# 安装依赖：pip install anthropic

import anthropic
import base64
from pathlib import Path

client = anthropic.Anthropic()

def encode_image(image_path: str) -> str:
    """将截图转成 base64"""
    with open(image_path, "rb") as f:
        return base64.standard_b64encode(f.read()).decode("utf-8")


def run_computer_use_agent(task: str, screenshot_path: str):
    """
    运行 Computer Use Agent

    注意：实际使用时需要在沙箱环境中运行，
    并且需要将截图和实际操作连接起来。
    这里只展示 API 调用方式。
    """
    # 获取当前屏幕截图
    image_data = encode_image(screenshot_path)

    tools = [
        {
            "type": "computer_20241022",
            "name": "computer",
            "display_width_px": 1920,
            "display_height_px": 1080,
            "display_number": 1,
        }
    ]

    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "image/png",
                        "data": image_data,
                    },
                },
                {
                    "type": "text",
                    "text": task
                }
            ],
        }
    ]

    # 循环执行直到任务完成
    while True:
        response = client.messages.create(
            model="claude-opus-4-5",
            max_tokens=4096,
            tools=tools,
            messages=messages,
        )

        # 如果没有工具调用，说明任务完成或需要人工介入
        if response.stop_reason == "end_turn":
            return response.content[-1].text

        # 处理工具调用（实际项目中这里要执行真实的鼠标键盘操作）
        tool_use = next(
            block for block in response.content
            if block.type == "tool_use"
        )

        action = tool_use.input.get("action")
        print(f"Agent 决定执行操作：{action}")

        # ... 实际执行操作，截新图，继续循环
        # 这里省略了实际操作执行的代码
        break


# 用法示例
result = run_computer_use_agent(
    task="打开浏览器，访问 example.com，找到联系我们页面",
    screenshot_path="current_screen.png"
)

# 安装依赖：pip install playwright openai beautifulsoup4
# playwright install chromium

import asyncio
from playwright.async_api import async_playwright
from openai import OpenAI
from bs4 import BeautifulSoup

client = OpenAI()


def extract_interactive_elements(html: str) -> str:
    """从 HTML 中提取交互元素，减少 token 消耗"""
    soup = BeautifulSoup(html, "html.parser")

    elements = []

    # 提取表单元素
    for el in soup.find_all(["input", "textarea", "select", "button", "a"]):
        attrs = {}
        for attr in ["id", "name", "type", "placeholder", "value", "href", "class"]:
            if el.get(attr):
                attrs[attr] = el.get(attr)

        text = el.get_text(strip=True)
        elements.append(f"<{el.name} {attrs}>{text}</{el.name}>")

    return "\n".join(elements)


def ask_llm_for_action(task: str, page_elements: str, completed_steps: list) -> dict:
    """让 LLM 根据当前页面状态决定下一步操作"""

    completed_str = "\n".join([f"- {s}" for s in completed_steps]) if completed_steps else "无"

    prompt = f"""你正在用 Playwright 控制浏览器完成以下任务：
{task}

已完成的步骤：
{completed_str}

当前页面的交互元素：
{page_elements}

请决定下一步操作，返回 JSON 格式：
{{
    "action": "click" | "fill" | "select" | "done" | "navigate",
    "selector": "CSS 选择器或元素描述",
    "value": "如果是 fill 或 select，这里填入内容",
    "url": "如果是 navigate，这里填 URL",
    "description": "这一步在做什么"
}}

如果任务已完成，返回 action: "done"。
只返回 JSON，不要其他内容。"""

    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
        response_format={"type": "json_object"}
    )

    import json
    return json.loads(response.choices[0].message.content)


async def run_web_agent(task: str, start_url: str):
    """
    基于 Playwright + LLM 的网页自动化 Agent
    比 Computer Use 快 10 倍，成本低 90%
    """
    async with async_playwright() as p:
        browser = await p.chromium.launch(headless=False)
        page = await browser.new_page()

        await page.goto(start_url)
        await page.wait_for_load_state("networkidle")

        completed_steps = []
        max_steps = 20  # 防止无限循环

        for step_num in range(max_steps):
            # 获取当前页面的 HTML
            html = await page.content()
            elements = extract_interactive_elements(html)

            # 让 LLM 决定下一步
            action = ask_llm_for_action(task, elements, completed_steps)
            print(f"步骤 {step_num + 1}: {action['description']}")

            if action["action"] == "done":
                print("任务完成！")
                break
            elif action["action"] == "navigate":
                await page.goto(action["url"])
            elif action["action"] == "click":
                await page.click(action["selector"])
            elif action["action"] == "fill":
                await page.fill(action["selector"], action["value"])
            elif action["action"] == "select":
                await page.select_option(action["selector"], action["value"])

            completed_steps.append(action["description"])
            await page.wait_for_timeout(1000)  # 等待页面响应

        await browser.close()


# 使用示例
asyncio.run(run_web_agent(
    task="在搜索框中搜索 'Python 异步编程'，点击第一个结果",
    start_url="https://www.baidu.com"
))