feat: Expose the Python code run by the code interpreter in the logs

Co-authored-by: Eliza Huang <heliza@google.com>
PiperOrigin-RevId: 826521391

Author: lizzij

src/google/adk/code_executors/agent_engine_sandbox_code_executor.py

@@ -125,6 +125,7 @@ def execute_code(
             input_data=input_data,
         )
     )
+    logger.debug('Executed code:\n```\n%s\n```', code_execution_input.code)
     saved_files = []
     stdout = ''
     stderr = ''

src/google/adk/code_executors/container_code_executor.py

@@ -131,6 +131,7 @@ def execute_code(
         ['python3', '-c', code_execution_input.code],
         demux=True,
     )
+    logger.debug('Executed code:\n```\n%s\n```', code_execution_input.code)
 
     if exec_result.output and exec_result.output[0]:
       output = exec_result.output[0].decode('utf-8')

src/google/adk/code_executors/gke_code_executor.py

@@ -162,6 +162,7 @@ def execute_code(
       logger.info(
           f"Submitted Job '{job_name}' to namespace '{self.namespace}'."
       )
+      logger.debug("Executing code:\n```\n%s\n```", code_execution_input.code)
       return self._watch_job_completion(job_name)
 
     except ApiException as e:

src/google/adk/code_executors/unsafe_local_code_executor.py

@@ -16,6 +16,7 @@
 
 from contextlib import redirect_stdout
 import io
+import logging
 import re
 from typing import Any
 
@@ -27,6 +28,8 @@
 from .code_execution_utils import CodeExecutionInput
 from .code_execution_utils import CodeExecutionResult
 
+logger = logging.getLogger('google_adk.' + __name__)
+
 
 def _prepare_globals(code: str, globals_: dict[str, Any]) -> None:
   """Prepare globals for code execution, injecting __name__ if needed."""
@@ -60,6 +63,7 @@ def execute_code(
       invocation_context: InvocationContext,
       code_execution_input: CodeExecutionInput,
   ) -> CodeExecutionResult:
+    logger.debug('Executing code:\n```\n%s\n```', code_execution_input.code)
     # Execute the code.
     output = ''
     error = ''

src/google/adk/code_executors/vertex_ai_code_executor.py

@@ -152,6 +152,7 @@ def execute_code(
         code_execution_input.input_files,
         code_execution_input.execution_id,
     )
+    logger.debug('Executed code:\n```\n%s\n```', code_execution_input.code)
 
     # Save output file as artifacts.
     saved_files = []
@@ -187,11 +188,13 @@ def execute_code(
         )
 
     # Collect the final result.
-    return CodeExecutionResult(
+    result = CodeExecutionResult(
         stdout=code_execution_result.get('execution_result', ''),
         stderr=code_execution_result.get('execution_error', ''),
         output_files=saved_files,
     )
+    logger.debug('Code execution result: %s', result)
+    return result
 
   def _execute_code_interpreter(
       self,

src/google/adk/flows/llm_flows/_code_execution.py

@@ -20,6 +20,7 @@
 import copy
 import dataclasses
 import datetime
+import logging
 import os
 import re
 from typing import AsyncGenerator
@@ -47,6 +48,8 @@
 if TYPE_CHECKING:
   from ...models.llm_request import LlmRequest
 
+logger = logging.getLogger('google_adk.' + __name__)
+
 
 @dataclasses.dataclass
 class DataFileUtil:
@@ -245,6 +248,7 @@ async def _run_pre_processor(
             ),
         ),
     )
+    logger.debug('Executed code:\n```\n%s\n```', code_str)
     # Update the processing results to code executor context.
     code_executor_context.update_code_execution_result(
         invocation_context.invocation_id,
@@ -352,6 +356,7 @@ async def _run_post_processor(
           ),
       ),
   )
+  logger.debug('Executed code:\n```\n%s\n```', code_str)
   code_executor_context.update_code_execution_result(
       invocation_context.invocation_id,
       code_str,

tests/unittests/flows/llm_flows/test_code_execution.py

@@ -20,7 +20,9 @@
 from unittest.mock import patch
 
 from google.adk.agents.llm_agent import Agent
+from google.adk.code_executors.base_code_executor import BaseCodeExecutor
 from google.adk.code_executors.built_in_code_executor import BuiltInCodeExecutor
+from google.adk.code_executors.code_execution_utils import CodeExecutionResult
 from google.adk.flows.llm_flows._code_execution import response_processor
 from google.adk.models.llm_response import LlmResponse
 from google.genai import types
@@ -105,3 +107,44 @@ async def test_builtin_code_executor_image_artifact_creation(mock_datetime):
       == f'Saved as artifact: {expected_filename2}. '
   )
   assert not llm_response.content.parts[2].inline_data
+
+
+@pytest.mark.asyncio
+@patch('google.adk.flows.llm_flows._code_execution.logger')
+async def test_logs_executed_code(mock_logger):
+  """Test that the response processor logs the code it executes."""
+  mock_code_executor = MagicMock(spec=BaseCodeExecutor)
+  mock_code_executor.code_block_delimiters = [('```python\n', '\n```')]
+  mock_code_executor.error_retry_attempts = 2
+  mock_code_executor.stateful = False
+  mock_code_executor.execute_code.return_value = CodeExecutionResult(
+      stdout='hello'
+  )
+
+  agent = Agent(name='test_agent', code_executor=mock_code_executor)
+  invocation_context = await testing_utils.create_invocation_context(
+      agent=agent, user_content='test message'
+  )
+  invocation_context.artifact_service = MagicMock()
+  invocation_context.artifact_service.save_artifact = AsyncMock()
+
+  llm_response = LlmResponse(
+      content=types.Content(
+          parts=[
+              types.Part(text='Here is some code:'),
+              types.Part(text='```python\nprint("hello")\n```'),
+          ]
+      )
+  )
+
+  _ = [
+      event
+      async for event in response_processor.run_async(
+          invocation_context, llm_response
+      )
+  ]
+
+  mock_code_executor.execute_code.assert_called_once()
+  mock_logger.debug.assert_called_once_with(
+      'Executed code:\n```\n%s\n```', 'print("hello")'
+  )