AGENT SKILL

browser-use

Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, or extract information from web pages.

License 许可
AGPL-3.0
最近更新
2026-06-17
Stars
★ 1
Forks
0

01 前言 Front Matter

name
browser-use
description
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, or extract information from web pages.

02 资源文件 4

03 SKILL.md 正文

Browser Automation with browser-use CLI

tabyAgent

  • Run every browser-use command via terminal_run (one per call, or chain with &&).
  • CLI is preinstalled in the image. Do not pip install unless browser-use doctor fails after an upgrade.
  • Docker has no display — always use default headless mode. Do not use --headed, connect, or --profile.
  • Anti-bot stealth is auto-applied at browser start. No extra step after open.

The daemon keeps the browser open between commands (~50ms per call).

Core workflow

  1. browser-use open <url>
  2. browser-use state — clickable elements with indices
  3. Interact by index (click, input, keys, …)
  4. browser-use state or browser-use screenshot to verify
  5. browser-use close when done (or leave open for the next command)

If a command fails: browser-use close, then retry.

Commands

# Navigation
browser-use open <url>
browser-use back
browser-use scroll down          # --amount N for pixels
browser-use scroll up

# Page state — run state first
browser-use state
browser-use screenshot [path.png]   # --full for full page

# Interactions — indices from state
browser-use click <index>
browser-use click <x> <y>
browser-use type "text"
browser-use input <index> "text"
browser-use input <index> ""        # clear field
browser-use keys "Enter"            # also "Control+a", etc.
browser-use select <index> "option"
browser-use upload <index> <path>
browser-use hover <index>
browser-use dblclick <index>
browser-use rightclick <index>

# Data
browser-use eval "js code"
browser-use get title
browser-use get html [--selector "h1"]
browser-use get text <index>
browser-use get value <index>
browser-use get attributes <index>

# Wait
browser-use wait selector "css"     # --timeout ms, --state visible|hidden
browser-use wait text "text"

# Session
browser-use close
browser-use close --all

Command chaining

browser-use open https://example.com && browser-use state
browser-use input 5 "user@example.com" && browser-use input 6 "password" && browser-use click 7

Chain when you do not need intermediate output. Run state separately first when you need to discover indices.

Blocked sites

Stealth patches apply on browser start. If still blocked:

  1. browser-use close then reopen (stale sessions may miss patches)
  2. Do not fall back to curl/wget — use browser automation or URL mirrors (see system prompt for X links)
  3. Captcha / login-only walls — tell the user; you cannot complete those for them

Tips

  1. Always run state before clicking — indices change after navigation
  2. Browser persists between commands in the same session
  3. Aliases: bu, browser, browseruse
  4. On failure: browser-use close then retry

Troubleshooting

  • Browser won't start → browser-use close then browser-use open <url>
  • Element not found → browser-use scroll down then browser-use state
  • Install issues → browser-use doctor

04 同仓库更多 Skills