beswarm 0.1.61__tar.gz → 0.1.63__tar.gz

{beswarm-0.1.61 → beswarm-0.1.63}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: beswarm
-Version: 0.1.61
+Version: 0.1.63
 Summary: MAS
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm/aient/setup.py

@@ -4,7 +4,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="aient",
-    version="1.1.16",
+    version="1.1.17",
     description="Aient: The Awakening of Agent.",
     long_description=Path.open(Path("README.md"), encoding="utf-8").read(),
     long_description_content_type="text/markdown",

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm/aient/src/aient/core/request.py

@@ -4,7 +4,7 @@ import httpx
 import base64
 import urllib.parse
 
-from .models import RequestModel
+from .models import RequestModel, Message
 from .utils import (
     c3s,
     c3o,
@@ -50,7 +50,12 @@ async def get_gemini_payload(request, engine, provider, api_key=None):
     systemInstruction = None
     system_prompt = ""
     function_arguments = None
-    for msg in request.messages:
+
+    try:
+        request_messages = [Message(role="user", content=request.prompt)]
+    except:
+        request_messages = request.messages
+    for msg in request_messages:
         if msg.role == "assistant":
             msg.role = "model"
         tool_calls = None
@@ -104,9 +109,10 @@ async def get_gemini_payload(request, engine, provider, api_key=None):
         elif msg.role == "system":
             content[0]["text"] = re.sub(r"_+", "_", content[0]["text"])
             system_prompt = system_prompt + "\n\n" + content[0]["text"]
-    systemInstruction = {"parts": [{"text": system_prompt}]}
+    if system_prompt.strip():
+        systemInstruction = {"parts": [{"text": system_prompt}]}
 
-    if any(off_model in original_model for off_model in gemini_max_token_65k_models):
+    if any(off_model in original_model for off_model in gemini_max_token_65k_models) or original_model == "gemini-2.0-flash-preview-image-generation":
         safety_settings = "OFF"
     else:
         safety_settings = "BLOCK_NONE"
@@ -160,6 +166,7 @@ async def get_gemini_payload(request, engine, provider, api_key=None):
         'top_logprobs',
         'response_format',
         'stream_options',
+        'prompt',
     ]
     generation_config = {}
 
@@ -214,6 +221,12 @@ async def get_gemini_payload(request, engine, provider, api_key=None):
         else:
             payload["generationConfig"]["maxOutputTokens"] = 8192
 
+        if original_model == "gemini-2.0-flash-preview-image-generation":
+            payload["generationConfig"]["response_modalities"] = [
+                "Text",
+                "Image",
+            ]
+
     if "gemini-2.5" in original_model:
         payload["generationConfig"]["thinkingConfig"] = {
             "includeThoughts": True,
@@ -241,7 +254,7 @@ async def get_gemini_payload(request, engine, provider, api_key=None):
             if key == request.model:
                 for k, v in value.items():
                     payload[k] = v
-            elif all(_model not in request.model.lower() for _model in ["gemini", "gpt", "claude"]):
+            elif all(_model not in request.model.lower() for _model in ["gemini", "gpt", "claude", "deepseek"]) and "-" not in key:
                 payload[key] = value
 
     return url, headers, payload

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm/aient/src/aient/core/response.py

@@ -36,6 +36,7 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model):
         candidatesTokenCount = 0
         totalTokenCount = 0
         parts_json = ""
+        image_base64 = ""
         # line_index = 0
         # last_text_line = 0
         # if "thinking" in model:
@@ -53,6 +54,8 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model):
                 if line and '\"finishReason\": \"' in line:
                     if "stop" not in line.lower():
                         logger.error(f"finishReason: {line}")
+                    sse_string = await generate_sse_response(timestamp, model, stop="stop")
+                    yield sse_string
                     is_finish = True
                 if is_finish and '\"promptTokenCount\": ' in line:
                     json_data = parse_json_safely( "{" + line + "}")
@@ -67,17 +70,25 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model):
                 if (line and '"parts": [' in line or parts_json != "") and is_finish == False:
                     parts_json += line
                 if parts_json != "" and line and '],' == line.strip():
-                    tmp_parts_json =  "{" + parts_json.split("}        ]      },")[0].strip().rstrip("}], ").replace("\n", "\\n").lstrip("{") + "}]}"
+                    # tmp_parts_json =  "{" + parts_json.split("}        ]      },")[0].strip().rstrip("}], ").replace("\n", "\\n").lstrip("{") + "}]}"
+                    tmp_parts_json =  "{" + parts_json.split("}        ]      },")[0].strip().rstrip("}], ").replace("\n", "\\n").lstrip("{")
+                    if "inlineData" in tmp_parts_json:
+                        tmp_parts_json = tmp_parts_json + "}}]}"
+                    else:
+                        tmp_parts_json = tmp_parts_json + "}]}"
                     try:
                         json_data = json.loads(tmp_parts_json)
 
