fix(models): treat empty GenerateContentResponse without prompt feedback as successful

Previously, an empty `candidates` list without `prompt_feedback` resulted in an `UNKNOWN_ERROR`. This change updates the logic to handle such cases as a successful completion with no generated content, which is valid for certain model interactions like tool-driven turns.

Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 916115022

Author: GWeale

src/google/adk/models/llm_response.py

@@ -14,6 +14,7 @@
 
 from __future__ import annotations
 
+import logging
 from typing import Any
 from typing import Optional
 
@@ -216,9 +217,15 @@ def create(
             model_version=generate_content_response.model_version,
         )
       else:
+        # Some model backends can legitimately complete a turn without
+        # candidates (for example, tool-driven UI turns with no text). Treat
+        # this as an empty successful response rather than an unknown error.
+        logging.warning(
+            'Received empty candidates and no prompt feedback in model '
+            'response. Treating as a successful empty response.'
+        )
         return LlmResponse(
-            error_code='UNKNOWN_ERROR',
-            error_message='Unknown error.',
+            content=types.Content(role='model', parts=[]),
             usage_metadata=usage_metadata,
             model_version=generate_content_response.model_version,
         )

tests/unittests/models/test_llm_response.py

@@ -107,6 +107,31 @@ def test_llm_response_create_no_candidates():
   assert response.error_message == 'Prompt blocked for safety'
 
 
+def test_llm_response_create_no_candidates_without_prompt_feedback():
+  """Test LlmResponse.create() for empty successful model responses."""
+  usage_metadata = types.GenerateContentResponseUsageMetadata(
+      prompt_token_count=10,
+      candidates_token_count=0,
+      total_token_count=10,
+  )
+  generate_content_response = types.GenerateContentResponse(
+      candidates=[],
+      usage_metadata=usage_metadata,
+      model_version='gemini-2.5-flash',
+  )
+
+  response = LlmResponse.create(generate_content_response)
+
+  assert response.error_code is None
+  assert response.error_message is None
+  assert response.finish_reason is None
+  assert response.content is not None
+  assert response.content.role == 'model'
+  assert not response.content.parts
+  assert response.usage_metadata == usage_metadata
+  assert response.model_version == 'gemini-2.5-flash'
+
+
 def test_llm_response_create_with_concrete_logprobs_result():
   """Test LlmResponse.create() with detailed logprobs_result containing actual token data."""
   # Create realistic logprobs data