sixb-ai/sixb/browser

Use a Node .cjs script with require("/opt/sixb/js/node_modules/playwright"). The matching Chromium headless shell is already installed at $PLAYWRIGHT_BROWSERS_PATH; use chromium.launch({ headless: true }). Do not download another browser or select an uninstalled Chrome channel. Close the browser in finally and set bounded navigation/action timeouts.

Local HTML can be loaded with page.setContent(html) or a file URL without external network access. External pages, images, fonts, and API calls depend on the sandbox network policy. Report a blocked source instead of repeatedly reinstalling tools. Requests and Beautiful Soup are also available for fetching permitted static pages and parsing HTML. Neither provides a search index.

Save screenshots and PDFs inside the workspace. Use view_file to inspect screenshots when image input is supported; check the resulting page state after interaction. Before printing a PDF, wait for fonts and relevant images, set print styles/page size, and inspect the resulting PDF page previews with sixb-render. A successful navigation or screenshot does not prove the page meets the task.

Deliver only requested final files: each with publish_artifact as a file artifact when you have that tool, otherwise as your instructions say. Browser traces, temporary HTML, and screenshots used for verification are working files unless requested.