dsh-browser

Many DeepSeek Harness tasks require an agent to interact with a real browser session - logging into enterprise dashboards, filling forms that depend on cookies, or capturing pages that carry an authenticated state. Pure API calls or headless rendering cannot cover these scenarios, and that is exactly the gap dsh-browser fills. It runs on bsk and plugs a genuine Chromium browser (with existing user logins intact) into the agent toolchain, so automation is no longer confined to a sandbox.

The plugin ships six core tools: browse_navigate (open any URL), browse_snapshot (read the page as an accessibility tree with @eN element references), browse_click (target an element by reference or CSS selector), browse_fill (type into an input), browse_screenshot (capture the current page), and browse_observe (semantic VOM-based page understanding). Together they let an agent complete the full interaction loop - open a page, comprehend its content, locate a button, click it, submit a form, and verify the result - just like a human would.

It is best suited for developers whose workflows depend on real-browser, end-to-end operations, especially tasks that rely on user login states, multi-step form interactions, or visual verification. As long as bsk is installed and the browser-skill extension is loaded, the plugin is ready to use immediately.