Synchronize Instances from AgentScope (#29)

browser_use/agent_browser/browser_agent.py

@@ -21,8 +21,7 @@ from agentscope.token import TokenCounterBase, OpenAITokenCounter
 _BROWSER_AGENT_DEFAULT_SYS_PROMPT = (
     "You are a helpful browser automation assistant. "
     "You can navigate websites, take screenshots, and interact with web pages."
-    "Always describe what you see and meta_planner_agent"
+    "Always describe what you see and plan your next steps clearly. "
-    " your next steps clearly. "
     "When taking actions, explain what you're doing and why."
 )
 _BROWSER_AGENT_REASONING_PROMPT = (
@@ -30,7 +29,7 @@ _BROWSER_AGENT_REASONING_PROMPT = (
     "The snapshot (and screenshot) of the current webpage is (are) given "
     "below. Since you can only view the latest webpage, "
     "you must promptly summarize current status, record required data, "
-    "and meta_planner_agent your next steps."
+    "and plan your next steps."
 )
 
 

browser_use/agent_browser/main.py

@@ -3,12 +3,12 @@
 import asyncio
 import os
 
-from agentscope.agent import UserAgent
 from agentscope.formatter import DashScopeChatFormatter
-from agentscope.mcp import StdIOStatefulClient
 from agentscope.memory import InMemoryMemory
 from agentscope.model import DashScopeChatModel
 from agentscope.tool import Toolkit
+from agentscope.mcp import StdIOStatefulClient
+from agentscope.agent import UserAgent
 
 from browser_agent import BrowserAgent  # pylint: disable=C0411
 
@@ -69,7 +69,7 @@ if __name__ == "__main__":
     print(
         "The browser agent will use "
         "playwright-mcp (https://github.com/microsoft/playwright-mcp)."
-        "Make sure the MCP server can be installed "
+        "Make sure the MCP server is can be install "
         "by `npx @playwright/mcp@latest`",
     )
 

deep_research/agent_deep_research/built_in_prompt/promptmodule.py

@@ -20,7 +20,7 @@ class SubtasksDecomposition(BaseModel):
     working_plan: str = Field(
         description=(
             "A logically ordered step-by-step working "
-            "meta_planner_agent (3-5 steps), each step starting with "
+            "plan (3-5 steps), each step starting with "
             "its number (1., 2., etc), including both "
             "core and expansion steps. Expanded steps "
             "should be clearly marked with (EXPANSION) "
@@ -85,7 +85,7 @@ class ReflectFailure(BaseModel):
             "subtask needs to be rephrased due "
             "to a design flaw or misunderstanding. "
             "If rephrasing is needed, provide the "
-            "modified working meta_planner_agent with only the "
+            "modified working plan with only the "
             "inappropriate subtask replaced by its "
             "improved version."
         ),
@@ -95,14 +95,14 @@ class ReflectFailure(BaseModel):
                 "properties": {
                     "need_rephrase": {
                         "type": "boolean",
-                        "description": "Set to 'true' if the failed"
+                        "description": "Set to 'true' if the failed subtask "
-                        " subtask needs to be rephrased due to a design "
+                        "needs to be rephrased due to a design "
                         "flaw or misunderstanding; otherwise, 'false'.",
                     },
                     "rephrased_plan": {
                         "type": "string",
-                        "description": "The modified working "
+                        "description": "The modified working plan "
-                        "meta_planner_agent with only the inappropriate "
+                        "with only the inappropriate "
                         "subtask replaced by its improved version. If no "
                         "rephrasing is needed, provide an empty string.",
                     },

deep_research/agent_deep_research/deep_research_agent.py

@@ -8,9 +8,7 @@ import asyncio
 from typing import Type, Optional, Any, Tuple
 from datetime import datetime
 from copy import deepcopy
-
 import shortuuid
-
 from pydantic import BaseModel
 
 from built_in_prompt.promptmodule import (
@@ -25,6 +23,7 @@ from utils import (
     get_dynamic_tool_call_json,
     get_structure_output,
 )
+
 from agentscope import logger, setup_logger
 from agentscope.mcp import StatefulClientBase
 from agentscope.agent import ReActAgent
@@ -43,6 +42,7 @@ from agentscope.message import (
     ToolResultBlock,
 )
 
+
 _DEEP_RESEARCH_AGENT_DEFAULT_SYS_PROMPT = "You're a helpful assistant."
 
