chore: remove tracker references from comments and docstrings

Co-authored-by: George Weale <gweale@google.com> PiperOrigin-RevId: 956763571
2026-07-30 15:20:35 -07:00
parent d776f22c8e
commit 83b71e68a9
35 changed files with 71 additions and 75 deletions
@@ -1008,7 +1008,7 @@ class LlmAgent(BaseAgent, abc.ABC):
    __maybe_save_output_to_state skips them and the text on those events
    is dropped from output_key. Accumulate every non-partial text-bearing
    event from this agent across the model turn so the segments survive
-    in session state. See issue #5590.
+    in session state.

    No-op when accumulation doesn't apply (different author, no
    output_key, output_schema set, partial event, no content, no text).
@@ -426,7 +426,7 @@ def _validate_single_config(
      }

    # Step 3: Additional structural validation
-    # TODO: b/455645705 - Remove once the frontend performs these validations before calling
+    # TODO: Remove once the frontend performs these validations before calling
    # this tool.
    name_warning = _normalize_agent_name_field(config_dict, path)
    structural_validation = _validate_structure(config_dict, path)
@@ -308,8 +308,9 @@ class GeminiLlmConnection(BaseLlmConnection):
    last_grounding_metadata = None
    tool_call_metadata = None
    async with Aclosing(self._gemini_session.receive()) as agen:
-      # TODO(b/440101573): Reuse StreamingResponseAggregator to accumulate
-      # partial content and emit responses as needed.
+      # Pending cleanup: reuse StreamingResponseAggregator to accumulate
+      # partial content and emit responses as needed, once that aggregator
+      # handles the live-connection message shapes.
      async for message in agen:
        logger.debug('Got LLM Live message: %s', message)
        live_session_id = self._gemini_session.session_id
@@ -1870,7 +1870,7 @@ class _SpanRecord:
    with ``GOOGLE_CLOUD_AGENT_ENGINE_ENABLE_TELEMETRY=true``), those
    plugin-owned spans were exported to Cloud Trace alongside the
    framework's real spans — producing a duplicate-span view for
-    every BQAA-instrumented operation.  See haiyuan-eng-google/BQAA-SDK#94.
+    every BQAA-instrumented operation.

    The plugin already tracked all parent / child relationships on
    this internal stack, so the OTel span object was incidental to
@@ -5616,7 +5616,7 @@ class BigQueryAgentAnalyticsPlugin(BasePlugin):
    ``InvocationContext.agent.name`` with no None guard, but ``agent`` is
    legitimately ``None`` for workflow-driven invocations with deterministic
    nodes. Reading it at row-build time then raised ``AttributeError``, which
-    ``@_safe_callback`` swallowed, silently dropping the row (issue #6063).
+    ``@_safe_callback`` swallowed, silently dropping the row.

    Resolution order:

@@ -6563,7 +6563,7 @@ class BigQueryAgentAnalyticsPlugin(BasePlugin):
    try:
      # Capture trace_id BEFORE popping the invocation-root span so
      # that INVOCATION_COMPLETED shares the same trace_id as all
-      # earlier events in this invocation (fixes #4645).
+      # earlier events in this invocation.
      callback_ctx = CallbackContext(invocation_context)
      trace_id = TraceManager.get_trace_id(callback_ctx)

@@ -34,7 +34,6 @@ class MultimodalToolResultsPlugin(BasePlugin):

  Should be removed in favor of directly supporting FunctionResponsePart when these
  are supported outside of computer use tool.
-  For context see: https://github.com/google/adk-python/issues/3064#issuecomment-3463067459
  """

  def __init__(self, name: str = "multimodal_tool_results_plugin"):
@@ -20,8 +20,7 @@ v0 schema, see
 https://github.com/google/adk-python/blob/main/docs/upgrading_from_1_22_0.md.

