Sometimes a feature needs to store something new — a field, a list, a new kind of record. You don't do any of this by hand, but here's how that change reaches the live app safely.
Your practice workspace has its own copy of the data; the live app has the real one. The same app runs against both — it just points at a different one. The shape of the data travels between them; the actual records don't.
Practice data. Add fields, try things, break it — nothing real is here.
Real records. Gets the same new shape on release — the real data is never touched by hand.
You just ask Claude for what you need. Behind the scenes, it's careful and repeatable:
It adds the new field and records the exact change as a clear instruction the databases can follow.
So you can build and see the feature working, with fake data — no risk.
Checked automatically and approved by a teammate, alongside the feature that uses it.
The same recorded instruction updates the real database's shape. Existing records stay put; nothing is edited by hand.
Adding is routine. Changing or removing needs more care — that's where real data can be affected.
Just shows up on release; existing records are untouched. The everyday case, no special steps.
Renaming or deleting existing data is done in small steps, with a backup and approval first — because real records are involved. Claude follows that rule for you.
Some features need a bit of starting information — a list of categories, default settings. That ships with the change, so it's there in both your practice copy and the live app. Only real user data stays separate.