browser-task

统筹协调多工具协作的浏览器任务：先搜索再导航、先登录再下载、跨多页交互等。当代理需要执行以下操作时可派上用场：(1) 先搜索目标网站并与其互动；(2) 登录或填写表单；(3) 在多页间穿梭，精准定位目标内容；(4) 从动态页面中提取下载链接。对于简单场景：静态页面可使用 webfetch，已知文件 URL 则可直接调用 curl。

name: browser-task description: "Orchestrates browser tasks requiring multi-tool coordination: search-then-navigate, login-then-download, multi-page interactions. Use when the agent needs to: (1) Search for a website then interact with it, (2) Login or fill forms, (3) Navigate multiple pages to find a target, (4) Extract download links from dynamic pages. For simple cases: use webfetch for static pages, curl for known file URLs."

BrowserTask

Complete complex browser tasks requiring multi-step, multi-tool coordination.

Available Tools

Tool	Capability	Use When
Search tools	Find trusted URLs	Target website unknown
webfetch	Fetch page content/HTML	Static pages, parse links
agent-browser	Browser interaction	Click, fill forms, login, dynamic content
curl	Download files	File URL is known

Core Principles

•Search first, never guess URLs - Use search tools to get trusted URLs
•Right tool for the job - agent-browser for interaction, curl for download
•Extract link then download - agent-browser cannot download files directly; extract href then use curl

Standard Download Pattern

code

Search/known URL -> agent-browser navigate to download page -> get download button href -> curl download

Key commands:

bash

# Get download button's link
agent-browser get attr @download-button href

# Or use webfetch to get HTML and parse
webfetch <url> --format html

agent-browser Key Points

For detailed commands, load agent-browser skill. Common usage:

•agent-browser open <url> - Open page
•agent-browser snapshot -i - Get interactive elements
•agent-browser click @e1 - Click element
•agent-browser get attr @e1 href - Get link attribute

Important Notes

•agent-browser is slow to start; use webfetch/curl when possible
•Search results are more reliable than guessed URLs
•Always use curl for file downloads, never attempt with agent-browser