I would base this on Superset's import/export bundle flow, not on editing exported files with sed.

A practical GitOps shape is:

superset export_dashboards -f dashboards.zip
superset export_datasources -f datasources.zip

Commit those bundles, or commit the unpacked YAML if your workflow validates the bundle structure. In Kubernetes, run the corresponding import from an init job or release job after migrations and before users start relying on the dashboards:

superset import_datasources -p /path/datasources.zip -u admin
superset import_dashboards -p /path/dashboards.zip -u admin

The key is to make environment-specific things stable outside the dashboard JSON/YAML: database connections, secrets, SQLAlchemy URIs, and ownership. Import datasources/databases first, then dashboards. Avoid rewriting UUIDs or IDs unless you have a deterministic migration step and tests, because those identifiers are how the exported objects refer to one another.

For Helm, keep secrets in Kubernetes secrets or extraSecretEnv, and put import commands in a controlled job/hook rather than a long-running web pod startup script.