Skip to content
  • Home
  • Plugin manual
  • dsh-browseruse

yzd6552-commits/dsh-browseruse

30Last commit Aug 14, 2026

dsh-browseruse DSH plugin

This plugin enables web automation agents to browse the web like a human. It uses playwright-core to control a dedicated Google Chrome instance with independent profile and persistent cookies. It provides fine-grained tools like open, snapshot, click, type, scroll, navigate, tabs, fill form, download, as well as autonomous task and scheduling capabilities.

browser-web

How to install the dsh-browseruse DSH plugin

dsh plugin --profile web add github:yzd6552/dsh-browseruse

Copying does not run this command. Review the repository and version before installing the dsh-browseruse DSH plugin.

dsh-browseruse DSH plugin data source

dsh-browseruse DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-browseruse DSH plugin can do

  • Fine-grained browser tools: browser_open, browser_snapshot, browser_click, browser_type, browser_scroll, browser_navigate, browser_tabs, browser_fill_form, browser_download
  • Autonomous task execution via browser_task: agent internally loops 'see page → think → act' with a default 30-step limit, trackable via job_output
  • Scheduled tasks via browser_schedule: set a goal and a time (HH:MM, ISO, or seconds) to run automatically within the session
  • Dangerous operation confirmation: automatically detects sensitive actions like payment, order, delete, and asks user before proceeding
  • CAPTCHA human takeover: pauses on CAPTCHA detection, shows screenshot, user can handle in the dedicated Chrome window then resume
  • Persistent login state: saves cookies in an independent profile at $DSH_HOME/.browseruse/chrome-profile, login once and remember forever

Where the dsh-browseruse DSH plugin fits

  • Automating form filling and submission on websites
  • Scheduling timed tasks like flash sales or ticket grabbing
  • Scraping web content with human-like interaction to avoid detection
  • Testing web applications by driving a real browser with step-by-step commands
  • Allowing an AI agent to browse and summarize information from multiple pages

Who the dsh-browseruse DSH plugin is for

  • DeepSeek Harness users who need browser automation capabilities
  • Developers and power users wanting to automate web tasks with natural language goals
  • Users who need scheduled browser actions within a DSH session

dsh-browseruse DSH plugin limitations

  • Requires a local installed Google Chrome browser (not Chromium) and Node.js >= 20
  • Scheduled tasks are session-scoped; they are lost after DSH restart and must be re-registered
  • Dangerous operation confirmation and CAPTCHA handling depend on the session's ask_user_question channel; sub-agent contexts without it may block risky actions
  • The plugin only manages a dedicated Chrome instance; it does not affect your everyday Chrome browser

dsh-browseruse DSH plugin: from the repository README

Quoted from the yzd6552-commits/dsh-browseruse README, the upstream source of the dsh-browseruse DSH plugin. Copyright remains with the original authors.

browser-use 风格的浏览器自动化插件，专为 [DeepSeek Harness (DSH)](https://github.com/deepseek-ai/deepseek-harness) 打造。 通过 [playwright-core](https://www.npmjs.com/package/playwright-core) 驱动一个**专用 Google Chrome 实例**（独立资料目录、cookie 持久），像真人一样"看页面 → 决策 → 操作"，不影响你日常使用的 Chrome。 > A browser-use style automation plugin for DeepSeek Harness. It drives a dedicated Google Chrome instance (independent profile, persistent cookies) through playwright-core, so an agent can browse the web like a human — without touching your everyday Chrome. ## ✨ 功能 / Features | 能力 | 说明 | |---|---| | 👀 细粒度工具 | `browser_open` / `browser_snapshot`（文本+元素序号+截图）/ `browser_click` / `browser_type` / `browser_scroll` / `browser_navigate` / `browser_tabs` / `browser_fill_form` / `browser_download` | | 🧠 自主任务 | `browser_task("自然语言目标")` —— 插件内部循环"看页面 → 调用当前模型决策 → 操作"，默认上限 30 步，后台运行可用 `job_output` 跟踪 | | ⏰ 定时任务 | `browser_schedule("目标", "HH:MM" / ISO 时间 / 秒数)` —— 到点自动开跑（抢购/抢票场景），本会话内有效 | | 🛡️ 危险操作确认 | 支付/下单/删除/注销等敏感动作自动识别，动手前弹窗征求用户同意 | | 🧩 验证码人工接管 | 检测到验证码时暂停并截图询问，用户可在专用 Chrome 窗口亲自处理后继续 | | 🖼️ 截图回传 | 每步操作截图进入对话（模型支持图片输入时） | | 🍪 登录态持久 | 独立 profile（默认 `$DSH_HOME/.browseruse/chrome-profile`），登录一次长期记住，密码不落盘 | ## 📦 安装 / Install 需要本

Read the full READMERepository license: MIT

dsh-browseruse DSH plugin questions

How do I install dsh-browseruse as a profile bundle?

Run `dsh plugin --profile web add github:yzd6552/dsh-browseruse` or `dsh plugin --profile web add dsh-browseruse`. After installation, restart `dsh web` for the tools to become available. All browser_* tools will be visible in that profile's sessions.

What prerequisites are needed for this plugin?

You need Node.js version 20 or higher, and a locally installed Google Chrome browser (not Chromium). The plugin connects to Chrome via the DevTools Protocol (CDP), so it does not download a separate browser binary.

How can I keep my login session persistent across restarts?

The plugin saves cookies in an independent Chrome profile at `$DSH_HOME/.browseruse/chrome-profile` (defaults to `~/.dsh/...`). Simply log in once in the dedicated Chrome window, and the cookies will be remembered for future sessions. Passwords are never stored on disk.

Can I schedule a task to run at a specific time every day?

Yes, use `browser_schedule` with a goal and a time in HH:MM format (e.g., "20:00:00"). However, scheduled tasks are only valid within the current DSH session. If you restart DSH, you need to re-register the schedule. The plugin does not support persistent cron-like scheduling yet.

How does the plugin handle CAPTCHAs?

When a CAPTCHA is detected, the plugin pauses and sends a screenshot to the user. The user can then manually solve the CAPTCHA in the dedicated Chrome window and resume the work. This relies on the session's ask_user_question channel; if the channel is not available in a sub-agent context, the action may be blocked.