 _LOG_DIR = os.path.join(os.path.dirname(__file__), "log")
@@ -196,7 +196,7 @@ class DeepResearchAgent(ReActAgent):
             SubTaskItem(objective=self.user_query),
         )
 
-        # Identify the expected output and generate a meta_planner_agent
+        # Identify the expected output and generate a plan
         await self.decompose_and_expand_subtask()
         msg.content += (
             f"\nExpected Output:\n{self.current_subtask[0].knowledge_gaps}"
@@ -214,7 +214,7 @@ class DeepResearchAgent(ReActAgent):
             )
 
         for _ in range(self.max_iters):
-            # Generate the working meta_planner_agent first
+            # Generate the working plan first
             if not self.current_subtask[-1].working_plan:
                 await self.decompose_and_expand_subtask()
 
@@ -497,23 +497,20 @@ class DeepResearchAgent(ReActAgent):
 
     async def decompose_and_expand_subtask(self) -> ToolResponse:
         """Identify the knowledge gaps of the current subtask and generate a
-        working meta_planner_agent by subtask decomposition.
+        working plan by subtask decomposition. The working plan includes
-        The working meta_planner_agent includes
         necessary steps for task completion and expanded steps.
 
         Returns:
             ToolResponse:
-                The knowledge gaps and working meta_planner_agent
+                The knowledge gaps and working plan of the current subtask
-                of the current subtask in JSON format.
+                in JSON format.
         """
         if len(self.current_subtask) <= self.max_depth:
             decompose_sys_prompt = self.prompt_dict["decompose_sys_prompt"]
 
             previous_plan = ""
             for i, subtask in enumerate(self.current_subtask):
-                previous_plan += (
+                previous_plan += f"The {i}-th plan: {subtask.working_plan}\n"
-                    f"The {i}-th meta_planner_agent: {subtask.working_plan}\n"
-                )
             previous_plan_inst = self.prompt_dict[
                 "previous_plan_inst"
             ].format_map(
@@ -716,7 +713,7 @@ class DeepResearchAgent(ReActAgent):
 
     async def summarize_intermediate_results(self) -> ToolResponse:
         """Summarize the intermediate results into a report when a step
-        in working meta_planner_agent is completed.
+        in working plan is completed.
 
         Returns:
             ToolResponse:
@@ -740,9 +737,7 @@ class DeepResearchAgent(ReActAgent):
                         "user",
                         self.prompt_dict["summarize_hint"].format_map(
                             {
-                                "meta_planner_agent": self.current_subtask[
+                                "plan": self.current_subtask[-1].working_plan,
-                                    -1
-                                ].working_plan,
                             },
                         ),
                         role="user",
@@ -953,12 +948,11 @@ class DeepResearchAgent(ReActAgent):
 
     async def reflect_failure(self) -> ToolResponse:
         """Reflect on the failure of the action and determine to rephrase
-        the meta_planner_agent or deeper decompose the current step.
+        the plan or deeper decompose the current step.
 
         Returns:
             ToolResponse:
-                The reflection about meta_planner_agent
+                The reflection about plan rephrasing and subtask decomposition.
-                rephrasing and subtask decomposition.
         """
         reflect_sys_prompt = self.prompt_dict["reflect_sys_prompt"]
         conversation_history = ""
@@ -974,7 +968,7 @@ class DeepResearchAgent(ReActAgent):
         reflect_inst = self.prompt_dict["reflect_instruction"].format_map(
             {
                 "conversation_history": conversation_history,
-                "meta_planner_agent": self.current_subtask[-1].working_plan,
+                "plan": self.current_subtask[-1].working_plan,
             },
         )
         try:

deep_research/agent_deep_research/main.py

@@ -3,14 +3,14 @@
 import asyncio
 import os
 
+from deep_research_agent import DeepResearchAgent
+
 from agentscope import logger
 from agentscope.formatter import DashScopeChatFormatter
-from agentscope.mcp import StdIOStatefulClient
 from agentscope.memory import InMemoryMemory
-from agentscope.message import Msg
 from agentscope.model import DashScopeChatModel
-
+from agentscope.message import Msg
-from deep_research_agent import DeepResearchAgent
+from agentscope.mcp import StdIOStatefulClient
 