                         content = safe_get(json_data, "parts", 0, "text", default="")
+                        b64_json = safe_get(json_data, "parts", 0, "inlineData", "data", default="")
+                        if b64_json:
+                            image_base64 = b64_json
 
                         is_thinking = safe_get(json_data, "parts", 0, "thought", default=False)
                         if is_thinking:
                             sse_string = await generate_sse_response(timestamp, model, reasoning_content=content)
                             yield sse_string
-                        else:
+                        elif not image_base64:
                             sse_string = await generate_sse_response(timestamp, model, content=content)
                             yield sse_string
                     except json.JSONDecodeError:
@@ -93,6 +104,10 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model):
 
                     function_full_response += line
 
+        if image_base64:
+            yield await generate_no_stream_response(timestamp, model, content=content, tools_id=None, function_call_name=None, function_call_content=None, role=None, total_tokens=totalTokenCount, prompt_tokens=promptTokenCount, completion_tokens=candidatesTokenCount, image_base64=image_base64)
+            return
+
         if need_function_call:
             function_call = json.loads(function_full_response)
             function_call_name = function_call["functionCall"]["name"]
@@ -535,9 +550,13 @@ async def fetch_response(client, url, headers, payload, engine, model):
         # print("parsed_data", json.dumps(parsed_data, indent=4, ensure_ascii=False))
         content = ""
         reasoning_content = ""
+        image_base64 = ""
         parts_list = safe_get(parsed_data, 0, "candidates", 0, "content", "parts", default=[])
         for item in parts_list:
             chunk = safe_get(item, "text")
+            b64_json = safe_get(item, "inlineData", "data", default="")
+            if b64_json:
+                image_base64 = b64_json
             is_think = safe_get(item, "thought", default=False)
             # logger.info(f"chunk: {repr(chunk)}")
             if chunk:
@@ -571,7 +590,7 @@ async def fetch_response(client, url, headers, payload, engine, model):
         function_call_content = safe_get(parsed_data, -1, "candidates", 0, "content", "parts", 0, "functionCall", "args", default=None)
 
         timestamp = int(datetime.timestamp(datetime.now()))
-        yield await generate_no_stream_response(timestamp, model, content=content, tools_id=None, function_call_name=function_call_name, function_call_content=function_call_content, role=role, total_tokens=total_tokens, prompt_tokens=prompt_tokens, completion_tokens=candidates_tokens, reasoning_content=reasoning_content)
+        yield await generate_no_stream_response(timestamp, model, content=content, tools_id=None, function_call_name=function_call_name, function_call_content=function_call_content, role=role, total_tokens=total_tokens, prompt_tokens=prompt_tokens, completion_tokens=candidates_tokens, reasoning_content=reasoning_content, image_base64=image_base64)
 
     elif engine == "claude":
         response_json = response.json()

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm/aient/src/aient/core/utils.py

@@ -112,7 +112,7 @@ def get_engine(provider, endpoint=None, original_model=""):
     if provider.get("engine"):
         engine = provider["engine"]
 
-    if endpoint == "/v1/images/generations" or "stable-diffusion" in original_model:
+    if engine != "gemini" and (endpoint == "/v1/images/generations" or "stable-diffusion" in original_model):
         engine = "dalle"
         stream = False
 
