update code

metagpt/roles/minecraft/action_developer.py

@@ -2,8 +2,6 @@
 # @Date    : 2023/9/23 12:45
 # @Author  : stellahong (stellahong@fuzhi.ai)
 # @Desc    :
-import copy
-
 from metagpt.logs import logger
 from metagpt.roles.minecraft.minecraft_base import Minecraft as Base
 from metagpt.schema import Message, HumanMessage, SystemMessage
@@ -40,14 +38,13 @@ class ActionDeveloper(Base):
         # Initialize actions specific to the Action role
         self._init_actions([GenerateActionCode])
         
-        
         # Set events or actions the ActionAgent should watch or be aware of
         # 需要根据events进行自己chest_observation的更新
         self._watch([RetrieveSkills])
         self.rollout_num_iter = 0
         self.task_max_retries = 4
         self.finish_state = len(self._actions)
-        self.critic_reviewer = None  # self._rc.env.roles["Task Reviewer"]
+        self.critic_reviewer = None  # self._rc.env.roles["Task Reviewer"]      
     
     def render_system_message(self, skills=[], *args, **kwargs):
         """
@@ -208,6 +205,7 @@ class ActionDeveloper(Base):
     async def handle_add_new_skills(
             self, task, program_name, program_code, skills, *args, **kwargs
     ):
+        skills = self.game_memory.skills
         skill_desp = self.game_memory.skill_desp
         new_skills_info = await AddNewSkills().run(
             task, program_name, program_code, skills, skill_desp
@@ -231,11 +229,9 @@ class ActionDeveloper(Base):
         context = self.game_memory.context
         
         # 更新生成的代码和对应程序名称
-        code, program_name = await GenerateActionCode().run(
+        program_code, code, program_name = await GenerateActionCode().run(
             human_msg, system_msg, *args, **kwargs
         )
-        # logger.warning(type(code))
-        # logger.info(f"Code is Here:{code}")
         
         if code is not None:
             # fixme：若有独立的mc code执行入口函数，使用独立的函数
@@ -298,6 +294,8 @@ class ActionDeveloper(Base):
             "success": self.game_memory.runtime_status,
         }
         logger.info(f"info is {info}")
+        
+        self.perform_game_info_callback(program_code, self.game_memory.update_program_code)
         self.perform_game_info_callback(code, self.game_memory.update_code)
         self.perform_game_info_callback(
             program_name, self.game_memory.update_program_name
@@ -306,11 +304,10 @@ class ActionDeveloper(Base):
         return system_msg, human_msg, 0, done, info
     
     async def generate_action_code(self, human_msg, system_msg, *args, **kwargs):
-        code, program_name = await GenerateActionCode().run(
+        program_code, code, program_name = await GenerateActionCode().run(
             human_msg, system_msg, *args, **kwargs
         )
-        # logger.warning(type(code))
-        # logger.info(f"Code is Here:{code}")
+        self.perform_game_info_callback(program_code, self.game_memory.update_program_code)
         self.perform_game_info_callback(code, self.game_memory.update_code)
         self.perform_game_info_callback(
             program_name, self.game_memory.update_program_name