 
 async def main(user_query: str) -> None:

deep_research/agent_deep_research/utils.py

@@ -8,6 +8,7 @@ from pydantic import BaseModel
 
 from agentscope.tool import Toolkit, ToolResponse
 
+
 TOOL_RESULTS_MAX_WORDS = 5000
 
 
@@ -219,7 +220,7 @@ def load_prompt_dict() -> dict:
 
     prompt_dict["reasoning_prompt"] = (
         "## Current Subtask:\n{objective}\n"
-        "## Working Plan:\n{meta_planner_agent}\n"
+        "## Working Plan:\n{plan}\n"
         "{knowledge_gap}\n"
         "## Research Depth:\n{depth}"
     )
@@ -281,17 +282,14 @@ def load_prompt_dict() -> dict:
 
     prompt_dict["summarize_hint"] = (
         "Based on your work history above, examine which step in the "
-        "following working meta_planner_agent has been "
+        "following working plan has been completed. Mark the completed "
-        "completed. Mark the completed "
         "step with [DONE] at the end of its line (e.g., k. step k [DONE]) "
         "and leave the uncompleted steps unchanged. You MUST return only "
-        "the updated meta_planner_agent, preserving exactly "
+        "the updated plan, preserving exactly the same format as the "
-        "the same format as the "
+        "original plan. Do not include any explanations, reasoning, "
-        "original meta_planner_agent. Do not include any "
-        "explanations, reasoning, "
         "or section headers such as '## Working Plan:', just output the"
-        "updated meta_planner_agent itself."
+        "updated plan itself."
-        "\n\n## Working Plan:\n{meta_planner_agent}"
+        "\n\n## Working Plan:\n{plan}"
     )
 
     prompt_dict["summarize_inst"] = (
@@ -307,19 +305,17 @@ def load_prompt_dict() -> dict:
         "following report that consolidates and summarizes the essential "
         "findings:\n {intermediate_report}\n\n"
         "Such report has been saved to the {report_path}. "
-        "I will now **proceed to the next item** "
+        "I will now **proceed to the next item** in the working plan."
-        "in the working meta_planner_agent."
     )
 
     prompt_dict["save_report_hint"] = (
-        "The milestone results of the current "
+        "The milestone results of the current item in working plan "
-        "item in working meta_planner_agent "
         "are summarized into the following report:\n{intermediate_report}"
     )
 
     prompt_dict["reflect_instruction"] = (
         "## Work History:\n{conversation_history}\n"
-        "## Working Plan:\n{meta_planner_agent}\n"
+        "## Working Plan:\n{plan}\n"
     )
 
     prompt_dict["subtask_complete_hint"] = (

games/game_werewolves/game.py

@@ -2,31 +2,37 @@
 # pylint: disable=too-many-branches, too-many-statements, no-name-in-module
 """A werewolf game implemented by agentscope."""
 import numpy as np
-from agentscope.agent import ReActAgent
+
-from agentscope.pipeline import MsgHub, fanout_pipeline, sequential_pipeline
-from prompt import EnglishPrompts as Prompts
 from utils import (
-    MAX_DISCUSSION_ROUND,
-    MAX_GAME_ROUND,
-    EchoAgent,
-    Players,
     majority_vote,
     names_to_str,
+    EchoAgent,
+    MAX_GAME_ROUND,
+    MAX_DISCUSSION_ROUND,
+    Players,
 )
-
 from structured_model import (
     DiscussionModel,
-    WitchResurrectModel,
-    get_hunter_model,
-    get_poison_model,
-    get_seer_model,
     get_vote_model,
+    get_poison_model,
+    WitchResurrectModel,
+    get_seer_model,
+    get_hunter_model,
 )
+from prompt import EnglishPrompts as Prompts
 
 # Uncomment the following line to use Chinese prompts
 # from prompt import ChinesePrompts as Prompts
 
 
+from agentscope.agent import ReActAgent
+from agentscope.pipeline import (
+    MsgHub,
+    sequential_pipeline,
+    fanout_pipeline,
+)
+
+
 moderator = EchoAgent()
 