@@ -449,7 +449,7 @@ end_of_line = "\n\n"
 
 import random
 import string
-async def generate_sse_response(timestamp, model, content=None, tools_id=None, function_call_name=None, function_call_content=None, role=None, total_tokens=0, prompt_tokens=0, completion_tokens=0, reasoning_content=None):
+async def generate_sse_response(timestamp, model, content=None, tools_id=None, function_call_name=None, function_call_content=None, role=None, total_tokens=0, prompt_tokens=0, completion_tokens=0, reasoning_content=None, stop=None):
     random.seed(timestamp)
     random_str = ''.join(random.choices(string.ascii_letters + string.digits, k=29))
 
@@ -467,7 +467,7 @@ async def generate_sse_response(timestamp, model, content=None, tools_id=None, f
                 "index": 0,
                 "delta": delta_content,
                 "logprobs": None,
-                "finish_reason": None if content else "stop"
+                "finish_reason": None if content or reasoning_content else "stop"
             }
         ],
         "usage": None,
@@ -484,14 +484,19 @@ async def generate_sse_response(timestamp, model, content=None, tools_id=None, f
         total_tokens = prompt_tokens + completion_tokens
         sample_data["usage"] = {"prompt_tokens": prompt_tokens, "completion_tokens": completion_tokens, "total_tokens": total_tokens}
         sample_data["choices"] = []
+    if stop:
+        sample_data["choices"][0]["delta"] = {}
+        sample_data["choices"][0]["finish_reason"] = stop
+
     json_data = json.dumps(sample_data, ensure_ascii=False)
+    # print("json_data", json.dumps(sample_data, indent=4, ensure_ascii=False))
 
     # 构建SSE响应
     sse_response = f"data: {json_data}" + end_of_line
 
     return sse_response
 
-async def generate_no_stream_response(timestamp, model, content=None, tools_id=None, function_call_name=None, function_call_content=None, role=None, total_tokens=0, prompt_tokens=0, completion_tokens=0, reasoning_content=None):
+async def generate_no_stream_response(timestamp, model, content=None, tools_id=None, function_call_name=None, function_call_content=None, role=None, total_tokens=0, prompt_tokens=0, completion_tokens=0, reasoning_content=None, image_base64=None):
     random.seed(timestamp)
     random_str = ''.join(random.choices(string.ascii_letters + string.digits, k=29))
     message = {
@@ -554,11 +559,25 @@ async def generate_no_stream_response(timestamp, model, content=None, tools_id=N
             "system_fingerprint": "fp_4691090a87"
         }
 
+    if image_base64:
+        sample_data = {
+            "created": timestamp,
+            "data": [{
+                "b64_json": image_base64
+            }],
+            # "usage": {
+            #     "total_tokens": 100,
+            #     "input_tokens": 50,
+            #     "output_tokens": 50,
+            # }
+        }
+
     if total_tokens:
         total_tokens = prompt_tokens + completion_tokens
         sample_data["usage"] = {"prompt_tokens": prompt_tokens, "completion_tokens": completion_tokens, "total_tokens": total_tokens}
 
     json_data = json.dumps(sample_data, ensure_ascii=False)
+    # print("json_data", json.dumps(sample_data, indent=4, ensure_ascii=False))
 
     return json_data
 

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm/tools/__init__.py

@@ -1,6 +1,6 @@
 from .think import think
 from .edit_file import edit_file
-from .worker import worker
+from .worker import worker, worker_gen
 
 from .search_arxiv import search_arxiv
 from .repomap import get_code_repo_map
@@ -26,6 +26,7 @@ __all__ = [
     "think",
     "edit_file",
     "worker",
+    "worker_gen",
     "search_arxiv",
     "get_code_repo_map",
     # aient.plugins