metagpt/roles/minecraft/curriculum_agent.py

@@ -18,29 +18,29 @@ class CurriculumDesigner(Base):
     """
     CurriculumDesigner is the automatic curriculum in paper, refer to the code voyager/agents/curriculum.py
     """
-
+    
     def __init__(
-        self,
-        name: str = "David",
-        profile: str = "Expertise in minecraft task design and curriculum development.",
-        goal: str = " Collect and integrate learner feedback to improve and refine educational content and pathways",
-        constraints: str = "Limited budget and resources for the development of educational content and technology tools.",
+            self,
+            name: str = "David",
+            profile: str = "Expertise in minecraft task design and curriculum development.",
+            goal: str = " Collect and integrate learner feedback to improve and refine educational content and pathways",
+            constraints: str = "Limited budget and resources for the development of educational content and technology tools.",
     ) -> None:
         super().__init__(name, profile, goal, constraints)
         # Initialize actions specific to the Action role
         self._init_actions([DesignTask, DesignCurriculum])
-
+        
         # Set events or actions the ActionAgent should watch or be aware of
         self._watch([PlayerActions, DesignTask])
         logger.info(self._actions)
         self.finish_state = len(self._actions)
-
+    
     def render_curriculum_observation(self, *, events, chest_observation):
         """
         Returns: observation for curriculum
         Refer to @ https://github.com/MineDojo/Voyager/blob/main/voyager/agents/curriculum.py
         """
-
+        
         assert events[-1][0] == "observe", "Last event must be observe"
         event = events[-1][1]
         biome = event["status"]["biome"]
@@ -54,31 +54,31 @@ class CurriculumDesigner(Base):
         equipment = event["status"]["equipment"]
         inventory_used = event["status"]["inventoryUsed"]
         inventory = event["inventory"]
-
+        
         if not any(
-            "dirt" in block
-            or "log" in block
-            or "grass" in block
-            or "sand" in block
-            or "snow" in block
-            for block in voxels
+                "dirt" in block
+                or "log" in block
+                or "grass" in block
+                or "sand" in block
+                or "snow" in block
+                for block in voxels
         ):
             biome = "underground"
-
+        
         other_blocks = ", ".join(
             list(
                 set(block_records).difference(set(voxels).union(set(inventory.keys())))
             )
         )
-
+        
         other_blocks = other_blocks if other_blocks else "None"
-
+        
         nearby_entities = (
             ", ".join([k for k, v in sorted(entities.items(), key=lambda x: x[1])])
             if entities
             else "None"
         )
-
+        
         completed_tasks = (
             ", ".join(self.game_memory.completed_tasks)
             if self.game_memory.completed_tasks
@@ -89,18 +89,18 @@ class CurriculumDesigner(Base):
             if self.game_memory.failed_tasks
             else "None"
         )
-
+        
         # filter out optional inventory items if required
         if (
-            self.game_memory.progress
-            < self.game_memory.warm_up["optional_inventory_items"]
+                self.game_memory.progress
+                < self.game_memory.warm_up["optional_inventory_items"]
         ):
             inventory = {
                 k: v
                 for k, v in inventory.items()
                 if self.game_memory.core_inv_items_regex.search(k) is not None
             }
-
+        
         observation = {
             "context": "",
             "biome": f"Biome: {biome}\n\n",
@@ -118,25 +118,31 @@ class CurriculumDesigner(Base):
             "failed_tasks": f"Failed tasks that are too hard: {failed_tasks}\n\n",
         }
         return observation
-
+    
