New Env: Online-Mind2Web

[Genteki]

Original #156

A hud remote-browser based environment for Online-Mind2Web dataset.

---

Note

Introduces a new Online-Mind2Web environment with a remote-browser MCP server, multi-provider support, setup/eval tools, recording, and Docker packaging.

• Environment: environments/online_mind2web
  • Adds pyproject, Dockerfile, README, .gitignore, and test_task.json.
• MCP Server (src/hud_controller/server.py)
  • Boots a remote-browser MCP server, attaches a persistent context, initializes provider, Playwright, and computer tools; exposes telemetry and graceful shutdown.
• Persistent Context (context.py)
  • New context server to persist browser/provider state, telemetry, and Playwright handle across hot-reloads.
• Providers (providers/)
  • Implements AnchorBrowserProvider, BrowserBaseProvider, HyperBrowserProvider, SteelProvider with launch/close/status APIs and live view URLs; registry and proxy helper.
• Tools (tools/)
  • BrowserExecutor: maps computer actions to Playwright.
  • AnthropicComputerToolWithRecord and OpenAIComputerToolWithRecord: add screenshot/history recording to /screenshot and /action_history.
• Setup Hub (setup/)
  • Tools for navigation, cookies, and basic interactions (navigate_to_url, set_cookies/clear_cookies, click_element/fill_input/select_option).
• Evaluation Hub (evaluate/)
  • webjudge (multi-screenshot + keypoint analysis via GPT-4o), autonomous_eval (single-screenshot VLM check), and overall_judge (aggregate) returning EvaluationResult.
• Problems (problems/)
  • Registry/decorator plus sample problems: navigate/verify, form interaction, button click, Google search.
• Packaging & Run
  • Docker image to start context + MCP server; script entry hud-om2w; instructions to eval local JSON or HuggingFace dataset.

^{Written by Cursor Bugbot for commit f70d067. This will update automatically on new commits. Configure here.}

[promptless]

📝 Documentation updates detected!

New suggestion: Add comprehensive Online-Mind2Web environment documentation for PR #168
Updated existing suggestion: Add comprehensive Mind2Web evaluation documentation (updated for PR #156)