In the development and debugging of DeepSeek Harness (DSH), operations and state tracking in browser environments are often cumbersome. This plugin aims to provide a complete set of web debugging and operation capabilities, including independent browser windows, multi-tab management, accessibility snapshots, and console and network data collection.

Plugin Purpose

This is a plugin that provides client-side web operation and debugging capabilities for DeepSeek Harness. Through independent browser windows or tabs, it gives agents capabilities such as multi-session management, element manipulation, snapshot reading, and network data collection. The plugin is maintained in the community catalog and is integrated into a profile through cordis.patch.yml.

Core Features

  • Independent browser windows and multiple tabs: Supports opening new controlled tabs, navigation, forward/back in history, and tab activation, closing, claiming, and handover.
  • Element operations and waiting: Provides operations such as clicking, filling, key pressing, and scrolling, and supports waiting for text to appear, elements to disappear, the page to stabilize, or a specified amount of time.
  • Debugging and collection: Supports accessibility snapshots, screenshots, console log collection, and network request capture.
  • Advanced capabilities: Provides ref operations to restore old references or obtain real-time coordinates, supports manual takeover mode, and executes advanced commands through CDP / Electron provider.

Installation and Activation

The plugin is integrated into a profile through cordis.patch.yml. The desktop edition uses the Electron provider by default, while external Chrome uses the CDP provider (disabled by default in the factory configuration).

Installation steps are as follows:

  1. First use on desktop edition:
    Download dsh-webops-desktop-v<版本>-win-x64-portable.zip from GitHub Releases. After extracting, run 启动.cmd.

  2. Existing desktop portable version:
    Download dsh-webops-plugin-update-v<版本>.zip, overwrite the root directory of the portable version while keeping home, then restart.

  3. If you already have your own dsh:
    Download dsh-webops-plugin-v<版本>-win-x64-portable.zip, and install it into the profile according to the plugin installation instructions.

Typical Usage

  • Opening and navigation: Use webpage_open or webpage_navigate to open a new controlled tab and navigate or move forward/back in history.
  • Reading page information: Use webpage_snapshot to read the page outline, or use webpage_find to search the cache.
  • Element operations: Use webpage_click, fill, press, or scroll to manipulate elements, or use webpage_locate to obtain real-time coordinates to restore an old ref.
  • Waiting and collection: Use webpage_wait to wait for the page to stabilize or for a specific condition, and use webpage_screenshot, console, or network for debugging and collection.
  • CDP execution: Use webpage_execute to execute one CDP command from the allowlist.

Applicable Scenarios and Notes

  • Tab exclusivity: Each controlled tab can be occupied by at most one conversation at a time. Popups inherit the ownership of their family.
  • Permissions and takeover: Supports manual takeover. During takeover, the Agent cannot operate; after handback, it regains a valid ref. Restarting does not restore permissions from previous conversations.
  • Security notice: Page content should be treated as untrusted data. Runtime.evaluate will execute page code, so use it cautiously.
  • Scope: Suitable for agent development scenarios that require fine-grained control of browser behavior, web page automated testing, or debugging.

By providing a standardized toolset and dual-provider architecture, this plugin lowers the operational barrier for DeepSeek Harness in browser environments. For more details and source code, see its GitHub repository.