     # --------------------------------Design Task Prepare---------------------------------------
-    def render_design_task_human_message(
-        self, events, chest_observation, *args, **kwargs
+    async def render_design_task_human_message(
+            self, events, chest_observation, *args, **kwargs
     ):
         """
         Returns: observation for curriculum
         Refer to @ https://github.com/MineDojo/Voyager/blob/main/voyager/agents/curriculum.py
         """
-
+        
         content = ""
         warm_up = self.game_memory.mf_instance.warm_up
         observation = self.render_curriculum_observation(
             events=events, chest_observation=chest_observation
         )
         if self.game_memory.progress >= warm_up["context"]:
-            questions, answers = DesignCurriculum.generate_qa(
+            # if self.game_memory.progress >= 0: # TEST ONLY
+            human_msg = self.render_design_curriculum_human_message(
                 events=events, chest_observation=chest_observation
+            ).content
+            system_msg = [self.render_design_curriculum_system_message().content]
+            questions, answers = await self.curriculum_design_action.generate_qa(
+                events=events, human_msg=human_msg, system_msg=system_msg
             )
+            logger.debug(f"Generate_qa result is HERE: Ques: {questions}, Ans: {answers}")
             i = 1
             for question, answer in zip(questions, answers):
                 if "Answer: Unknown" in answer or "language model" in answer:
@@ -146,7 +152,7 @@ class CurriculumDesigner(Base):
                 i += 1
                 if i > 5:
                     break
-
+        
         for key in CURRICULUM_OB:
             if self.game_memory.progress >= warm_up[key]:
                 if warm_up[key] != 0:
@@ -155,20 +161,20 @@ class CurriculumDesigner(Base):
                     should_include = True
                 if should_include:
                     content += observation[key]
-
+        
         logger.info(f"Curriculum Agent human message\n{content}")
         return HumanMessage(content=content)
-
+    
     def render_design_task_system_message(self, *args, **kwargs):
         return SystemMessage(content=load_prompt("curriculum"))
-
-    def encapsule_design_task_message(self, events, chest_observation, *args, **kwargs):
-        human_msg = self.render_design_task_human_message(
+    
+    async def encapsule_design_task_message(self, events, chest_observation, *args, **kwargs):
+        human_msg = await self.render_design_task_human_message(
             events=events, chest_observation=chest_observation, *args, **kwargs
         )
         system_msg = self.render_design_task_system_message(*args, **kwargs)
         return {"system_msg": [system_msg.content], "human_msg": human_msg.content}
-
+    
     def generate_task_if_inventory_full(self, events, chest_observation):
         """
         TODO: Try if this could be done with prompt
@@ -187,15 +193,15 @@ class CurriculumDesigner(Base):
         else:
             task = "Craft 1 chest"
         return task
-
+    
     # -----------------------------------------------------------------------------------------
-
+    
     # --------------------------------Design Curriculum Prepare--------------------------------
     def render_design_curriculum_system_message(self, *args, **kwargs):
         return SystemMessage(content=load_prompt("curriculum_qa_step1_ask_questions"))
-
+    
     def render_design_curriculum_human_message(
-        self, events, chest_observation, *args, **kwargs
+            self, events, chest_observation, *args, **kwargs
     ):
         observation = self.render_curriculum_observation(
             events=events, chest_observation=chest_observation
@@ -204,16 +210,16 @@ class CurriculumDesigner(Base):
         for key in CURRICULUM_OB:
             content += observation[key]
         return HumanMessage(content=content)
-
+    
     def encapsule_design_curriculum_message(
-        self, events, chest_observation, *args, **kwargs
