AgentSkillsCN

browser

浏览器操作技能，适用于 Chrome、Firefox、Safari、Edge 等主流浏览器。当操作浏览器应用或进行网页相关任务时激活。

中文原作
SKILL.md
--- frontmatter
name: browser
description: 浏览器操作技能，适用于 Chrome、Firefox、Safari、Edge 等主流浏览器。当操作浏览器应用或进行网页相关任务时激活。

Browser Skill

操作规则

  • •像人类一样浏览网页 - 你看到的只是屏幕上的一部分内容，更多内容需要滚动才能看到
  • •使用 middle_click 打开链接到新标签页：保持当前页面不变，方便对比和返回
  • •地址栏操作：先聚焦地址栏，再输入 URL，最后按 enter
  • •搜索操作：点击搜索框 → 输入关键词 → 按 enter → 等待结果加载
  • •表单提交：填写完成后按 enter 或点击提交按钮

滚动浏览 (CRITICAL)

You can only see what's visible on screen. Web pages are like long scrolls - you MUST scroll to see more content.

Search Results Page

  • •Search engines typically show 10 results per page
  • •First screen may only show 3-5 results - scroll down to see more
  • •If you need "top 3 results", make sure you can actually SEE 3 results before clicking
  • •Scroll the search results page first if needed

Article/Content Pages

  • •Articles can be very long - multiple screens worth of content
  • •Don't assume you've seen everything after one scroll
  • •Keep scrolling until you reach: article footer, comments section, "related articles", or page bottom
  • •Important information may be in the middle or end of the article

操作示例

打开新网址

code
1. hotkey: "ctrl l" (Windows/Linux) 或 "cmd l" (macOS) - 聚焦地址栏
2. type: "https://example.com"
3. hotkey: "enter"
4. wait: 1000

搜索操作

code
1. click: 搜索框位置
2. type: "搜索关键词"
3. hotkey: "enter"
4. wait: 1000

打开多个搜索结果

code
1. middle_click: 第一个结果链接
2. middle_click: 第二个结果链接
3. middle_click: 第三个结果链接
4. wait: 1000

截取多个页面内容 (IMPORTANT)

Process ONE page at a time - Do NOT batch screenshot operations across multiple tabs.

Complete workflow for extracting page content:

code
# Step 1: Open the page
1. click: target tab - switch to the page
2. wait: 1000 - wait for page to fully load

# Step 2: Handle obstructions FIRST
3. Check screenshot for any blocking elements:
   - Cookie consent banners -> click "Accept" or "X"
   - Login popups -> click "X" or press "escape"
   - Ad overlays -> find and click close button
   - Newsletter popups -> press "escape" or close

# Step 3: Scroll and read - A CONTINUOUS PROCESS (NOT just one scroll!)
4. Read the visible content on screen
5. scroll down in page content area
6. wait: 300
7. Read more content
8. scroll down again
9. wait: 300
10. Read more content
11. KEEP SCROLLING AND READING until you see:
    - Article footer / author info
    - Comments section
    - "Related articles" section
    - Page bottom / no more content

**Scrolling is like reading a book - you turn pages CONTINUOUSLY until you finish the chapter, not just once!**

# Step 4: MUST save extracted content with write_file
12. write_file(file_path="workspace/[topic].md", content="...")
   - YOU MUST CALL write_file TOOL - just thinking about content is NOT enough!
   - Include key information, quotes, and facts
   - This file will be your reference for the final summary
   - WITHOUT saving to file, you will FORGET the content when moving to next page!

# Step 5: Close and move to next page
13. hotkey: "cmd w" - close current tab
14. Process next page in the NEXT response (not same response)

[CRITICAL] Scrolling is a CONTINUOUS process, NOT a single action!

  • •One scroll only moves the page a little bit - you need MULTIPLE scrolls to read a full article
  • •Think of it like reading a physical newspaper - you keep scrolling down as you read
  • •A typical article requires 3-10 scrolls to read completely
  • •STOP scrolling only when you reach the END of the article (footer, comments, related articles)
  • •If you only scroll once, you've only seen ~20% of the content!

[CRITICAL] You MUST call write_file tool to save content!

  • •Thinking "I'll remember this" is NOT enough - you WILL forget
  • •Screenshots are images, not searchable text
  • •The saved .md file is your ONLY reliable reference for final summary
  • •If you don't call write_file, the information is LOST

Why save to file?

  • •Screenshots alone are not searchable or quotable
  • •Saved content can be referenced when writing the final summary
  • •Ensures you don't lose information when closing tabs

Why one page at a time?

  • •Pages may have ads, login popups, or cookie banners that block content
  • •You need to see the actual page to handle these obstacles
  • •Page length varies - you must scroll until reaching the end
  • •Batching across tabs often results in capturing incomplete content

滚动操作注意事项

CRITICAL: Scroll coordinate must be in the PAGE CONTENT area, NOT the tab bar or toolbar!

  • •Tab bar / Toolbar area (top of browser window): Scrolling here has NO effect
  • •Page content area (main webpage area, center of screen): Scroll here to navigate page content

Always scroll in the center of the visible webpage content, not near the top of the browser window.

处理页面遮挡

Many pages have elements that block content:

  • •Cookie consent banners: Look for "Accept" or "X" button to dismiss
  • •Login popups: Look for "X" close button, or press escape
  • •Ad overlays: Look for close button, usually in corner
  • •Newsletter popups: Press escape or find close button

Always check the screenshot after loading a page. If content is blocked, handle the obstruction first before scrolling/capturing.

技巧

  • •批量打开链接：使用 middle_click 连续打开多个链接，然后逐个处理
  • •快速导航：使用快捷键比点击更快（地址栏、刷新、后退等）
  • •等待加载：页面跳转后务必 wait，等待内容加载完成
  • •滚动查看：长页面需要滚动并多次截图才能获取完整内容
  • •标签页管理：处理完一个标签页后及时关闭，避免混乱

注意事项

  • •点击链接前确认是否需要在新标签页打开（middle_click vs click）
  • •页面加载需要时间，操作后要适当等待
  • •弹窗和对话框可能阻挡操作，需要先处理
  • •某些网站有反爬机制，操作过快可能触发验证