 The latest schema is defined in `v1.py`. That module uses JSON serialization
-for the EventActions data as well as other fields in the `events` table. See
-https://github.com/google/adk-python/discussions/3605 for more details.
+for the EventActions data as well as other fields in the `events` table.
 """

 from __future__ import annotations
@@ -17,8 +17,6 @@
 This module defines SQLAlchemy models for storing session and event data
 in a relational database with the "events" table using JSON
 serialization for Event data.
-
-See https://github.com/google/adk-python/discussions/3605 for more details.
 """

 from __future__ import annotations
@@ -142,9 +142,8 @@ def trace_agent_invocation(
    agent: Agent from which attributes are gathered.
    ctx: InvocationContext from which attributes are gathered.

-  Inference related fields are not set, due to their planned removal from
-    invoke_agent span:
-  https://github.com/open-telemetry/semantic-conventions/issues/2632
+  Inference related fields are not set, because the OpenTelemetry semantic
+    conventions plan to remove them from the invoke_agent span.

  `gen_ai.agent.id` is not set because currently it's unclear what attributes
    this field should have, specifically:
@@ -296,7 +295,8 @@ def trace_merged_tool_calls(
  span.set_attribute(GEN_AI_TOOL_DESCRIPTION, "(merged tools)")
  span.set_attribute(GEN_AI_TOOL_CALL_ID, response_event_id)

-  # TODO(b/441461932): See if these are still necessary
+  # Pending cleanup: drop these placeholder attributes once no downstream
+  # consumer reads them.
  span.set_attribute("gcp.vertex.agent.tool_call_args", "N/A")
  span.set_attribute("gcp.vertex.agent.event_id", response_event_id)
  try:
@@ -222,7 +222,8 @@ def build_function_declaration(
        )
    )
    # Add response schema only for VERTEX_AI
-    # TODO(b/421991354): Remove this check once the bug is fixed.
+    # Pending cleanup: remove this check once the Gemini API accepts
+    # response_json_schema.
    if variant != GoogleLLMVariant.VERTEX_AI:
      declaration.response_json_schema = None
    return declaration
@@ -1076,7 +1076,7 @@ class TestA2AToEventConverters:


 class TestRoleMappingRegression:
-  """Regression tests for issue #5186: role mapping in A2A→ADK conversion."""
+  """Regression tests for role mapping in A2A→ADK conversion."""

  def setup_method(self):
    """Set up test fixtures."""
@@ -445,7 +445,7 @@ class TestConvertGenaiPartToA2aPart:
  def test_convert_empty_text_part(self):
    """Test that Part(text='') is preserved, not dropped.

-    Regression test for #5341: empty-string text parts are valid and
+    Regression test: empty-string text parts are valid and
    must not fall through to the unsupported-part warning.
    """
    # Arrange
@@ -270,8 +270,6 @@ class TestGeminiContextCacheManager:
  async def test_create_cache_gates_on_prefix_not_full_prompt(self):
    """Cache creation is gated on the cacheable prefix, not the full prompt.

-    Regression test for https://github.com/google/adk-python/issues/5847.
-
    On a long conversation the previous-prompt token count
    (``cacheable_contents_token_count``) can be well above Gemini's 4096-token
    minimum while the cached prefix ``contents[:cache_contents_count]`` is far
@@ -362,7 +362,7 @@ def test_allow_transfer_by_default():
  assert not agent.disallow_transfer_to_peers


-# TODO(b/448114567): Remove TestCanonicalTools once the workaround
+# Pending cleanup: remove TestCanonicalTools once the workaround
 # is no longer needed.
 class TestCanonicalTools:
  """Unit tests for canonical_tools in LlmAgent."""
@@ -57,7 +57,7 @@ def _event(

@pytest.mark.asyncio
 async def test_run_async_accumulates_text_around_tool_calls():
-  """Regression test for issue #5590.
+  """Regression test for dropped output_key text around tool calls.

