AI Agent Hub
Back to plugins
🤖

dsh-browser-runtime

Model Inference Updated 2026.08.27

Run the following command in DeepSeek Harness:

dsh plugin install LHKong7/dsh-browser-runtime

Paste the following prompt into your AI chat to install this plugin:

Fetch from https://github.com/LHKong7/dsh-browser-runtime and run dsh plugin install LHKong7/dsh-browser-runtime to mount the isolated browser runtime, Playwright Provider, and browser_* tools onto your DSH Agent in one bundle.

About this plugin

Give an existing DSH Agent real browser capabilities without bolting on a separate orchestration layer. dsh-browser-runtime ships as a single bundle that mounts a Chromium runtime, a Playwright Provider, eighteen browser_* tools, and usage guidance onto the Agent you already have—covering navigation, clicking, form filling, scrolling, waiting, screenshots, and structured extraction.

Rather than a loose set of Playwright calls, the runtime owns Agent isolation, browser lifecycle, serialized operations, stale-reference checks, resumable checkpoints, before-and-after transition evidence, and a credential-safe fill channel. Each Agent object gets its own BrowserContext so parallel Agents never share cookies, navigation state, or element refs. Observations are ranked into tiers—form controls first, then pagination, site navigation, record titles—so the model references fresh refs from the latest observation instead of hand-writing CSS selectors or executing JavaScript.

Built for developers who need an Agent that reads dynamic pages, fills forms, extracts lists or tables, saves screenshots, or runs multiple browser tasks in parallel with fully isolated state. The Provider interface is replaceable, so swapping in another browser engine later keeps the Agent-facing contract stable.

Use Cases

  • End-to-end web tasks: navigate, fill forms, extract data, and save screenshots without manual orchestration
  • Run multiple browser tasks in parallel with fully isolated cookies, navigation state, and element refs per Agent
  • Extract lists, tables, or articles from dynamic pages using ranked observation refs instead of hand-written selectors

Best For

  • Developers giving a DeepSeek Harness Agent real browser capabilities
  • Engineering teams running parallel browser tasks with strict state isolation
  • App developers who prefer observation-ref interaction over maintaining large selector libraries