Retry prompt model response template (#2008)

Author: hovi

pydantic_ai_slim/pydantic_ai/messages.py

@@ -501,7 +501,10 @@ class RetryPromptPart:
     def model_response(self) -> str:
         """Return a string message describing why the retry is requested."""
         if isinstance(self.content, str):
-            description = self.content
+            if self.tool_name is None:
+                description = f'Validation feedback:\n{self.content}'
+            else:
+                description = self.content
         else:
             json_errors = error_details_ta.dump_json(self.content, exclude={'__all__': {'ctx'}}, indent=2)
             description = f'{len(self.content)} validation errors: {json_errors.decode()}'

tests/models/test_instrumented.py

@@ -240,6 +240,7 @@ async def test_instrumented_model(capfire: CaptureLogfire):
             {
                 'body': {
                     'content': """\
+Validation feedback:
 retry_prompt2
 
 Fix the errors and try again.\
@@ -619,6 +620,7 @@ async def test_instrumented_model_attributes_mode(capfire: CaptureLogfire):
                                 {
                                     'event.name': 'gen_ai.user.message',
                                     'content': """\
+Validation feedback:
 retry_prompt2
 
 Fix the errors and try again.\

tests/test_agent.py

@@ -301,7 +301,7 @@ def return_tuple(_: list[ModelMessage], info: AgentInfo) -> ModelResponse:
                 parts=[
                     ToolCallPart(tool_name='final_result', args='{"response": ["foo", "bar"]}', tool_call_id=IsStr())
                 ],
-                usage=Usage(requests=1, request_tokens=72, response_tokens=8, total_tokens=80),
+                usage=Usage(requests=1, request_tokens=74, response_tokens=8, total_tokens=82),
                 model_name='function:return_tuple:',
                 timestamp=IsNow(tz=timezone.utc),
             ),
@@ -930,7 +930,7 @@ def call_tool(messages: list[ModelMessage], info: AgentInfo) -> ModelResponse:
             ),
             ModelResponse(
                 parts=[TextPart(content='Mexico City')],
-                usage=Usage(requests=1, request_tokens=68, response_tokens=5, total_tokens=73),
+                usage=Usage(requests=1, request_tokens=70, response_tokens=5, total_tokens=75),
                 model_name='function:call_tool:',
                 timestamp=IsDatetime(),
             ),
@@ -1584,7 +1584,7 @@ def call_tool(messages: list[ModelMessage], info: AgentInfo) -> ModelResponse:
             ),
             ModelResponse(
                 parts=[TextPart(content='{"city": "Mexico City"}')],
-                usage=Usage(requests=1, request_tokens=68, response_tokens=11, total_tokens=79),
+                usage=Usage(requests=1, request_tokens=70, response_tokens=11, total_tokens=81),
                 model_name='function:call_tool:',
                 timestamp=IsDatetime(),
             ),