  Under StreamingMode.SSE with tools, an LlmAgent emits text in several
  non-partial events: some carry text only, others carry text alongside a
@@ -1615,7 +1615,7 @@ class TestRemoteA2aAgentMessageHandling:
  ):
    """Test streaming A2A response handling when content/parts are missing.

-    This verifies the fix for issue #3769 where the code could raise when it
+    This verifies the fix for the case where the code could raise when it
    tried to read parts[0] without checking for empty/missing content.
    """
    mock_a2a_task = create_autospec(A2ATask, instance=True)
@@ -1898,7 +1898,7 @@ class TestRemoteA2aAgentMessageHandling:


 class TestRemoteA2aAgentStreamingArtifactChunks:
-  """Regression tests for chunked artifact streams (#6343)."""
+  """Regression tests for chunked artifact streams."""

  def setup_method(self):
    """Setup test fixtures."""
@@ -1338,7 +1338,7 @@ class TestLiveSessionCallbacks:


 def test_convert_events_preserves_tool_calls_when_skip_summarization():
-  """Regression test for #5410.
+  """Regression test for tool calls dropped from invocation_events.

  When an event has skip_summarization=True, is_final_response() returns True
  even if the event contains function calls.  Previously such an event was
@@ -632,7 +632,7 @@ def test_generate_final_eval_status_doesn_t_throw_on(eval_service):
 async def test_mcp_stdio_agent_no_runtime_error(mocker):
  """Test that LocalEvalService can handle MCP stdio agents without RuntimeError.

-  This is a regression test for GitHub issue #2196:
+  This is a regression test for the reported failure:
  "RuntimeError: Attempted to exit cancel scope in a different task than it was
  entered in"

@@ -491,7 +491,7 @@ def test_evaluate_invocations_invocation_events_format_exact_match(
 ):
  """InvocationEvents intermediate_data format should score 1.0 on exact match.

-  Regression test for #5410: tool_trajectory_avg_score returned 0.0 even when
+  Regression test: tool_trajectory_avg_score returned 0.0 even when
  tool name and args were identical because function-call events with
  skip_summarization=True were incorrectly excluded from invocation_events.
  """
@@ -172,7 +172,7 @@ async def test_preprocess_handles_mixed_tools_and_toolsets():
  assert mock_toolset.process_llm_request_called


-# TODO(b/448114567): Remove the following test_preprocess_with_google_search
+# Pending cleanup: remove the following test_preprocess_with_google_search
 # tests once the workaround is no longer needed.
@pytest.mark.asyncio
 async def test_preprocess_with_google_search_only():
@@ -490,7 +490,7 @@ class _AsyncProcessLlmRequestTool:
      self._on_process(self.name)


-# TODO(b/448114567): Remove the following
+# Pending cleanup: remove the following
 # test_handle_after_model_callback_grounding tests once the workaround
 # is no longer needed.
 def dummy_tool():
@@ -1681,7 +1681,7 @@ def _make_agent_tree():

@pytest.mark.asyncio
 async def test_empty_stop_after_tool_call_surfaces_error_event():
-  """Regression test for empty Gemini turn after a successful tool call (#5631).
+  """Regression test for an empty Gemini turn after a successful tool call.

  Turn 1 returns a function_call which executes successfully, then turn 2
  returns Content(role='model', parts=[]) with finish_reason=STOP and no error.
@@ -207,7 +207,7 @@ def test_data_file_helper_lib_defines_crop():
  assert crop('x' * 100, max_chars=10) == 'x' * 7 + '...'
  assert crop('abcdef', max_chars=2) == 'ab'

-  # Regression for #4011: explore_df raised NameError when crop was undefined.
+  # Regression: explore_df raised NameError when crop was undefined.
  namespace['explore_df'](pd.DataFrame({'a': [1, 2], 'b': ['x', 'y']}))


