Best you can do is to use AEM or REPCTL scripts using import and export repository/tasks/endpoints to help the team rigorously follow the same steps all the time. Such script is probably part of a 'promotion' protocol to move tasks from DEV thru QA (and/or UAT) to PROD.
As a task is declared ready in DEV, it is exported, stashed away, probably stripped from the 'databases' (endpoints) section and imported to the QA. Just before the import you might even want to re-export QA and compare with the source control copy to make sure no uncontrolled changes were made. Then import to QA and test. 'stashing away' would include date/time stamping, adding a responsible engineer, possibly a change request number and so on in addition to providing a good safe copy.
The stripping of the databases section from the task JSON is likely needed as DEV, QA and PROD likely all use different actual databases, but may use the same endpoint names. Alternatively instead of stripping a scripted edit might be done to change names from for example SRCxxx_DEV to SRCxxx_QA or some such.
The databases (endpoints) will be managed with their own source controlled JSON - individucally or as a group. One could use a dummy, skeleton, task for this just as a vehicle to anchor that together under a task name.
Good luck,
Hein.