+            self, events, chest_observation, *args, **kwargs
     ):
         human_msg = self.render_design_curriculum_human_message(
             events=events, chest_observation=chest_observation, *args, **kwargs
         )
         system_msg = self.render_design_curriculum_system_message(*args, **kwargs)
         return {"system_msg": [system_msg.content], "human_msg": human_msg.content}
-
+    
     def generate_context_if_inventory_full(self, events, chest_observation):
         """
         TODO: Try if this could be done with prompt
@@ -244,9 +250,9 @@ class CurriculumDesigner(Base):
         else:
             context = "Craft 1 chest with 8 planks of any kind of wood."
         return context
-
+    
     # -----------------------------------------------------------------------------------------
-
+    
     async def handle_task_design(self, human_msg, system_msg, *args, **kwargs):
         """
         Args:
@@ -260,7 +266,7 @@ class CurriculumDesigner(Base):
         events = self.game_memory.event
         chest_observation = self.game_memory.chest_observation
         inventoryUsed = events[-1][1]["status"]["inventoryUsed"]
-
+        
         if self.game_memory.progress == 0:
             task = self.game_memory.current_task
         elif inventoryUsed >= 33:
@@ -270,12 +276,12 @@ class CurriculumDesigner(Base):
         else:
             task = await DesignTask().run(human_msg, system_msg, *args, **kwargs)
         logger.info(f"Handle_task_design result is Here: {task}")
-
+        
         self.perform_game_info_callback(task, self.game_memory.update_task)
         return Message(
             content=f"{task}", instruct_content="task_design", role=self.profile
         )
-
+    
     async def handle_curriculum_design(self, human_msg, system_msg, *args, **kwargs):
         """
         refer to the context generation in voyager
@@ -292,7 +298,7 @@ class CurriculumDesigner(Base):
         chest_observation = self.game_memory.chest_observation
         inventoryUsed = events[-1][1]["status"]["inventoryUsed"]
         task = self.game_memory.current_task
-
+        
         if self.game_memory.progress == 0:
             context = self.game_memory.context
         elif inventoryUsed >= 33:
@@ -300,37 +306,34 @@ class CurriculumDesigner(Base):
                 self, events=events, chest_observation=chest_observation
             )
         else:
-            context = await DesignCurriculum().run(
+            context = await self.DesignCurriculum().run(
                 task, human_msg, system_msg, *args, **kwargs
             )
+        
         self.perform_game_info_callback(context, self.game_memory.update_context)
         return Message(
             content=f"{context}",
             instruct_content="curriculum_design",
             role=self.profile,
         )
-
+    
     async def _act(self) -> Message:
         todo = self._rc.todo
         logger.debug(f"Todo is {todo}")
-        self.maintain_actions(todo) 
+        self.maintain_actions(todo)
+        
         # 获取最新的游戏周边环境信息
         # events = await self._obtain_events()
         events = self.game_memory.event
         chest_observation = self.game_memory.chest_observation
-
-        # DesignCurriculum.set_qa_cache(self.game_memory.qa_cache)
-
-        # msg = self._rc.memory.get(k=1)[0]
-        # query = msg.content
-
-        design_task_message = self.encapsule_design_task_message(
+        
+        design_task_message = await self.encapsule_design_task_message(
             events, chest_observation
         )
         design_curriculum_message = self.encapsule_design_curriculum_message(
             events, chest_observation
         )
-
+        
         handler_map = {
             DesignTask: self.handle_task_design,
             DesignCurriculum: self.handle_curriculum_design,
@@ -345,5 +348,5 @@ class CurriculumDesigner(Base):
             msg.round_id = self.round_id
             self._publish_message(msg)
             return msg
-
+        
         raise ValueError(f"Unknown todo type: {type(todo)}")

