Ask Claude to Write Data Migration Scripts for Backfills and Cleanups

Schema migrations change your database structure. Data migrations change what's in it — and they're harder to get right because bad data transforms are hard to undo. Claude reads your schema and writes scripts that handle the edge cases.

Write a data migration that backfills the new 'full_name' column 
from the existing 'first_name' and 'last_name' columns. Handle nulls, 
empty strings, and names with multiple spaces. Process in batches of 1000.

Claude reads your model and writes a script that handles the edge cases you'd forget:

// Claude generates framework-appropriate batch processing
User::whereNull('full_name')
    ->chunkById(1000, function ($users) {
        foreach ($users as $user) {
            $parts = array_filter([
                $user->first_name,
                $user->last_name,
            ]);
            $user->update([
                'full_name' => implode(' ', $parts) ?: null,
            ]);
        }
    });

This works for all kinds of data transformations:

# Clean up inconsistent data
Find all phone numbers in the users table that aren't in E.164 format 
and normalise them. Log any that can't be parsed.

# Merge duplicates
Find duplicate users by email (case-insensitive). Keep the one with the 
most recent login, merge the order history from the others, then soft-delete 
the duplicates.

# Split a column
The 'address' field contains freetext addresses. Parse them into 
street, city, state, and postcode columns. Flag any that can't be 
parsed reliably for manual review.

# Encrypt sensitive data
Write a migration that encrypts all plain-text SSN values in the 
customers table using the app's encryption key. Make it idempotent 
so it can be re-run safely.

Always ask Claude to make the script safe:

Make this data migration:
- Idempotent (safe to run twice)
- Batched (don't load the whole table into memory)
- Logged (report progress every 1000 records)
- Reversible if possible (store the old values somewhere)

Schema migrations are well-tooled. Data migrations are where things go wrong — let Claude handle the edge cases, batching, and safety checks you'd skip if you wrote it yourself.

via Claude Code