dsh-plugin-chrome

When agents perform web browsing and automation tasks, users often struggle with a lack of real-time visualization, difficulty in manual intervention, or opaque browser operations. The dsh-plugin-chrome plugin solves these issues by launching a real, visible Chrome window for each DeepSeek Harness session. It allows agents to drive the browser directly through a suite of 16 chrome_* tools (such as open, navigate, screenshot, click, and fill) while streaming a live view of the window into a Chrome tab in the Web GUI. This enables users to observe every agent action in real time and manually take over control at any moment for seamless human-agent collaboration. The plugin also features accessibility-tree snapshots for stable element targeting, dual-channel screenshots (where images enter the model context and are saved to history), along with security isolation and automatic resource cleanup. It is ideal for developers, researchers, and automation enthusiasts who need browser automation, real-time monitoring of agent behavior, or web testing, especially in scenarios that combine agent intelligence with human oversight.