fix a bug in model eval

rdagent/components/coder/data_science/model/prompts.yaml

@@ -78,6 +78,8 @@ model_eval:
         You should evaluate the code given by the user. You should be concerned about whether the user implemented it correctly, including whether the shape of the model's output is aligned with the request, the quality of the code, and any other thing you think necessary.
         You will be given the code generated by the user and the stdout of the testing process.
         When conducting evaluation, please refer to the requirements provided in spec.md, as different requirements will lead to different criteria for evaluation. 
+
+        Only if there is "Model code test passed successfully." in the stdout, then the model is considered successfu, or else there must be some issues with the model.
     
         Please respond with your feedback in the following JSON format and order:
         ```json