dsh-browser-tool

DeepSeek Harness agents are strong at text reasoning but lack a way to see what is actually on a web page. dsh-browser-tool closes that gap by letting the model open a Chromium tab, run JavaScript, interact with the DOM, take a screenshot, and receive both the text stream and the image block back in the conversation context, completing an observe-act-observe loop entirely within the agent session.

The plugin is built around a simple open / run / close contract. open supports four browser sources: an in-process headless Chromium (with optional stealth injection), a detached desktop app attached via its idle CDP port, an existing CDP endpoint, and a local Browser Relay that hijacks the user's live Chrome session with tab-group queuing and debugger-ban recovery. run executes an async function body inside the tab with full access to the Tab API (goto, observe, screenshot, click, type, fill, press, scroll, drag, waitFor, evaluate, uploadFile, and more). Screenshots are resized to the model pixel budget via sharp before being returned as image content blocks. Chromium resolution prefers system installs and existing Chrome for Testing builds from other tools' caches, downloading only as a last resort.

This plugin is aimed at developers who need web-level interaction inside an agent workflow: automated information extraction, end-to-end UI verification, competitive-page inspection, multi-step form filling, and the like. It ships as a self-contained npm package (no local build required) that mounts as a DSH profile bundle under the MIT license.