The big thing to test for is that errors from within the toolbar get sent to our own sentry project, and errors from the surrounding page get sent to the customers sentry project.

No errors should escape our toolbar stacktraces.

Multiple SDK instances/versions could probably exist on the page; like if the customer has not updated to the latest SDK the toolbar could be using a newer version.

The toolbar should not instrument replay, whether we can or should instrument spans/traces is something to sort out.

Edit:
seems like we can skip calling Sentry.init() and instead call new Sentry.Client and then try/catch manually.

There's the sentry-toolbar project already setup to use: https://sentry.sentry.io/projects/sentry-toolbar