@@ -1324,7 +1324,7 @@ async def test_adk_function_call_ids_preserved_for_interactions_model():
@pytest.mark.asyncio
 async def test_adk_function_call_ids_preserved_for_anthropic_model():
  """Anthropic ids must round-trip through replay so Claude can match
-  tool_use blocks with their tool_result blocks (issue #5074).
+  tool_use blocks with their tool_result blocks.
  """
  from google.adk.models.anthropic_llm import AnthropicLlm

@@ -1875,8 +1875,8 @@ def test_recover_compacted_function_calls_uses_latest_sibling_response():
 def test_get_contents_recovers_compacted_long_running_call_on_resume():
  """A long-running call compacted before resume is restored during assembly.

-  Reproduces issue #5602: the call and its intermediate placeholder response are
-  summarized away, then the real result arrives on resume. Without recovery,
+  The call and its intermediate placeholder response are summarized away, then
+  the real result arrives on resume. Without recovery,
  assembly raises because the resumed response has no matching call.
  """
  compaction = EventCompaction(
@@ -795,7 +795,7 @@ class TestCallToolInThreadPool:

  @pytest.mark.asyncio
  async def test_sync_tool_returning_none_runs_exactly_once(self):
-    """Regression test for issue #5284.
+    """Regression test for double invocation of a None-returning sync tool.

    A sync FunctionTool whose underlying function returns None must not
    be re-invoked through the run_async fallback path.
@@ -162,7 +162,7 @@ class NonOverriddenBuiltInPlanner(BuiltInPlanner):
 async def test_overridden_subclass_process_planning_response_called():
  """Test that subclasses overriding process_planning_response have it called.

-  Regression test for issue #4133.
+  Regression test: the base implementation used to be called instead.
  """
  planner = OverriddenBuiltInPlanner(thinking_config=types.ThinkingConfig())
  agent = Agent(name='test_agent', planner=planner)
@@ -5019,8 +5019,8 @@ async def test_finish_reason_propagation(
 def test_model_response_to_generate_content_response_no_message_with_finish_reason():
  """Test response with no message but finish_reason returns empty LlmResponse.

-  This test covers issue #3618: when a turn ends with tool calls and no final
-  message, we should return an empty LlmResponse instead of raising ValueError.
+  When a turn ends with tool calls and no final message, we should return an
+  empty LlmResponse instead of raising ValueError.
  """
  response = ModelResponse(
      model="test_model",
@@ -659,7 +659,7 @@ class TestBigQueryAgentAnalyticsPlugin:
      dummy_arrow_schema,
      mock_asyncio_to_thread,
  ):
-    """Regression test for cross-region writes (issue #262).
+    """Regression test for cross-region writes.

    The Storage Write API streaming AppendRows RPC does not
    auto-populate the request-routing header, so writes to a dataset
@@ -2111,7 +2111,7 @@ class TestBigQueryAgentAnalyticsPlugin:
      callback_context,
      dummy_arrow_schema,
  ):
-    """Regression for #6063: None agent falls back to source event author."""
+    """Regression: None agent falls back to source event author."""
    # Workflow-driven invocations leave ``InvocationContext.agent`` as None.
    # Reading ``callback_context.agent_name`` then raised ``AttributeError``,
    # which ``@_safe_callback`` swallowed, silently dropping the BigQuery row.
@@ -2143,7 +2143,7 @@ class TestBigQueryAgentAnalyticsPlugin:
      callback_context,
      dummy_arrow_schema,
  ):
-    """Regression for #6063: callback-only row with no agent writes null."""
+    """Regression: callback-only row with no agent writes null."""
    callback_context._invocation_context.agent = None

    await bq_plugin_inst._log_event(
@@ -2573,7 +2573,7 @@ class TestBigQueryAgentAnalyticsPlugin:
    """Verify no quota_project_id is set when credentials don't provide one.

