Show HN：BrowserAct——面向 AI Agent 的浏览器层
英文标题：🔶 Show HN: BrowserAct: Browser Layer for Your AI Agent
BrowserAct 是一款面向 AI Agent 的浏览器自动化 CLI，主打反反爬、人工接管、多账号隔离及并发任务执行。它通过环境、执行和人工三层机制应对网站拦截，并提供 Chrome 复用、隐私模式和固定身份 stealth 等浏览器模式。该项目为 LLM 推理设计了索引式交互、紧凑文本输出和语义记忆，降低 Agent 操作浏览器时的 DOM 解析与 Token 开销。基础功能大多免费且开源，收费项目主要包括托管代理以及超过 5 个的 stealth 浏览器；此外，Skill Forge 可为重复性数据采集生成可部署的 Skill。
8.7 KB
3.2 KB
2.7 KB
1.2 KB
1.3 KB
1.7 KB
1.3 KB
2.0 KB
58.5 KB
中文
专为 AI Agent 打造的浏览器自动化 CLI。可突破反机器人拦截墙；Agent 卡住时，可在不同平台间交接给人工；支持并行运行任务且不会相互污染；还能在相互独立的浏览器中隔离多个账号。
AI Agent 所需的浏览器，必须能抵达标准工具无法访问的地方；在 Agent 卡住时让人工无缝接管；防止并行任务相互污染；并且应为 LLM 推理而非人工编写的脚本而设计。面向 Agent 的浏览器必须做好四件事。
1. 突破拦截——三个渐进层级
- 环境层——隐身指纹伪装、TLS 轮换、代理切换。绝大多数拦截根本不会被触发。
- 执行层——
solve-captcha自动解决 CAPTCHA；stealth-extract用一条命令提取受保护页面。 - 人工层——
remote-assist生成实时 URL；用户可从任何设备接管，完成后 Agent 无缝继续执行。
2. 三种浏览器模式——按真实世界场景划分
| 模式 | 场景 | 关键特性 |
|---|---|---|
chrome |
复用本地 Chrome 登录状态 | 导入 Profile 或通过 CDP 附加 |
stealth 隐私模式 |
无需登录的无阻力批量抓取 | 每个会话使用全新指纹 + 代理轮换，不留痕迹 |
stealth 固定身份 |
已登录账号 · 多浏览器并行 | 稳定指纹 + 稳定 IP，账号身份稳定，不会被标记为机器人 |
3. 零干扰并发——每个 Agent 各行其道
- 跨浏览器并行——独立的 Cookie、指纹和代理。网站无法将它们关联起来。
- 同浏览器多会话——共享登录状态，独立执行，任务彼此不会阻塞。
- 隐私模式——每个会话都有全新指纹和空白 Profile，结束后不留痕迹。
4. 为 Agent 推理设计——而非为人工脚本设计
- 紧凑文本输出——索引化文本格式，相比 JSON 或 HTML 的 Token 效率高出数倍。
- 索引式交互——
state返回索引列表；click 3/input 2 "..."。无需解析 DOM。 - 语义记忆——每个浏览器都带有一个
desc，通过语义与任务匹配。 - 并发安全——会话所有权 + 显式命名。多 Agent 操作永不冲突。
安全性：确认门控——敏感操作（创建/删除浏览器、导入 Profile、修改代理、安全与隐私开关）均需用户明确批准。此前的批准不会延续。该机制在 Skill 层强制执行，而不是一个配置开关。
- 更好的无头模式——默认无头运行且不干扰用户；隐身无头模式不会被检测。
- 跨平台远程交接——任何设备均可打开链接接管，之后 Agent 无缝继续执行。
告诉你的 AI Agent：
安装 browser-act。Skill 源码：https://github.com/browser-act/skills/tree/main/browser-act 。安装后验证其是否正常工作。
# Extract protected page content (zero config)
browser-act stealth-extract https://example.com
# Full browser automation
browser-act --session my-task browser open <id> https://example.com
browser-act --session my-task state # See clickable elements
browser-act --session my-task click 3 # Click by index
browser-act --session my-task input 2 "hi" # Type into a field
Agent 会在每个会话开始时运行 get-skills——通过一次调用获取环境状态、浏览器列表和命令：
browser-act get-skills core --skill-version 2.0.2
**操作系统：**Windows、macOS、Linux
**Agent：**Claude Code · Cursor · VS Code · OpenCode · OpenClaw · Codex · Gemini CLI——适用于任何能够执行 Shell 命令并加载 Skills 的 Agent。
几乎所有功能都免费。只有两项功能需要付费：托管代理（Dynamic / Static）以及超过前 5 个额度的 stealth 浏览器。
功能 | 免费
（无需注册） | 免费
（仅需登录） | 付费
---|---|---|---
浏览器自动化、Chrome / Chrome-direct | ✓ | ✓ | ✓
