Currently the workflow described in docs is something like this:

docker.image('alpine').inside {
  sh 'X'
  sh 'Y'
}

It does not address the situations when you need to:

1. run X  in container
2. run Y in host agent
3. run Z in the same container, preserving changes from 1

The only solution to that, AFAIK, is something like this:

docker.image('alpine').withRun { container ->
  sh "docker exec ${container.id} X"
  sh "Y"
  sh "docker exec ${container.id} Z"
}

As mentioned in JENKINS-39746 , this is kinda ugly, even if it could be rewritten as container.exec("X").

It would be nice to have the ability to use .inside{...} not only on Image objects (source), but also on Container objects, to allow workflow like this:

docker.image('alpine').withRun { container ->
  container.inside {
    sh "X"
  }
  sh "Y"
  container.inside {
    sh "Z"
  }
}