beswarm-0.1.63/beswarm/tools/worker.py

@@ -0,0 +1,271 @@
+import os
+import re
+import copy
+import json
+import platform
+from datetime import datetime
+
+from ..aient.src.aient.models import chatgpt
+from ..aient.src.aient.plugins import register_tool, get_function_call_list
+from ..aient.src.aient.prompt import system_prompt, instruction_system_prompt
+from ..utils import extract_xml_content, get_current_screen_image_message
+
+@register_tool()
+async def worker(goal, tools, work_dir, cache_messages=None):
+    """
+    启动一个 **工作智能体 (Worker Agent)** 来自动完成指定的任务目标 (`goal`)。
+
+    这个工作智能体接收一个清晰的任务描述、一组可供调用的工具 (`tools`)，以及一个工作目录 (`work_dir`)。
+    它会利用语言模型的能力，结合可用的工具，自主规划并逐步执行必要的操作，直到最终完成指定的任务目标。
+    核心功能是根据输入的目标，驱动整个任务执行流程。
+
+    Args:
+        goal (str): 需要完成的具体任务目标描述。工作智能体将围绕此目标进行工作。必须清晰、具体。
+        tools (list[str]): 一个包含可用工具函数对象的列表。工作智能体在执行任务时可能会调用这些工具来与环境交互（例如读写文件、执行命令等）。
+        work_dir (str): 工作目录的绝对路径。工作智能体将在此目录上下文中执行操作。
+
+    Returns:
+        str: 当任务成功完成时，返回字符串 "任务已完成"。
+    """
+
+    tools_json = [value for _, value in get_function_call_list(tools).items()]
+    work_agent_system_prompt = system_prompt.format(
+        os_version=platform.platform(),
+        workspace_path=work_dir,
+        shell=os.getenv('SHELL', 'Unknown'),
+        tools_list=tools_json
+    )
+
+    work_agent_config = {
+        "api_key": os.getenv("API_KEY"),
+        "api_url": os.getenv("BASE_URL"),
+        "engine": os.getenv("MODEL"),
+        "system_prompt": work_agent_system_prompt,
+        "print_log": True,
+        # "max_tokens": 8000,
+        "temperature": 0.5,
+        "function_call_max_loop": 100,
+    }
+    if cache_messages:
+        work_agent_config["cache_messages"] = cache_messages
+
+    instruction_agent_config = {
+        "api_key": os.getenv("API_KEY"),
+        "api_url": os.getenv("BASE_URL"),
+        "engine": os.getenv("MODEL"),
+        "system_prompt": instruction_system_prompt.format(os_version=platform.platform(), tools_list=tools_json, workspace_path=work_dir, current_time=datetime.now().strftime("%Y-%m-%d %H:%M:%S")),
+        "print_log": False,
+        # "max_tokens": 4000,
+        "temperature": 0.7,
+        "use_plugins": False,
+    }
+
+    # 工作agent初始化
+    work_agent = chatgpt(**work_agent_config)
+    async def instruction_agent_task():
+        while True:
+
+            instruction_prompt = f"""
+</work_agent_conversation_end>
+任务目标: {goal}
+
+在 tag <work_agent_conversation_start>...</work_agent_conversation_end> 之前的对话历史都是工作智能体的对话历史。
+
+根据以上对话历史和目标，请生成下一步指令。如果任务已完成，请回复"任务已完成"。
+            """
+            # 让指令agent分析对话历史并生成新指令
+            instruction_agent = chatgpt(**instruction_agent_config)
+            conversation_history = copy.deepcopy(work_agent.conversation["default"])
+
+            cache_dir = os.path.join(work_dir, ".beswarm")
+            os.makedirs(cache_dir, exist_ok=True)
+            cache_file = os.path.join(cache_dir, "work_agent_conversation_history.json")
+            with open(cache_file, "w", encoding="utf-8") as f:
+                f.write(json.dumps(conversation_history, ensure_ascii=False, indent=4))
+
+            work_agent_system_prompt = conversation_history.pop(0)
+            if conversation_history:
+                # 获取原始内容
+                original_content = work_agent_system_prompt["content"]
+
+                # 定义正则表达式
+                regex = r"<latest_file_content>(.*?)</latest_file_content>"
+
+                # 进行匹配
+                match = re.search(regex, original_content, re.DOTALL)
+
+                # 提取内容或设置为空字符串
+                if match:
+                    extracted_content = f"<latest_file_content>{match.group(1)}</latest_file_content>\n\n"
