dsh-b2us-chrome-tool

The hard part of letting an Agent drive a real browser is not the automation itself but the security boundary that must hold while the Agent clicks, types, and reads pages. dsh-b2us-chrome-tool solves exactly that problem: through a Manifest V3 extension and an authenticated 127.0.0.1 loopback bridge it gives DeepSeek Harness agents control over the Chrome session you already use, while every byte of page data stays locked inside an isolated Browser Worker context that the parent Agent can only see through tightly typed, bounded results.

The capability surface covers the full browser-automation loop. Semantic accessible names, roles, and states replace raw page-source dumps. page.find ranks elements inside the page runtime before anything reaches the model, and page.locator generates CSS and XPath candidates that it proves are unique in the live DOM. Interaction primitives include click, exact-name click, non-sensitive text input, select, scroll, wait, and structured extraction. Screenshots and bounded Fetch or XHR metadata arrive as opaque artifacts with redaction, quotas, and expiry. Browser execution and file delivery are reported so that a failed artifact copy can be retried without repeating a login or form submission.

The plugin is aimed at developers and small teams working in local or trusted environments who want an auditable, recoverable, and bounded layer of browser interaction for their agents. It does not replace a Selenium or Playwright CI pipeline; it gives the Agent a capability that is scoped, observable, and safe by construction, so you enjoy automation without handing over an unlimited DOM handle.