metagpt/roles/minecraft/skill_manager.py

@@ -19,41 +19,40 @@ from metagpt.utils.minecraft import load_prompt
 @agent_registry.register("skill_manager")
 class SkillManager(Base):
     def __init__(
-        self,
-        name: str = "John",
-        profile: str = "Skills Management Specialist",
-        goal: str = "To oversee and optimize the acquisition, development, and utilization of skills within the organization, ensuring workforce competence and efficiency.",
-        constraints: str = "Resource allocation, training budgets, and alignment with organizational goals.",
+            self,
+            name: str = "John",
+            profile: str = "Skills Management Specialist",
+            goal: str = "To oversee and optimize the acquisition, development, and utilization of skills within the organization, ensuring workforce competence and efficiency.",
+            constraints: str = "Resource allocation, training budgets, and alignment with organizational goals.",
     ) -> None:
         super().__init__(name, profile, goal, constraints)
-
+        
         # Initialize actions specific to the SkillManager role
-        self._init_actions([RetrieveSkills, GenerateSkillDescription, AddNewSkills]) #AddNewSkills])#先去掉add
+        self._init_actions([RetrieveSkills, GenerateSkillDescription, AddNewSkills])
         
         # Set events or actions the SkillManager should watch or be aware of
         self._watch(
             [DesignCurriculum, GenerateActionCode, RetrieveSkills, GenerateSkillDescription]
         )
-        
         self.finish_state = len(self._actions)
-
+    
     def encapsule_message(self, program_code, program_name, *args, **kwargs):
         system_msg = self.render_system_message(load_prompt("skill"))
         human_msg = self.render_human_message(
             program_code + "\n\n" + f"The main function is `{program_name}`."
         )
         return {"system_msg": [system_msg.content], "human_msg": human_msg.content}
-
+    
     async def retrieve_skills(self, query, skills, *args, **kwargs):
         retrieve_skills = await RetrieveSkills().run(query, skills)
         logger.info(f"Render Action Agent system message with {len(retrieve_skills)} skills")
         self.perform_game_info_callback(retrieve_skills, self.game_memory.update_retrieve_skills)
-        return Message(content=f"{retrieve_skills}", instruct_content="retrieve_skills", 
+        return Message(content=f"{retrieve_skills}", instruct_content="retrieve_skills",
                        role=self.profile, send_to=agent_registry.entries["action_developer"]()._setting.name)
         # return Message(
         #     content=f"{skills}", instruct_content="retrieve_skills", role=self.profile
         # )  # Unit test only
-
+    
     async def generate_skill_descp(self, human_msg, system_msg, *args, **kwargs):
         program_name = self.game_memory.program_name
         desp = await GenerateSkillDescription().run(program_name, human_msg, system_msg)
@@ -63,16 +62,16 @@ class SkillManager(Base):
             instruct_content="generate_skill_descp",
             role=self.profile,
         )
-
+    
     async def handle_add_new_skills(
-        self, task, program_name, program_code, skills, *args, **kwargs
+            self, task, program_name, program_code, skills, *args, **kwargs
     ):
         if not self.game_memory.runtime_status:
             return Message(
-            content="",
-            instruct_content="handle_add_new_skills",
-            role=self.profile,
-        )
+                content="",
+                instruct_content="handle_add_new_skills",
+                role=self.profile,
+            )
         
         skill_desp = self.game_memory.skill_desp
         new_skills_info = await AddNewSkills().run(
@@ -84,43 +83,41 @@ class SkillManager(Base):
             instruct_content="handle_add_new_skills",
             role=self.profile,
         )
-
+    
     async def _act(self) -> Message:
         todo = self._rc.todo
         logger.debug(f"Todo is {todo}")
-        self.maintain_actions(todo)    
+        self.maintain_actions(todo)
         # 获取最新的游戏周边信息
+        
         context = self.game_memory.context
         task = self.game_memory.current_task
         
         code = self.game_memory.code
         self.perform_game_info_callback(self.game_memory.event, self.game_memory.summarize_chatlog)
         event_summary = self.game_memory.event_summary
-        try:
-            program_code = code["program_code"] # TODO: Handle code is None, cuz first round DesignCurriculum(code is None) trigger this 
-        except (KeyError, TypeError):
-            program_code = ""
-
+        program_code = self.game_memory.program_code
+        
         program_name = self.game_memory.program_name
         skills = self.game_memory.skills
-
+        
         # msg = self._rc.memory.get(k=1)[0]
-
+        
         retrieve_skills_message_step1 = {"query": context, "skills": skills}
         logger.info(f"check query {context}")
         logger.info(f"check event summary {event_summary}")
         
         retrieve_skills_message_step2 = {"query": context + "\n\n" + event_summary, "skills": skills}
-
+        
         generate_skill_message = self.encapsule_message(program_code, program_name)
-
+        
         add_new_skills_message = {
             "task": task,
             "program_name": program_name,
             "program_code": program_code,
             "skills": skills,
         }
-
+        
         handler_map = {
             DesignCurriculum: self.retrieve_skills,
             RetrieveSkills: self.retrieve_skills,
@@ -139,10 +136,10 @@ class SkillManager(Base):
                 msg = await handler(**generate_skill_message)
             else:
                 msg = await handler(**add_new_skills_message)
-
+            
             msg.cause_by = type(todo)
             msg.round_id = self.round_id
             self._publish_message(msg)
             return msg
-
+        
         raise ValueError(f"Unknown todo type: {type(todo)}")