browser-use

Browser Use 技能

概述

此技能支持使用 Playwright 浏览网站并与网页交互。

快速开始 - 使用脚本

重要提示： 所有 uv run 命令都必须在 scripts 文件夹中执行（而非项目根目录）：

cd .claude/skills/browser-use/scripts

# === MANUAL BROWSING (opens browser for user interaction) ===
# Open browser for manual use (waits for user to close)
uv run browser.py open https://example.com --wait 60
uv run browser.py open https://example.com --account myaccount --wait 120

# === AUTOMATION (headless, returns content) ===
# Navigate to a URL and get content
uv run browser.py auto https://example.com
uv run browser.py auto https://example.com --account myaccount

# Take a screenshot
uv run browser.py screenshot https://example.com -o screenshot.png

# Extract text from a page
uv run browser.py text https://example.com --selector "h1"

# Get all links from a page
uv run browser.py links https://example.com

# Save page as PDF
uv run browser.py pdf https://example.com -o page.pdf

# Download a file by clicking a download link
uv run browser.py download https://example.com/downloads "a.download-btn" -o ~/Downloads

# Upload file(s) to a page
uv run browser.py upload https://example.com/upload "input[type=file]" myfile.pdf
uv run browser.py upload https://example.com/upload "input[type=file]" file1.txt file2.txt --submit "button[type=submit]"

# Upload via file chooser dialog (for dynamic inputs)
uv run browser.py upload-chooser https://example.com/upload "button.upload-trigger" myfile.pdf

# Click an element
uv run browser.py click https://example.com "button.submit"

# Fill an input field and press Enter
uv run browser.py fill https://google.com "input[name=q]" "search term" --press Enter

# Extract attribute from elements
uv run browser.py extract https://example.com "img" --attr src --all

图片下载命令

以优化的性能从网站下载图片：

cd .claude/skills/browser-use/scripts

# Download images directly from src attribute
uv run browser.py download-images https://example.com "img.gallery" -n 10 -o ./images

# Download from Google Images (19x faster with regex extraction)
uv run browser.py download-from-gallery \
  "https://www.google.com/search?q=keyword&tbm=isch&tbs=isz:l" \
  "div[data-id] img" \
  "img[jsname='kn3ccd']" \
  -n 100 \
  -o ~/Downloads \
  -a myaccount

# Search Google Images with size filter
uv run browser.py google-image "landscape wallpaper" -n 50 -o ~/Downloads -s Large

# Download 4K images (3840px+ minimum)
uv run browser.py google-image "wallpaper" -n 20 -o ~/Downloads -s 4k

# Download FullHD images (1920px+ minimum)
uv run browser.py google-image "wallpaper" -n 50 -o ~/Downloads -s fullhd

Google 图片尺寸筛选

CLI 尺寸选项（按最小尺寸筛选）：

• -s 4k - 最小 3840px（4K 分辨率）
• -s fullhd - 最小 1920px（全高清）
• -s Large - 最小 1000px
• -s Medium - 最小 400px
• -s Icon - 无最小尺寸限制

Google URL 参数（内部使用）：

• tbs=isz:l - 大尺寸图片
• tbs=isz:m - 中尺寸图片
• tbs=isz:i - 图标尺寸

YouTube 命令

搜索并下载 YouTube 视频，可按时长和日期筛选：

cd .claude/skills/browser-use/scripts

# Search YouTube and get video URLs (returns JSON)
uv run browser.py youtube-search "python tutorial" -n 10
uv run browser.py youtube-search "lofi music" -n 5 -o results.json -s screenshot.png

# Search with duration filter (4-20 minutes)
uv run browser.py youtube-search "lofi music" -n 5 -min 4 -max 20

# Search videos from the last week
uv run browser.py youtube-search "news" -n 10 -t week

# Search with custom date range (YYYYMMDD format)
uv run browser.py youtube-search "tutorial" -n 10 -df 20240101 -dt 20241231

# Download a single video by URL
uv run browser.py youtube-download "https://www.youtube.com/watch?v=dQw4w9WgXcQ" -o ~/Downloads

# Download with quality options
uv run browser.py youtube-download "https://youtube.com/watch?v=..." -q 720p -o ~/Downloads
uv run browser.py youtube-download "https://youtube.com/watch?v=..." -q 1080p -o ~/Downloads

# Download audio only (mp3)
uv run browser.py youtube-download "https://youtube.com/watch?v=..." -a -o ./music

# Search and download in one command
uv run browser.py youtube-download "lofi hip hop" --search -o ~/Downloads
uv run browser.py youtube-download "python tutorial" --search -n 3 -o ~/Downloads