Stealth 浏览器（≤ 5）、stealth-extract、solve-captcha、remote-assist、隐私模式、Skill Forge | — | ✓ | ✓
Stealth 浏览器（> 5）、Dynamic / Static 代理 | — | — | ✓
完整文档涵盖反拦截、浏览器模式、会话与并发、无头模式与远程交接、Agent 设计、Skills 系统以及完整命令参考。
Skill Forge——你的专属数据抓取工程师
需要大规模重复从同一网站提取数据？无需手写爬虫。Skill Forge 会探索网站一次，发现其 API 和数据模式，生成可部署的 Skill 包，然后无需再次探索即可稳定运行——无论是 500 条还是 5,000 条记录，都通过同一条稳定路径处理。
任何网站。任何数据。一条命令即可开始：
安装 browser-act-skill-forge。Skill 源码：https://github.com/browser-act/skills/tree/main/browser-act-skill-forge 。安装后验证其是否正常工作。
然后告诉你的 Agent 你需要什么：
“打造一个从 LinkedIn 提取职位列表的 Skill——职位名称、公司、薪资、URL。之后我会运行 300 个关键词。”
已有超过 30 个由 Skill Forge 生成的预构建 Skill，可随时安装和运行。覆盖 Amazon、Google Maps、YouTube、Reddit、微信、知乎等。
找不到上述所需内容？几分钟内即可为任何网站生成自定义 Skill——无需编程。只需描述你想获取的数据或希望执行的操作，Skill Forge 会处理其余工作。
BrowserAct Skills 免费且开源。如果它为你节省了时间，请给我们一个 ⭐ Star——这有助于项目持续运营，并帮助我们发布更多 Skill。
🎁 **额外福利：**在你 Star 该仓库后，可以加入我们的 Discord，并在 #claim-500-credits 频道发帖，即可获得 500 免费积分！
🤝 社区与支持
由 BrowserAct Team 倾心打造 ❤️
English / 原文
Browser automation CLI built for AI agents. Get past anti-bot walls, hand off to humans across platforms when stuck, run parallel tasks without cross-contamination, and isolate multiple accounts in independent browsers.
The browser an AI agent needs has to reach places standard tools can't, let a human seamlessly take over when the agent is stuck, keep parallel tasks from cross-contaminating, and be designed for LLM reasoning — not human-written scripts. A browser for agents must get four things right.
1. Break through blocks — three progressive layers
- Environment layer — stealth fingerprint spoofing, TLS rotation, proxy switching. The vast majority of blocks never trigger.
- Execution layer —
solve-captchaauto-solves CAPTCHAs;stealth-extractpulls protected pages in one command. - Human layer —
remote-assistgenerates a live URL; the user takes over from any device, and the agent continues seamlessly when done.
2. Three browser modes — by real-world scenario
| Mode | Scenario | Key trait |
|---|---|---|
chrome |
Reuse local Chrome login state | Profile import or CDP attach |
stealth privacy mode |
Frictionless batch scraping without login | Fresh fingerprint per session + proxy rotation, zero residue |
stealth fixed identity |
Logged-in accounts · multi-browser parallel | Stable fingerprint + stable IP, stable account identity, not flagged as bots |
3. Zero-interference concurrency — every agent in its own lane
- Cross-browser parallel — independent cookies, fingerprints, proxies. Sites cannot correlate them.