+                else:
+                    extracted_content = ""
+
+                conversation_history[0]["content"] = extracted_content + conversation_history[0]["content"]
+
+            instruction_agent.conversation["default"][1:] = conversation_history
+            if "find_and_click_element" in str(tools_json):
+                instruction_prompt = await get_current_screen_image_message(instruction_prompt)
+            next_instruction = await instruction_agent.ask_async(instruction_prompt)
+            print("\n🤖 指令智能体生成的下一步指令:", next_instruction)
+            if "fetch_gpt_response_stream HTTP Error', 'status_code': 404" in next_instruction:
+                raise Exception(f"Model: {instruction_agent_config['engine']} not found!")
+            if "'status_code': 413" in next_instruction:
+                raise Exception(f"The request body is too long, please try again.")
+            next_instruction = extract_xml_content(next_instruction, "instructions")
+            if not next_instruction:
+                print("\n❌ 指令智能体生成的指令不符合要求，请重新生成。")
+                continue
+            else:
+                if conversation_history == []:
+                    next_instruction = (
+                        "任务描述：\n"
+                        f"{goal}\n\n"
+                        "现在开始执行第一步：\n"
+                        f"{next_instruction}"
+                    )
+                break
+        return next_instruction
+
+    need_instruction = True
+    while True:
+        next_instruction = ''
+        if need_instruction:
+            next_instruction = await instruction_agent_task()
+
+            # 检查任务是否完成
+            if "任务已完成" in next_instruction:
+                print("\n✅ 任务已完成！")
+                break
+        if "find_and_click_element" in str(tools_json):
+            next_instruction = await get_current_screen_image_message(next_instruction)
+        result = await work_agent.ask_async(next_instruction)
+        if result.strip() == '' or result.strip() == '</content>\n</write_to_file>':
+            print("\n❌ 工作智能体回复为空，请重新生成指令。")
+            need_instruction = False
+            continue
+        print("✅ 工作智能体回复:", result)
+        need_instruction = True
+
+    return "任务已完成"
+
+async def worker_gen(goal, tools, work_dir, cache_messages=None):
+    tools_json = [value for _, value in get_function_call_list(tools).items()]
+    work_agent_system_prompt = system_prompt.format(
+        os_version=platform.platform(),
+        workspace_path=work_dir,
+        shell=os.getenv('SHELL', 'Unknown'),
+        tools_list=tools_json
+    )
+
+    work_agent_config = {
+        "api_key": os.getenv("API_KEY"),
+        "api_url": os.getenv("BASE_URL"),
+        "engine": os.getenv("MODEL"),
+        "system_prompt": work_agent_system_prompt,
+        "print_log": True,
+        # "max_tokens": 8000,
+        "temperature": 0.5,
+        "function_call_max_loop": 100,
+    }
+    if cache_messages:
+        work_agent_config["cache_messages"] = cache_messages
+
+    instruction_agent_config = {
+        "api_key": os.getenv("API_KEY"),
+        "api_url": os.getenv("BASE_URL"),
+        "engine": os.getenv("MODEL"),
+        "system_prompt": instruction_system_prompt.format(os_version=platform.platform(), tools_list=tools_json, workspace_path=work_dir, current_time=datetime.now().strftime("%Y-%m-%d %H:%M:%S")),
+        "print_log": False,
+        # "max_tokens": 4000,
+        "temperature": 0.7,
+        "use_plugins": False,
+    }
+
+    # 工作agent初始化
+    work_agent = chatgpt(**work_agent_config)
+    async def instruction_agent_task():
+        while True:
