agent-openai-agents-sdk: databricks.yml ships with blank experiment_id and fails on databricks bundle deploy

[mdbecker]

I hit what looks like a template footgun in agent-openai-agents-sdk.

The current template is documented as including MLflow-based evaluation/tracing and a built-in chat UI, and Databricks’ bundle resource docs show the recommended pattern for app/experiment wiring is to define an experiments resource and reference it from the app. (Microsoft Learn)

However, the template databricks.yml includes an app resource binding like this:

resources:
  apps:
    ...
      resources:
        - name: "experiment"
          experiment:
            experiment_id: ""
            permission: "CAN_MANAGE"

In my case, databricks bundle validate passed, but databricks bundle deploy failed during app creation with:

Error: failed to create app
For input string: ""

That makes the template look deployable at first glance, but it fails at deploy time unless the user manually fixes the blank experiment_id.

Suggested fix
Update the template to use a bundle-managed experiment resource instead of a blank placeholder, e.g.:

resources:
  experiments:
    app_experiment:
      name: /Users/${workspace.current_user.userName}/agents-on-apps

  apps:
    <app_name>:
      ...
      resources:
        - name: "experiment"
          experiment:
            experiment_id: ${resources.experiments.app_experiment.id}
            permission: "CAN_MANAGE"

That seems more consistent with the current bundle resource docs and would remove a pretty sharp footgun for first-time users. (Microsoft Learn)

If helpful, I can open a PR with the change.