Token导航 LogoToken导航TokenDH.com
研究检索敏感数据github未标认证来源可访问许可证需确认审计异常

real-browser真正的浏览器

Agent Skill

real-browser 用于处理浏览器自动化、网页检查和页面信息提取,适合在 Codex、Claude、Cursor、Gemini CLI 中需要让 Agent 打开页面、读取网页或验证前端流程时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

364

周安装

15

GitHub Stars

5

下载量

119
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:real-browser(真正的浏览器)
来源仓库:https://github.com/xxww0098/skills-hub
仓库路径:skills/real-browser
安装命令:
npx skills add https://github.com/xxww0098/skills-hub --skill real-browser
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/xxww0098/skills-hub --skill real-browser

简介

real-browser 用于处理浏览器自动化、网页检查和页面信息提取,适合在 Codex、Claude、Cursor、Gemini CLI 中让 Agent 打开页面或读取内容。

  • 适用于前端测试、数据抓取和页面验证等场景,可模拟用户操作并提取关键信息。
  • 通过 npx skills add 命令从 GitHub 仓库安装,需确认权限范围和维护状态。
  • 建议使用前核实是否会触发联网、命令执行或文件读写,避免安全风险。
  • 适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

SKILL.md

Real Browser — Login-State CDP Bridge

This skill = launch Chrome with login state + CDP port. agent-browser skill = ALL browser commands (click, fill, type, scroll, eval, etc.). You MUST load BOTH skills to operate a browser.

Step 0: Load the agent-browser Skill

Before running ANY browser command, you MUST read and load the agent-browser skill. It contains the full command reference with 50+ commands you will need: click, fill, type, press, scroll, select, hover, check, eval, snapshot, diff, screenshot, download, upload, auth, sessions, and more.

Without loading the agent-browser skill, you will not know how to interact with the browser. This skill only teaches you how to LAUNCH and CONNECT.

Step 1: Launch Chrome

# Default: headless on port 9851 (recommended for automation)
scripts/real_browser.sh

# Visible window (only if user explicitly asks to see the browser)
scripts/real_browser.sh --headed

# Custom port (headless)
scripts/real_browser.sh 9888

# Visible window on custom port
scripts/real_browser.sh --headed 9888

# Incremental updates: preserve login state between runs (faster startup)
scripts/real_browser.sh --keep-profile

# Custom logging level (debug, info, warn, error)
scripts/real_browser.sh --log-level debug

# Machine-readable output (for scripting): outputs CDP_URL=<url>
scripts/real_browser.sh --machine-readable

# Combine options as needed
scripts/real_browser.sh --keep-profile --log-level info --headed 9852

Environment Variables for Advanced Customization

VariablePurposeExample
REAL_BROWSER_BINOverride the Chrome binary path (e.g., for Brave, Edge, or custom Chrome builds)REAL_BROWSER_BIN=/usr/bin/brave-browser
REAL_BROWSER_EXTRA_ARGSAdditional arguments passed to Chrome (space-separated)REAL_BROWSER_EXTRA_ARGS="--no-sandbox --disable-gpu --disable-dev-shm-usage"
REAL_BROWSER_LOG_LEVELSet the logging level (debug, info, warn, error)REAL_BROWSER_LOG_LEVEL=warn
REAL_BROWSER_CDP_TIMEOUTSet the CDP readiness timeout in units of 0.25 seconds (default 160 = 40s)REAL_BROWSER_CDP_TIMEOUT=200 (50s)

Example with environment variables:

REAL_BROWSER_BIN=/opt/google/chrome/chrome \
REAL_BROWSER_EXTRA_ARGS="--no-sandbox --disable-web-security" \
REAL_BROWSER_LOG_LEVEL=debug \
scripts/real_browser.sh --machine-readable 9851

Step 2: Interact via agent-browser --cdp <port>

After launch, prepend --cdp <port> to every agent-browser command. This is the ONLY difference from standard agent-browser usage. All commands, patterns, and workflows from the agent-browser skill apply as-is.

agent-browser --cdp 9851 open https://x.com
agent-browser --cdp 9851 snapshot -i
agent-browser --cdp 9851 click @e3
💡 Note: If you used a custom port when launching (e.g., scripts/real_browser.sh 9888), then use --cdp 9888 in all agent-browser commands.

Visibility Guard

If commands fail or return unexpected results, the CDP target may be on a chrome:// internal page. Diagnose and fix:

agent-browser --cdp 9851 get url       # what page am I on?
agent-browser --cdp 9851 tab list      # list all tabs
agent-browser --cdp 9851 tab <index>   # switch to the correct tab

Login State Notes

Site typeExamplesLogin state
Normal sitesX, Reddit, GitHub, 知乎, V2EX✅ Preserved
High-security (DBSC/Keychain)Google, Claude, ChatGPT⚠️ May log out
⚠️ Note: When using --keep-profile, the login state is preserved between runs. Over time, the profile directory (~/.chrome-cdp-profile) may grow in size. Periodically remove it to reset the state if needed.

Rules

  1. --cdp <port> on EVERY agent-browser command. No exceptions.
  2. Load the agent-browser skill for all available commands.
  3. open → always wait --load networkidle immediately after.
  4. snapshot -i before any interaction to get fresh refs.
  5. Re-snapshot after any page navigation or DOM change.
  6. Never mix chrome-devtools tool with agent-browser.
  7. Port is 9851 unless user passed a custom port to the launch script.

Troubleshooting

lsof -iTCP:9851 -sTCP:LISTEN    # find process on port (macOS / most Linux)
ss -tlnp | grep :9851            # alternative if lsof is unavailable (Linux)
kill <pid>                       # free the port
rm -f /tmp/.real_browser.lock    # remove stale lock
scripts/real_browser.sh          # re-launch fresh
🔍 Additional checks: - If using environment variables, verify they are set correctly (e.g., echo $REAL_BROWSER_EXTRA_ARGS). - If Chrome fails to launch, check the binary path and extra arguments. - For machine-readable output, ensure no other logging interferes (use --log-level error to minimize noise).

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.12%
按下载量换算41

Claude

31.39%
按下载量换算37

Cursor

17.27%
按下载量换算21

Gemini CLI

8.55%
按下载量换算10

安全审计

Gen Agent Trust Hub

未通过

Socket

可疑

Snyk

可疑

权限和风险

敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills