Adjust prompt to new model

codegpt/conversation.py

@@ -4,11 +4,11 @@
 
 
 class Model(Enum):
-    GPT_4 = "gpt-4"
+    GPT_4 = "gpt-4-0613"
 
 
 def argparse_model_type(model_str):
-    if model_str.lower() == "gpt-4":
+    if model_str.lower() == "gpt-4-0613":
         return Model.GPT_4
     else:
         raise argparse.ArgumentTypeError(f"Model {model_str} is not supported.")

codegpt/messages_builder.py

@@ -22,6 +22,7 @@ def system_prompt(context: Context):
 Because of technical limitations, Sky only has a limited context. So it might be that previous output from user or terminal are cut off.
 Format: `action(parameter)`, Sky should never directly speak to user output (the output will be written by a software, the user will not see).
 The parameters are never quoted neither with single nor double quotes, so Sky will never output an action like this: `action('parameter')` but rather like this: `action(parameter)`.
+Sky will always output an action at the end of its answer.
 Action 1: ask_for_refinement(question) 
 - Sky asks the user for a refinement of the software. Where ask_for_refinement is the action and question represents the question Sky asks the user.
 - The output will be the response of the user.
@@ -36,7 +37,8 @@ def system_prompt(context: Context):
 - If Sky runs into the same error multiple times, Sky will try to fix the error. A quick self reflection, where Sky thinks step by step and evaluates solutions to the problem.
 Action 3: end_programming_session()
 - Sky ends the programming session.
-- Can only be executed after the user is happy with the software.
+- This action can only be executed by sky if the user is satisfied with the software.
+- The user has to answer with a ask_for_refinement action.
 ---
 Sky will program software iteratively in small steps.
 If possible, sky will use Test Driven Development automatically without being explicitly asked to do so.