 
@@ -113,7 +119,7 @@ async def werewolves_game(agents: list[ReActAgent]) -> None:
                         names_to_str(players.current_alive),
                     ),
                 ),
-                name="game_werewolves",
+                name="werewolves",
             ) as werewolves_hub:
                 # Discussion
                 n_werewolves = len(players.werewolves)

games/game_werewolves/main.py

@@ -4,15 +4,16 @@
 import asyncio
 import os
 
+from game import werewolves_game
+
 from agentscope.agent import ReActAgent
 from agentscope.formatter import DashScopeMultiAgentFormatter
 from agentscope.model import DashScopeChatModel
 from agentscope.session import JSONSession
-from game import werewolves_game
 
 
 def get_official_agents(name: str) -> ReActAgent:
-    """Get the official game_werewolves game agents."""
+    """Get the official werewolves game agents."""
     agent = ReActAgent(
         name=name,
         sys_prompt=f"""You're a werewolf game player named {name}.
@@ -21,9 +22,9 @@ def get_official_agents(name: str) -> ReActAgent:
 Your target is to win the game with your teammates as much as possible.
 
 # GAME RULES
-- In werewolf game, players are divided into three game_werewolves, three villagers, one seer, one hunter and one witch.
+- In werewolf game, players are divided into three werewolves, three villagers, one seer, one hunter and one witch.
     - Werewolves: kill one player each night, and must hide identity during the day.
-    - Villagers: ordinary players without special abilities, try to identify and eliminate game_werewolves.
+    - Villagers: ordinary players without special abilities, try to identify and eliminate werewolves.
         - Seer: A special villager who can check one player's identity each night.
         - Witch: A special villager with two one-time-use potions: a healing potion to save a player from being killed at night, and a poison to eliminate one player at night.
         - Hunter: A special villager who can take one player down with them when they are eliminated.
@@ -39,18 +40,18 @@ Your target is to win the game with your teammates as much as possible.
 # GAME GUIDANCE
 - Try your best to win the game with your teammates, tricks, lies, and deception are all allowed, e.g. pretending to be a different role.
 - During discussion, don't be political, be direct and to the point.
-- The day phase voting provides important clues. For example, the game_werewolves may vote together, attack the seer, etc.
+- The day phase voting provides important clues. For example, the werewolves may vote together, attack the seer, etc.
 ## GAME GUIDANCE FOR WEREWOLF
 - Seer is your greatest threat, who can check one player's identity each night. Analyze players' speeches, find out the seer and eliminate him/her will greatly increase your chances of winning.
-- In the first night, making random choices is common for game_werewolves since no information is available.
+- In the first night, making random choices is common for werewolves since no information is available.
 - Pretending to be other roles (seer, witch or villager) is a common strategy to hide your identity and mislead other villagers in the day phase.
 - The outcome of the night phase provides important clues. For example, if witch uses the healing or poison potion, if the dead player is hunter, etc. Use this information to adjust your strategy.
 ## GAME GUIDANCE FOR SEER
-- Seer is very important to villagers, exposing yourself too early may lead to being targeted by game_werewolves.
+- Seer is very important to villagers, exposing yourself too early may lead to being targeted by werewolves.
 - Your ability to check one player's identity is crucial.
 - The outcome of the night phase provides important clues. For example, if witch uses the healing or poison potion, if the dead player is hunter, etc. Use this information to adjust your strategy.
 ## GAME GUIDANCE FOR WITCH
-- Witch has two powerful potions, use them wisely to protect key villagers or eliminate suspected game_werewolves.
+- Witch has two powerful potions, use them wisely to protect key villagers or eliminate suspected werewolves.
 - The outcome of the night phase provides important clues. For example, if the dead player is hunter, etc. Use this information to adjust your strategy.
 ## GAME GUIDANCE FOR HUNTER
 - Using your ability in day phase will expose your role (since only hunter can take one player down)

games/game_werewolves/prompt.py

@@ -108,13 +108,13 @@ class EnglishPrompts:
 
     to_all_wolf_win = (
         "There are {n_alive} players alive, and {n_werewolves} of them are "
-        "game_werewolves. "
+        "werewolves. "
-        "The game is over and game_werewolves win🐺🎉!"
+        "The game is over and werewolves win🐺🎉!"
         "In this game, the true roles of all players are: {true_roles}"
     )
 
     to_all_village_win = (
-        "All the game_werewolves have been eliminated."
+        "All the werewolves have been eliminated."
         "The game is over and villagers win🏘️🎉!"
         "In this game, the true roles of all players are: {true_roles}"
     )