+
+            instruction_prompt = f"""
+</work_agent_conversation_end>
+任务目标: {goal}
+
+在 tag <work_agent_conversation_start>...</work_agent_conversation_end> 之前的对话历史都是工作智能体的对话历史。
+
+根据以上对话历史和目标，请生成下一步指令。如果任务已完成，请回复"任务已完成"。
+            """
+            # 让指令agent分析对话历史并生成新指令
+            instruction_agent = chatgpt(**instruction_agent_config)
+            conversation_history = copy.deepcopy(work_agent.conversation["default"])
+
+            cache_dir = os.path.join(work_dir, ".beswarm")
+            os.makedirs(cache_dir, exist_ok=True)
+            cache_file = os.path.join(cache_dir, "work_agent_conversation_history.json")
+            with open(cache_file, "w", encoding="utf-8") as f:
+                f.write(json.dumps(conversation_history, ensure_ascii=False, indent=4))
+
+            work_agent_system_prompt = conversation_history.pop(0)
+            if conversation_history:
+                # 获取原始内容
+                original_content = work_agent_system_prompt["content"]
+
+                # 定义正则表达式
+                regex = r"<latest_file_content>(.*?)</latest_file_content>"
+
+                # 进行匹配
+                match = re.search(regex, original_content, re.DOTALL)
+
+                # 提取内容或设置为空字符串
+                if match:
+                    extracted_content = f"<latest_file_content>{match.group(1)}</latest_file_content>\n\n"
+                else:
+                    extracted_content = ""
+
+                conversation_history[0]["content"] = extracted_content + conversation_history[0]["content"]
+
+            instruction_agent.conversation["default"][1:] = conversation_history
+            if "find_and_click_element" in str(tools_json):
+                instruction_prompt = await get_current_screen_image_message(instruction_prompt)
+            next_instruction = await instruction_agent.ask_async(instruction_prompt)
+            print("\n🤖 指令智能体生成的下一步指令:", next_instruction)
+            if "fetch_gpt_response_stream HTTP Error', 'status_code': 404" in next_instruction:
+                raise Exception(f"Model: {instruction_agent_config['engine']} not found!")
+            if "'status_code': 413" in next_instruction:
+                raise Exception(f"The request body is too long, please try again.")
+            next_instruction = extract_xml_content(next_instruction, "instructions")
+            if not next_instruction:
+                print("\n❌ 指令智能体生成的指令不符合要求，请重新生成。")
+                continue
+            else:
+                if conversation_history == []:
+                    next_instruction = (
+                        "任务描述：\n"
+                        f"{goal}\n\n"
+                        "现在开始执行第一步：\n"
+                        f"{next_instruction}"
+                    )
+                break
+        return next_instruction
+
+    need_instruction = True
+    while True:
+        next_instruction = ''
+        if need_instruction:
+            next_instruction = await instruction_agent_task()
+
+            yield {"user": next_instruction}
+
+            # 检查任务是否完成
+            if "任务已完成" in next_instruction:
+                print("\n✅ 任务已完成！")
+                break
+        if "find_and_click_element" in str(tools_json):
+            next_instruction = await get_current_screen_image_message(next_instruction)
+        result = await work_agent.ask_async(next_instruction)
+        if result.strip() == '' or result.strip() == '</content>\n</write_to_file>':
+            print("\n❌ 工作智能体回复为空，请重新生成指令。")
+            need_instruction = False
+            continue
+        yield {"assistant": result}
+        print("✅ 工作智能体回复:", result)
+        need_instruction = True

