Cognition 首席执行官否认 SpaceX 试图收购该初创公司的传闻

+-------------------------------------------------------------+
|                        用户输入需求                          |
+------------------------------+------------------------------+
                               |
                               v
+------------------------------+------------------------------+
|                       规划器智能体                          |
|                 (将任务拆解为具体的代码步骤)                  |
+------------------------------+------------------------------+
                               |
                               v
+------------------------------+------------------------------+
|                       自主执行与纠错环                       |
|  +-------------------------------------------------------+  |
|  |                     1. 代码生成器                     |  |
|  |            (通过 n1n.ai API 调用最强代码模型)           |  |
|  +---------------------------+---------------------------+  |
|                              |                              |
|                              v                              |
|  +---------------------------+---------------------------+  |
|  |                     2. 沙箱运行器                     |  |
|  |                 (子进程安全执行 Python 脚本)          |  |
|  +---------------------------+---------------------------+  |
|                              |                              |
|                              v                              |
|  +---------------------------+---------------------------+  |
|  |                     3. 结果评估器                     |  |
|  |               (检测 Exit Code 及标准错误输出)          |  |
|  +---------------------------+---------------------------+  |
|                              |                              |
|                              v                              |
|  +---------------------------+---------------------------+  |
|  |                     4. 错误反馈修复                   |  |
|  |         (若报错，将 Traceback 重新输入给生成器)        |  |
|  +-------------------------------------------------------+  |
+------------------------------+------------------------------+
                               |
                               v
+------------------------------+------------------------------+
|                        输出可用代码                          |
+-------------------------------------------------------------+

pip install requests

import json
import os
import subprocess
import requests

# 配置 n1n.ai API 访问
API_KEY = os.getenv("N1N_API_KEY", "您的_n1n_api_key")
API_URL = "https://api.n1n.ai/v1/chat/completions"

def call_llm(prompt: str, system_instruction: str, model: str = "claude-3-5-sonnet") -> str:
    headers = {
        "Authorization": f"Bearer {API_KEY}",
        "Content-Type": "application/json"
    }

    payload = {
        "model": model,
        "messages": [
            {"role": "system", "content": system_instruction},
            {"role": "user", "content": prompt}
        ],
        "temperature": 0.1  # 较低的温度以确保代码生成的确定性
    }

    response = requests.post(API_URL, json=payload, headers=headers)
    if response.status_code != 200:
        raise Exception(f"API 调用失败: {response.status_code} - {response.text}")

    result = response.json()
    return result["choices"][0]["message"]["content"]

def extract_code(llm_output: str) -> str:
    # 提取 Markdown 格式的代码块
    if "```python" in llm_output:
        return llm_output.split("```python")[1].split("```")[0].strip()
    elif "```" in llm_output:
        return llm_output.split("```")[1].split("```")[0].strip()
    return llm_output.strip()

def run_and_test_code(code: str, filename: str = "sandbox_test.py") -> tuple[bool, str]:
    # 将代码写入临时文件
    with open(filename, "w", encoding="utf-8") as f:
        f.write(code)

    try:
        # 在子进程中执行代码，设置 10 秒超时防止死循环
        result = subprocess.run(
            ["python", filename],
            capture_output=True,
            text=True,
            timeout=10
        )
        if result.returncode == 0:
            return True, result.stdout
        else:
            return False, result.stderr
    except subprocess.TimeoutExpired:
        return False, "错误：代码执行超时（超过 10 秒）。请优化算法结构。"
    finally:
        if os.path.exists(filename):
            os.remove(filename)

def run_autonomous_agent(task_desc: str, max_tries: int = 5) -> str:
    sys_prompt = (
        "你是一个顶尖的 Python 软件工程师。请直接输出可执行的 Python 代码，"
        "不要包含任何解释性文字或 Markdown 以外的文本。代码必须包裹在 ```python 和 ``` 之间。"
    )

    current_user_prompt = f"请编写一个 Python 脚本完成以下任务：{task_desc}"

    for i in range(1, max_tries + 1):
        print(f"\n正在进行第 {i} 次迭代尝试...")

        # 首次或前几次迭代使用高性价比的 DeepSeek-V3 模型
        current_model = "deepseek-v3" if i < 3 else "claude-3-5-sonnet"
        print(f"正在使用模型: {current_model}")

        raw_output = call_llm(current_user_prompt, sys_prompt, model=current_model)
        generated_code = extract_code(raw_output)

        print("代码生成完毕，正在进行安全沙箱测试...")
        success, log_info = run_and_test_code(generated_code)

        if success:
            print("【测试成功】代码顺利运行！输出如下：")
            print(log_info)
            return generated_code
        else:
            print("【测试失败】检测到运行时错误。")
            print(f"错误详情: {log_info}")
            # 构建反馈 Prompt，引导模型进行自我修复
            current_user_prompt = (
                f"你之前生成的代码运行失败了。\n\n"
                f"### 失败的代码：\n{generated_code}\n\n"
                f"### 报错信息：\n{log_info}\n\n"
                f"请仔细分析上述报错原因（如语法错误、类型匹配错误或导入缺失），并给出修复后的完整 Python 代码。"
            )

    raise RuntimeError("智能体在最大尝试次数内未能生成无错代码。")

if __name__ == "__main__":
    target_task = "读取一个包含用户数据的 JSON 字符串，过滤出所有年龄大于 21 岁且邮箱格式合法的用户，并输出他们的名字。"
    try:
        final_code = run_autonomous_agent(target_task)
        print("\n最终生成的可用代码：")
        print(final_code)
    except Exception as error:
        print(f"智能体执行失败: {error}")