.get_eval_set.return_value = EvalSet(
+      eval_set_id="set-1",
+      eval_cases=[eval_case],
+  )
+
+  mock_generate = mocker.patch(
+      "google.adk.evaluation.local_eval_service.EvaluationGenerator._generate_inferences_from_root_agent",
+      new=mocker.AsyncMock(return_value=[]),
+  )
+
+  service = LocalEvalService(
+      root_agent=dummy_agent,
+      eval_sets_manager=mock_eval_sets_manager,
+  )
+
+  request = InferenceRequest(
+      app_name="test_app",
+      eval_set_id="set-1",
+      eval_case_ids=["case-1"],
+      inference_config=InferenceConfig(),
+  )
+  async for _ in service.perform_inference(inference_request=request):
+    pass
+
+  mock_generate.assert_awaited_once()
+  assert mock_generate.await_args.kwargs["app"] is None