Some tests that involve UI fail when taskcluster-worker runs as daemon. We need to find out how to run it in a keep alive basis that allow UI interactions.

Let's make a command that runs it under panicwrap. Then you have the option of restarting on panic, and send the panic messages (+stack trace) to sentry. This can be done with: https://github.com/mitchellh/panicwrap Which will fork the process...

How does this help the issue of needing to start the worker in a way that allows UI interactions?

Ok, we need to configure taskcluster-worker to start as a local agent running under an autologin account.