Playwright 浏览器工具

适用版本：4.0.8.1+

Playwright 是 Agently 的内置浏览器工具，适合处理“仅靠 HTTP 抓取不够”的页面：

• 需要执行前端 JS 后再读取内容
• 需要拿到页面标题、最终跳转 URL、状态码
• 需要可选截图或链接提取

1. 初始化参数

from agently.builtins.tools import Playwright

playwright = Playwright(
    headless=True,
    timeout=30000,
    proxy=None,
    user_agent=None,
    response_mode="markdown",  # "markdown" | "text"
    max_content_length=8000,
    include_links=False,
    max_links=120,
    screenshot_path=None,
)

核心参数：

• response_mode：markdown 会把 <a> 转成 markdown 链接；text 返回纯文本
• include_links：是否额外返回 links
• screenshot_path：设置后会保存全页截图

2. 直接调用

import asyncio
from agently.builtins.tools import Playwright

playwright = Playwright(headless=True, response_mode="markdown")

async def main():
    result = await playwright.open("https://agently.tech")
    print(result)

asyncio.run(main())

3. 作为 Agent 工具接入

from agently import Agently
from agently.builtins.tools import Playwright

agent = Agently.create_agent()
playwright = Playwright(headless=True, response_mode="markdown")

agent.use_tools([playwright.open])
result = agent.input("先浏览 Agently 官网，再总结 TriggerFlow 的作用").start()
print(result)

通过 tool_info_list 注册时，工具名为 playwright_open。

4. 返回结构（成功）

典型字段：

• ok
• requested_url
• normalized_url
• url（最终 URL）
• status
• title
• content_format
• content
• screenshot_path
• links（仅当 include_links=True）

失败时返回 ok=False 和 error。

5. 使用建议

• 首次使用前先安装浏览器驱动（playwright install）
• 抓取稳定性优先时建议设置 timeout 与 proxy
• 若后续需要精确读取元素，不建议只依赖 content 文本，可结合专用抓取流程