New Env: Online-Mind2Web #168
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
📝 Documentation updates detected! New suggestion: Add comprehensive Online-Mind2Web environment documentation for PR #168 |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
| # Note: Environment variables for browser providers should be set at runtime: | ||
| # - BROWSER_PROVIDER: anchorbrowser, steel, browserbase, hyperbrowser, kernel | ||
| # - Provider-specific API keys: ANCHOR_API_KEY, STEEL_API_KEY, etc. | ||
| # - GCP_CREDENTIALS_JSON: For Google Sheets functionality (if needed) |
There was a problem hiding this comment.
nit: remove this line, not relevant in OM2W
| class AnthropicComputerToolWithRecord(AnthropicComputerTool): | ||
| def __init__( |
There was a problem hiding this comment.
We should likely make this a first class tool in the SDK, but could be on a case by case basis with each environment.
Seems super valuable for LLM/VLM as judge.
2 participants
Original #156
A hud remote-browser based environment for Online-Mind2Web dataset.
Note
Introduces a Dockerized HUD environment for Online-Mind2Web with persistent context, multiple cloud browser providers, Playwright-based executor, setup/evaluation hubs, telemetry, and a test task.
environments/online_mind2web/withDockerfile,pyproject.toml, andREADME.mdto run a HUD remote-browser MCP server with persistent context (hud_controller.context) and main server (hud_controller.server).resourcetelemetry://liveand progress-enabled initialization; supports initial URL and graceful shutdown.AnchorBrowserProvider,BrowserBaseProvider,HyperBrowserProvider, andSteelProviderundersrc/hud_controller/providers/withBrowserProviderbase, status/telemetry, live view URLs, and proxy helper (helper/proxy.py).get_providerforBROWSER_PROVIDERselection.BrowserExecutorto drive Playwrightpagefor clicks/keys/scroll/drag and screenshots.AnthropicComputerToolWithRecordandOpenAIComputerToolWithRecord, saving screenshots to/screenshotand actions to/action_history.setup.navigate_to_urlfor navigation via Playwright.evaluate/:autonomous,webjudge, andoverall_judge(aggregates), leveraging OpenAI (gpt-4o) with screenshot(s) and action history.test_task.jsonand README instructions for running single tasks or HF dataset (Genteki/Online-Mind2Web).Written by Cursor Bugbot for commit 4c96563. This will update automatically on new commits. Configure here.