# Search and download with duration filter (4-20 min videos only)
uv run browser.py youtube-download "lofi music" --search -n 5 -min 4 -max 20 -o ~/Downloads

# Search and download videos from a date range
uv run browser.py youtube-download "tutorial" --search -n 5 -df 20240101 -dt 20241231 -o ~/Downloads

YouTube 时长筛选

• -min N - 最短时长（分钟）
• -max N - 最长时长（分钟）
• 筛选使用 YouTube URL 参数以提高速度，并以 Python 端验证作为备用

YouTube 日期筛选

• -t, --upload-date - 按上传日期筛选：hour、today、week、month、year
• -df, --date-from - 自定义日期范围的开始日期（YYYYMMDD 格式，例如 20240101）
• -dt, --date-to - 自定义日期范围的结束日期（YYYYMMDD 格式，例如 20241231）

YouTube 画质选项

• best - 最佳可用画质（默认）
• 1080p - 1080p 或更低
• 720p - 720p 或更低
• 480p - 480p 或更低
• 360p - 360p 或更低
• audio - 仅最佳音频

YouTube 搜索输出格式

[
  {
    "url": "https://www.youtube.com/watch?v=...",
    "title": "Video Title",
    "channel": "Channel Name",
    "duration": "10:30",
    "views": "1.2M views",
    "date": "2024-12-01"
  }
]

TikTok 命令

重要提示： TikTok 会拦截无头浏览器。搜索操作请使用 --no-headless。

cd .claude/skills/browser-use/scripts

# Login and save session for authenticated operations
uv run browser.py tiktok-login -a mytiktok
uv run browser.py tiktok-login -a mytiktok -w 180  # Extended wait for 2FA

# Search videos (--no-headless required for TikTok)
uv run browser.py tiktok-search "keyword" -n 10 --no-headless
uv run browser.py tiktok-search "#dance" -n 5 --no-headless  # By hashtag
uv run browser.py tiktok-search "cooking" -n 10 -a mytiktok --no-headless  # With account

# Download single video
uv run browser.py tiktok-download "https://tiktok.com/@user/video/123" -o ~/Downloads

# Search and download
uv run browser.py tiktok-download "funny cats" --search -n 5 -o ~/Downloads --no-headless
uv run browser.py tiktok-download "#cooking" --search -n 10 -o ~/Downloads -p 3 --no-headless

TikTok 搜索输出格式

[
  {
    "url": "https://www.tiktok.com/@user/video/123456789",
    "title": "Video description...",
    "author": "username",
    "views": "1.2M",
    "likes": "50K",
    "date": "2d ago"
  }
]

身份验证

保存并在不同浏览器命令之间重复使用登录会话。使用启用了隐身模式的 Chrome 绕过自动化检测（适用于 Google 等）：

# Run from scripts folder
cd .claude/skills/browser-use/scripts

# Step 1: Create a login session (opens Chrome for manual login)
# Browser will open maximized - login manually then CLOSE THE BROWSER when done
uv run browser.py create-login https://gemini.google.com/app --account myaccount

# Options:
#   --wait, -w    Seconds to wait for login (default: 120)
#   --channel, -c Browser: chrome, msedge, chromium (default: chrome)

# Step 2: List saved accounts
uv run browser.py accounts

# Step 3: Use saved account with commands
# Manual browsing (opens browser for user interaction)
uv run browser.py open https://gemini.google.com/app --account myaccount

# Automation (headless, returns content)
uv run browser.py auto https://gemini.google.com/app --account myaccount
uv run browser.py screenshot https://gemini.google.com/app -o gemini.png --account myaccount

存储位置

身份验证信息存储在用户的主目录 ~/.auth/ 中：

~/.auth/
├── account-name.json          # Storage state (cookies, localStorage)
└── profiles/
    └── account-name/          # Persistent browser profile

此位置由所有项目共享，且不会提交到 git。

最佳实践

• 自动化任务使用 headless=True
• 调试或需要目视确认时使用 headless=False
• 使用后务必关闭浏览器
• 尽可能优先使用特定选择器（id、data-testid），而非通用选择器
• 为动态内容添加适当的等待
• 使用 try/except 块妥善处理超时

常用选择器

• 按 ID：#element-id
• 按类：.class-name
• 按文本：text=Button Text
• 按角色：role=button[name="Submit"]
• 按 CSS：div.container > p.content
• 按 XPath：xpath=//div[@class='item']