    This is critical for Workload Identity Federation flows where setting
-    quota_project_id on the client breaks auth token refresh (issue #4370).
+    quota_project_id on the client breaks auth token refresh.
    """
    mock_creds = mock.create_autospec(
        google.auth.credentials.Credentials, instance=True, spec_set=True
@@ -2822,7 +2822,7 @@ class TestBigQueryAgentAnalyticsPlugin:
      self,
      callback_context,
  ):
-    """Regression guard for the duplicate-Cloud-Trace bug (issue #94).
+    """Regression guard for the duplicate-Cloud-Trace bug.

    The plugin must NOT call ``tracer.start_span(...)`` from
    ``push_span`` / ``pop_span``.  Any owned OTel span goes through
@@ -2858,7 +2858,7 @@ class TestBigQueryAgentAnalyticsPlugin:
  async def test_push_pop_does_not_export_spans_through_real_provider(
      self, callback_context
  ):
-    """End-to-end regression guard against #94 with a real OTel
+    """End-to-end guard against duplicate Cloud Trace spans with a real OTel

    provider + in-memory exporter.

@@ -2901,7 +2901,7 @@ class TestBigQueryAgentAnalyticsPlugin:
      assert exporter.get_finished_spans() == (), (
          "Plugin must not export OTel spans; any owned span would"
          " surface as a duplicate in Cloud Trace alongside the"
-          " framework's real spans (issue #94)."
+          " framework's real spans."
      )

    provider.shutdown()
@@ -5861,7 +5861,7 @@ class TestHITLTracing:


 # ==============================================================================
-# TEST CLASS: Span Hierarchy Isolation (Issue #4561)
+# TEST CLASS: Span Hierarchy Isolation
 # ==============================================================================


@@ -6724,7 +6724,7 @@ class TestAnalyticsViews:


 # ==============================================================================
-# Trace-ID Continuity Tests (Issue #4645)
+# Trace-ID Continuity Tests
 # ==============================================================================
 class TestTraceIdContinuity:
  """Tests for trace_id continuity across all events in an invocation.
@@ -6740,9 +6740,9 @@ class TestTraceIdContinuity:
  async def test_trace_id_continuity_no_ambient_span(self, callback_context):
    """All events share one trace_id when no ambient OTel span exists.

-    Simulates the #4645 scenario: OTel IS configured (real TracerProvider)
-    but the Runner's ambient span is NOT present (e.g. Agent Engine,
-    custom runners).
+    Simulates the broken-continuity scenario: OTel IS configured (real
+    TracerProvider) but the Runner's ambient span is NOT present (e.g. Agent
+    Engine, custom runners).
    """
    from opentelemetry.sdk.trace import TracerProvider as SdkProvider
    from opentelemetry.sdk.trace.export import SimpleSpanProcessor
@@ -7267,7 +7267,7 @@ class TestStackLeakSafety:
  def test_clear_stack_does_not_export_spans(self, callback_context):
    """``clear_stack()`` clears the internal records but does NOT

-    export any OTel spans (issue #94 regression guard).
+    export any OTel spans (duplicate-Cloud-Trace regression guard).

    Pre-fix, ``clear_stack()`` called ``record.span.end()`` for every
    owned record, which delivered the now-finished span to whatever
@@ -7304,11 +7304,11 @@ class TestStackLeakSafety:
      result = bigquery_agent_analytics_plugin._span_records_ctx.get()
      assert result == []

-      # Still no exported spans — the regression guard for #94.
+      # Still no exported spans — the duplicate-Cloud-Trace guard.
      assert exporter.get_finished_spans() == (), (
          "clear_stack() must not export OTel spans; any owned span"
          " would surface as a duplicate in Cloud Trace alongside the"
-          " framework's real spans (issue #94)."
+          " framework's real spans."
      )

    provider.shutdown()
@@ -8226,7 +8226,7 @@ class TestA2AInteractionLogging:


 # ================================================================
-# TEST CLASS: Dataset location handling (Issue #5476)
+# TEST CLASS: Dataset location handling
 # ================================================================
 class TestDatasetLocationHandling:
  """Tests that BQ client is created without a default location.
@@ -8324,7 +8324,7 @@ class TestDatasetLocationHandling:


 # ================================================================
-# TEST CLASS: Fork detection after pickle (Issue #86 / PR #5528)
+# TEST CLASS: Fork detection after pickle
 # ================================================================
 class TestForkDetectionAfterPickle:
  """Tests that unpickled plugins do not false-positive fork detection."""
@@ -8394,7 +8394,7 @@ class TestForkDetectionAfterPickle:


 # ================================================================
-# TEST CLASS: GCS offload unit mismatch fix (Issue #5561)
+# TEST CLASS: GCS offload unit mismatch fix
 # ================================================================
 class TestOffloadUnitSeparation:
  """Tests that byte-based inline limit and character-based truncation
@@ -8608,7 +8608,7 @@ class TestOffloadUnitSeparation:


 # ================================================================
-# TEST CLASS: AGENT_RESPONSE logging (Issue #87)
+# TEST CLASS: AGENT_RESPONSE logging
 # ================================================================
 class TestAgentResponseLogging:
  """Tests that final agent response events are captured correctly."""
@@ -222,12 +222,12 @@ def _create_function_response_content(name: str, call_id: str) -> types.Content:
 async def test_filter_preserves_function_call_response_pairs():
  """Tests that function_call and function_response pairs are kept together.

-  This tests the fix for issue #4027 where filtering could create orphaned
+  This tests the fix for the case where filtering could create orphaned
  function_response messages without their corresponding function_call.
  """
  plugin = ContextFilterPlugin(num_invocations_to_keep=2)

-  # Simulate conversation from issue #4027:
+  # Simulate the reported conversation:
  # user -> model -> user -> model(function_call) -> user(function_response)
  # -> model -> user -> model(function_call) -> user(function_response)
  contents = [
@@ -14,7 +14,7 @@

 """Tests for on_agent_error_callback and on_run_error_callback.

-Validates RFC #5044: agent-level and runner-level error callbacks.
+Validates the agent-level and runner-level error callback contract.
 """

 import asyncio
@@ -57,8 +57,8 @@ class CustomErrorExtractionPlugin(ReflectAndRetryToolPlugin):
    return None


-# Inheriting from IsolatedAsyncioTestCase ensures consistent behavior.
-# See https://github.com/pytest-dev/pytest-asyncio/issues/1039
+# Inheriting from IsolatedAsyncioTestCase ensures consistent behavior, because
+# pytest-asyncio's own event-loop scoping varies across versions.
 class TestReflectAndRetryToolPlugin(IsolatedAsyncioTestCase):
  """Comprehensive tests for ReflectAndRetryToolPlugin focusing on behavior."""

@@ -2014,7 +2014,7 @@ def test_database_session_service_visible_in_module_namespace():
  """DatabaseSessionService must be in dir() so Sphinx autodoc renders it.

  It is imported lazily via module __getattr__, so without an explicit
-  __dir__ it drops out of the generated API reference (issue #4331).
+  __dir__ it drops out of the generated API reference.
  """
  import google.adk.sessions as sessions_module

@@ -2131,7 +2131,7 @@ async def test_database_session_service_requires_one_argument():
 async def test_database_session_service_sqlite_file_timestamp_read_after_reopen(
    tmp_path,
 ):
-  """Regression test for #6352 (SQLite REAL-affinity timestamp reads)."""
+  """Regression test for SQLite REAL-affinity timestamp reads."""
  # SQLite REAL-affinity columns can end up storing raw Unix epoch floats
  # instead of the text format SQLAlchemy's DateTime type normally writes
  # (for example, if the row was written by a different code path than the
@@ -1170,7 +1170,7 @@ async def test_append_event():
 async def test_append_event_strips_unsupported_part_metadata(
    mock_api_client_instance: MockAsyncClient,
 ) -> None:
-  """part_metadata must not reach the Sessions API (#6014).
+  """part_metadata must not reach the Sessions API.

  ``Part.part_metadata`` is a Gemini Developer API-only field; the Vertex AI
  Agent Engine Sessions ``appendEvent`` API rejects it with 400 INVALID_ARGUMENT
@@ -1212,7 +1212,7 @@ async def test_append_event_strips_unsupported_part_metadata(
 async def test_append_event_with_part_metadata_round_trips(
    mock_api_client_instance: MockAsyncClient,
 ) -> None:
-  """Reconstruction side of #6014: an event carrying part_metadata appends and
+  """Reconstruction side: an event carrying part_metadata appends and
  reads back without error. part_metadata is dropped (unsupported on Vertex),
  but the session round-trips and the part text is preserved.
  """
@@ -2384,7 +2384,7 @@ EXPECTED_METRICS_V2: dict[str, frozenset[MetricPoint]] = {
 # ``error.type`` across the duration metrics (see the metric constants below).
 #
 # ``google.genai`` collapses every 4xx into ``ClientError`` / 5xx into
-# ``ServerError``, so before b/534739207 every such failure reported
+# ``ServerError``, so historically every such failure reported
 # ``error.type=ClientError``. ADK now uses the provider's HTTP status code
 # (e.g. ``429``), falling back to the exception class name for non-API errors
 # (e.g. ``ValueError``).
@@ -2728,7 +2728,7 @@ ALL_CASES: list[FunctionalTestCase] = [
            metric_points=EXPECTED_METRICS_V2,
        ),
    ),
-    # Inference failures (b/534739207): the mock raises before responding, so
+    # Inference failures: the mock raises before responding, so
    # the scenario aborts and the failure surfaces on ``error.type``. A 429
    # surfaces its HTTP status code ``429`` (not a blanket ``ClientError``); a
    # plain ``ValueError`` falls back to the class name.
@@ -1909,9 +1909,9 @@ class TestRunnerInferAgentOrigin:
  def test_infer_agent_origin_no_false_positive_for_direct_llm_agent(self):
    """Test that using LlmAgent directly doesn't trigger mismatch warning.

-    Regression test for GitHub issue #3143: Users who instantiate LlmAgent
-    directly and run from a directory that is a parent of the ADK installation
-    were getting false positive 'App name mismatch' warnings.
+    Regression test: users who instantiate LlmAgent directly and run from a
+    directory that is a parent of the ADK installation were getting false
+    positive 'App name mismatch' warnings.

    This also verifies that _infer_agent_origin returns None for ADK internal
    modules (google.adk.*).
@@ -680,7 +680,7 @@ class TestBuildFunctionDeclarationWithJsonSchema:
        get_data, variant=GoogleLLMVariant.GEMINI_API
    )

-    # GEMINI_API should not have response_json_schema due to bug b/421991354
+    # GEMINI_API should not have response_json_schema: the API rejects it.
    assert decl.response_json_schema is None

  @pytest.mark.parametrize(
@@ -31,7 +31,7 @@ function_call_no_schema = Part.from_function_call(
 grounding_metadata = types.GroundingMetadata(web_search_queries=['test query'])


-# TODO(b/448114567): Remove test_grounding_metadata_ tests once the workaround
+# Pending cleanup: remove test_grounding_metadata_ tests once the workaround
 # is no longer needed.


@@ -2139,7 +2139,7 @@ async def test_request_input_resume_after_earlier_invocation_completed(
 ):
  """A completed earlier invocation must not block a later HITL resume.

-  Regression test for #6497. The first invocation finishes on the `finish`
+  Regression test. The first invocation finishes on the `finish`
  branch. The second invocation takes the `clarify` branch and pauses for
  input. When the replay sequence was built from every event in the session,
  the terminal `finish` event of the first invocation entered the sequence