fix: use Modality enum for RunConfig.response_modalities

Merge #4889

## Description

Fixes #4869

`RunConfig.response_modalities` is typed as `Optional[list[str]]`, but it gets assigned to `LiveConnectConfig.response_modalities` which expects `Optional[list[Modality]]` (an enum from `google-genai`). This type mismatch causes Pydantic to emit a `PydanticSerializationUnexpectedValue` warning on every live streaming session, polluting production logs.

## Changes

- **`run_config.py`**: Changed `response_modalities` type from `Optional[list[str]]` to `Optional[list[types.Modality]]`
- **`runners.py`**: Updated two string literal `'AUDIO'` usages to `types.Modality.AUDIO` for consistency

This is backwards compatible — `Modality` is a string enum, so Pydantic automatically coerces string values like `'AUDIO'` to `Modality.AUDIO`.

## Testing Plan

Added 4 new unit tests in `test_run_config.py`:

- `test_response_modalities_accepts_enum` — enum values are stored correctly
- `test_response_modalities_coerces_string_to_enum` — string values still work (backwards compat)
- `test_response_modalities_coerces_lowercase_string_to_enum` — lowercase string values also work (backwards compat)
- `test_response_modalities_serialization_no_warning` — `LiveConnectConfig.model_dump()` produces no Pydantic warnings

Co-authored-by: Liang Wu <wuliang@google.com>
COPYBARA_INTEGRATE_REVIEW=#4889 from brucearctor:fix/response-modalities-type e67f715
PiperOrigin-RevId: 934659835

Author: 2 people

src/google/adk/agents/run_config.py

@@ -196,7 +196,7 @@ class RunConfig(BaseModel):
   speech_config: Optional[types.SpeechConfig] = None
   """Speech configuration for the live agent."""
 
-  response_modalities: Optional[list[str]] = None
+  response_modalities: Optional[list[types.Modality]] = None
   """The output modalities. If not set, it's default to AUDIO."""
 
   avatar_config: Optional[types.AvatarConfig] = None

src/google/adk/runners.py

@@ -1554,7 +1554,7 @@ async def run_live(
     # Some native audio models requires the modality to be set. So we set it to
     # AUDIO by default.
     if run_config.response_modalities is None:
-      run_config.response_modalities = ['AUDIO']
+      run_config.response_modalities = [types.Modality.AUDIO]
     if session is None and (user_id is None or session_id is None):
       raise ValueError(
           'Either session or user_id and session_id must be provided.'
@@ -2021,7 +2021,7 @@ def _new_invocation_context_for_live(
     # For live multi-agents system, we need model's text transcription as
     # context for the transferred agent.
     if hasattr(self.agent, 'sub_agents') and self.agent.sub_agents:
-      if 'AUDIO' in run_config.response_modalities:
+      if types.Modality.AUDIO in run_config.response_modalities:
         if not run_config.output_audio_transcription:
           run_config.output_audio_transcription = (
               types.AudioTranscriptionConfig()

tests/unittests/agents/test_run_config.py

@@ -15,6 +15,7 @@
 import sys
 from unittest.mock import ANY
 from unittest.mock import patch
+import warnings
 
 from google.adk.agents.run_config import RunConfig
 from google.genai import types
@@ -67,6 +68,37 @@ def test_audio_transcription_configs_are_not_shared_between_instances():
   )
 
 
+def test_response_modalities_accepts_enum():
+  config = RunConfig(response_modalities=[types.Modality.AUDIO])
+  assert config.response_modalities == [types.Modality.AUDIO]
+  assert isinstance(config.response_modalities[0], types.Modality)
+
+
+def test_response_modalities_coerces_string_to_enum():
+  config = RunConfig(response_modalities=["AUDIO"])
+  assert config.response_modalities == [types.Modality.AUDIO]
+  assert isinstance(config.response_modalities[0], types.Modality)
+
+
+def test_response_modalities_coerces_lowercase_string_to_enum():
+  config = RunConfig(response_modalities=["audio"])
+  assert config.response_modalities == [types.Modality.AUDIO]
+  assert isinstance(config.response_modalities[0], types.Modality)
+
+
+def test_response_modalities_serialization_no_warning():
+  config = RunConfig(response_modalities=[types.Modality.AUDIO])
+  live_config = types.LiveConnectConfig()
+  live_config.response_modalities = config.response_modalities
+  with warnings.catch_warnings(record=True) as w:
+    warnings.simplefilter("always")
+    live_config.model_dump()
+    pydantic_warnings = [
+        x for x in w if "PydanticSerializationUnexpectedValue" in str(x.message)
+    ]
+    assert len(pydantic_warnings) == 0
+
+
 def test_avatar_config_initialization():
   custom_avatar = types.CustomizedAvatar(
       image_mime_type="image/jpeg", image_data=b"image_bytes"