In the plugin ecosystem of DeepSeek Harness (DSH), enabling an agent to browse and interact with web pages requires solving two problems: how to obtain real page content, and how to manipulate page elements. dsh-browser introduces bsk (browser-skill) and provides DSH with an interface to directly operate a real Chromium browser.

What Is This?

ylwl1997/dsh-browser is a DSH plugin. Its core value lies in connecting DSH with bsk, allowing the agent to navigate web pages, click elements, fill out forms, and take screenshots like a human. This solves the pain point that pure API simulation cannot obtain real page states while preserving the login state.

Core Features

The plugin provides the following tools:

  • browse_navigate: Opens a specified URL.
  • browse_snapshot: Reads the page accessibility tree (including @eN references).
  • browse_click: Clicks page elements using @eN references or CSS selectors.
  • browse_fill: Enters text into an input field.
  • browse_screenshot: Takes a screenshot of the current page.
  • browse_observe: Performs semantic VOM observation.

Installation and Enabling

Before use, two prerequisites must be met:
1. Ensure that bsk is included in the system environment variable PATH.
2. Load the browser skill extension in DSH.

Once the prerequisites are met, install it by running the following command:

dsh plugin add dsh-browser

Typical Usage

Based on the example in the README, a typical instruction is as follows:

Open https://example.com, read the page, and tell me the title.

This instruction triggers browse_navigate to open the page, then uses browse_snapshot to read the page content and extract the title.

Notes

  • Dependency environment: Ensure that bsk is in PATH and that the browser skill extension is loaded.
  • Runtime permissions: The plugin runs with the permissions of the current DSH process and involves browser operations, so it is recommended to inspect the source code before deployment.
  • License: Usage is governed by the MIT License.

Summary

dsh-browser brings the browser automation capabilities of bsk into DSH, giving agents the ability to operate a real browser. If you need DSH agents to perform complex web interaction tasks, this is a usable tool.