- Same-browser multi-session — shared login state, independent execution, tasks don't block each other.
- Privacy mode — fresh fingerprint and empty profile per session, zero residue when done.
4. Designed for agent reasoning — not human scripts
- Compact text output — indexed text format, several times more token-efficient than JSON or HTML.
- Indexed interaction —
statereturns an indexed list;click 3/input 2 "...". No DOM parsing required. - Semantic memory — every browser carries a
desc, matched to tasks by meaning. - Concurrency-safe — session ownership + explicit naming. Multi-agent operation never conflicts.
Security: confirmation gating — sensitive operations (browser create / delete, Profile import, proxy changes, security and privacy toggles) require explicit user approval. Prior approvals do not carry over. Enforced at the Skill layer, not a configuration toggle.
- Better headless — Default headless without disrupting users; stealth headless that isn't detected.
- Cross-platform remote handoff — Any device opens the link to take over, and the agent continues seamlessly.
Tell your AI agent:
Install browser-act. Skill source: https://github.com/browser-act/skills/tree/main/browser-act . Verify it works after installation.
# Extract protected page content (zero config)
browser-act stealth-extract https://example.com
# Full browser automation
browser-act --session my-task browser open <id> https://example.com
browser-act --session my-task state # See clickable elements
browser-act --session my-task click 3 # Click by index
browser-act --session my-task input 2 "hi" # Type into a field
The agent runs get-skills at the start of each session — gets environment state, browser list, and commands in one call:
browser-act get-skills core --skill-version 2.0.2
How agents discover and use BrowserAct →
OS: Windows, macOS, Linux
Agents: Claude Code · Cursor · VS Code · OpenCode · OpenClaw · Codex · Gemini CLI — works with any agent that can execute shell commands and load Skills.
Almost everything is free. Only two features require payment: managed proxies (Dynamic / Static), and stealth browsers beyond the first 5.
Feature | Free
(No Signup) | Free
(Login Only) | Paid
---|---|---|---
Browser automation, Chrome / Chrome-direct | ✓ | ✓ | ✓
Stealth browser (≤ 5), stealth-extract, solve-captcha, remote-assist, privacy mode, Skill Forge | — | ✓ | ✓
Stealth browser (> 5), Dynamic / Static proxy | — | — | ✓
Full documentation covers anti-blocking, browser modes, sessions and concurrency, headless and remote handoff, agent design, the Skills system, and the complete command reference.
Skill Forge — Your Personal Scraping Engineer
Need to extract data from the same website repeatedly at scale? Don't write scrapers by hand. Skill Forge explores a site once, discovers its APIs and data patterns, generates a deploy-ready Skill package, then runs reliably without re-exploration — 500 or 5,000 records through the same stable path.
Any website. Any data. One command to start:
Install browser-act-skill-forge. Skill source: https://github.com/browser-act/skills/tree/main/browser-act-skill-forge . Verify it works after installation.
Then tell your agent what you need:
"Forge a Skill that extracts job listings from LinkedIn — title, company, salary, URL. I'll run 300 keywords later."
30+ pre-built Skills already generated by Skill Forge, ready to install and run. Covers Amazon, Google Maps, YouTube, Reddit, WeChat, Zhihu, and more.
Browse the full Solutions Catalog →
Can't find what you need above? Generate a custom Skill for any website in minutes — no coding required. Just describe what data you want or what action to perform, and Skill Forge handles the rest.
BrowserAct Skills is free and open source. If it saves you time, please give us a ⭐ Star — it keeps the project alive and helps us ship more skills.
🎁 Bonus: Once you star the repository, you can join our Discord and post in the #claim-500-credits channel to receive 500 free credits!
🤝 Community & Support
Built with ❤️ by the BrowserAct Team