Reference other node's data in alert() details of the current node

InfluxDB 1 Kapacitor
1

In my rule what I try to do is to calculate the overall count of points (agents_count) in a window (where each point stands for an alive server along with its metadata), and then to re-use the same window data to filter some data points having only some certain tags values (idle_agents), and calculate the percentage of the remaining points against the overall count. Finally, this % is evaluated in a threshold of an alert() node. The simplified script is:

var data = stream
    |from()
        .measurement(measurement)

var agents = data
    |window()
        .period(window_period)
        .every(window_every)
        .align()

var agents_count = agents
    |count('status_code')

var idle_agents = agents
    |where(lambda: "status" == 'Idle')

var idle_agents_count = idle_agents
    |count('status_code')

idle_agents_count
    |join(agents_count)
        .as('idle', 'all')
    |eval(lambda: float("idle.count") * 100.0 / float("all.count"))
        .as('idle.%')
        .keep()
    |alert()
        .crit(lambda: "idle.%" <= float(crit_percent) OR "idle.count" <= crit_count)

What I want to include into my alert details is the data of the data stream before it’s got processed with count() - variable idle_agents in the case above. So that I not only get the bare % of the remaining points, but also the details of those remaining points (pretty printed somehow). This is how the data looks there:

{
  "series": [
    {
      "name": "measurement",
      "columns": [
        "time",
        "build_server",
        "host",
        "identity",
        "name",
        "organization",
        "project",
        "status",
        "status_code"
      ],
      "values": [
        [
          "2018-08-24T16:19:51Z",
          "build02.host.corp",
          "24511ae0f374",
          "my-identity1",
          "build02-agent-10",
          "quality_assurance_and_support",
          "build",
          "Idle",
          1
        ],
        [
          "2018-08-24T16:19:51Z",
          "build02.host.corp",
          "24511ae0f374",
          "my-identity2",
          "build02-agent-6",
          "quality_assurance_and_support",
          "build",
          "Building",
          2
        ],
...

The point is that once I process the points with count() what I have remained in the node is only the following:

{
  "series": [
    {
      "name": "measurement",
      "columns": [
        "time",
        "count"
      ],
      "values": [
        [
          "2018-08-24T16:07:30Z",
          15
        ]
      ]
    }
  ]
}

So I can’t access the original data when I then work it out with join() and alert().

I tried some stupid thing as adding that variable reference containing all the data - .details(idle_agents), but of course ran into error:

invalid TICKscript: line 86 char 10: error calling func "details" on obj *pipeline.AlertNodeData: reflect.Set: value of type *pipeline.WindowNode is not assignable to type string

There are some other rules with which I ran into the same necessity too. What would be the suggestion to achieve that? So that alerts would be much more meaningful.