{beswarm-0.1.61 → beswarm-0.1.63}/beswarm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: beswarm
-Version: 0.1.61
+Version: 0.1.63
 Summary: MAS
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{beswarm-0.1.61 → beswarm-0.1.63}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "beswarm"
-version = "0.1.61"
+version = "0.1.63"
 description = "MAS"
 readme = "README.md"
 requires-python = ">=3.11"

beswarm-0.1.61/beswarm/tools/worker.py

@@ -1,130 +0,0 @@
-import os
-import copy
-import json
-import platform
-from datetime import datetime
-
-from ..aient.src.aient.models import chatgpt
-from ..aient.src.aient.plugins import register_tool, get_function_call_list
-from ..aient.src.aient.prompt import system_prompt, instruction_system_prompt
-from ..utils import extract_xml_content, get_current_screen_image_message
-
-@register_tool()
-async def worker(goal, tools, work_dir, cache_messages=None):
-    """
-    启动一个 **工作智能体 (Worker Agent)** 来自动完成指定的任务目标 (`goal`)。
-
-    这个工作智能体接收一个清晰的任务描述、一组可供调用的工具 (`tools`)，以及一个工作目录 (`work_dir`)。
-    它会利用语言模型的能力，结合可用的工具，自主规划并逐步执行必要的操作，直到最终完成指定的任务目标。
-    核心功能是根据输入的目标，驱动整个任务执行流程。
-
-    Args:
-        goal (str): 需要完成的具体任务目标描述。工作智能体将围绕此目标进行工作。必须清晰、具体。
-        tools (list[str]): 一个包含可用工具函数对象的列表。工作智能体在执行任务时可能会调用这些工具来与环境交互（例如读写文件、执行命令等）。
-        work_dir (str): 工作目录的绝对路径。工作智能体将在此目录上下文中执行操作。
-
-    Returns:
-        str: 当任务成功完成时，返回字符串 "任务已完成"。
-    """
-
-    tools_json = [value for _, value in get_function_call_list(tools).items()]
-    work_agent_system_prompt = system_prompt.format(
-        os_version=platform.platform(),
-        workspace_path=work_dir,
-        shell=os.getenv('SHELL', 'Unknown'),
-        tools_list=tools_json
-    )
-
-    work_agent_config = {
-        "api_key": os.getenv("API_KEY"),
-        "api_url": os.getenv("BASE_URL"),
-        "engine": os.getenv("MODEL"),
-        "system_prompt": work_agent_system_prompt,
-        "print_log": True,
-        # "max_tokens": 8000,
-        "temperature": 0.5,
-        "function_call_max_loop": 100,
-    }
-    if cache_messages:
-        work_agent_config["cache_messages"] = cache_messages
-
-    instruction_agent_config = {
-        "api_key": os.getenv("API_KEY"),
-        "api_url": os.getenv("BASE_URL"),
-        "engine": os.getenv("MODEL"),
-        "system_prompt": instruction_system_prompt.format(os_version=platform.platform(), tools_list=tools_json, workspace_path=work_dir, current_time=datetime.now().strftime("%Y-%m-%d %H:%M:%S")),
-        "print_log": False,
-        # "max_tokens": 4000,
-        "temperature": 0.7,
-        "use_plugins": False,
-    }
-
-    # 工作agent初始化
-    work_agent = chatgpt(**work_agent_config)
-    async def instruction_agent_task():
-        while True:
-
-            instruction_prompt = f"""
-</work_agent_conversation_end>
-任务目标: {goal}
-
-在 tag <work_agent_conversation_start>...</work_agent_conversation_end> 之前的对话历史都是工作智能体的对话历史。
-
-根据以上对话历史和目标，请生成下一步指令。如果任务已完成，请回复"任务已完成"。
-            """
-            # 让指令agent分析对话历史并生成新指令
-            instruction_agent = chatgpt(**instruction_agent_config)
-            conversation_history = copy.deepcopy(work_agent.conversation["default"])
-
-            cache_dir = os.path.join(work_dir, ".beswarm")
-            os.makedirs(cache_dir, exist_ok=True)
-            cache_file = os.path.join(cache_dir, "work_agent_conversation_history.json")
-            with open(cache_file, "w", encoding="utf-8") as f:
-                f.write(json.dumps(conversation_history, ensure_ascii=False, indent=4))
-
-            conversation_history.pop(0)
-            instruction_agent.conversation["default"][1:] = conversation_history
-            if "find_and_click_element" in str(tools_json):
-                instruction_prompt = await get_current_screen_image_message(instruction_prompt)
-            next_instruction = await instruction_agent.ask_async(instruction_prompt)
-            print("\n🤖 指令智能体生成的下一步指令:", next_instruction)
-            if "fetch_gpt_response_stream HTTP Error', 'status_code': 404" in next_instruction:
-                raise Exception(f"Model: {instruction_agent_config['engine']} not found!")
-            if "'status_code': 413" in next_instruction:
-                raise Exception(f"The request body is too long, please try again.")
-            next_instruction = extract_xml_content(next_instruction, "instructions")
-            if not next_instruction:
-                print("\n❌ 指令智能体生成的指令不符合要求，请重新生成。")
-                continue
-            else:
-                if conversation_history == []:
-                    next_instruction = (
-                        "任务描述：\n"
-                        f"{goal}\n\n"
-                        "现在开始执行第一步：\n"
-                        f"{next_instruction}"
-                    )
-                break
-        return next_instruction
-
-    need_instruction = True
-    while True:
-        next_instruction = ''
-        if need_instruction:
-            next_instruction = await instruction_agent_task()
-
-            # 检查任务是否完成
-            if "任务已完成" in next_instruction:
-                print("\n✅ 任务已完成！")
-                break
-        if "find_and_click_element" in str(tools_json):
-            next_instruction = await get_current_screen_image_message(next_instruction)
-        result = await work_agent.ask_async(next_instruction)
-        if result.strip() == '' or result.strip() == '</content>\n</write_to_file>':
-            print("\n❌ 工作智能体回复为空，请重新生成指令。")
-            need_instruction = False
-            continue
-        print("✅ 工作智能体回复:", result)
-        need_instruction = True
-
-    return "任务已完成"