mirror of
https://github.com/browseros-ai/BrowserOS.git
synced 2026-05-20 20:39:10 +00:00
Standalone script that runs the BrowserOS agent on tasks and captures rich execution traces for website showcase galleries: - Before/after screenshots per tool call via AI SDK callbacks - CSS crosshair overlays injected via browser.evaluate() for element-targeting tools (click, fill, hover, etc.) - Accessibility snapshot capture at each step - Per-task JSON manifests with execution metadata - R2 upload with URL rewriting for cloud-hosted galleries New files: - apps/eval/scripts/showcase/ (generate, executor, crosshair, manifest, uploader, types) - apps/eval/.env.example (R2 credentials template) Also adds Browser.getElementCenter() public method for coordinate resolution without triggering actions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>