talforsherClaude Code
curl -fsSL https://raw.githubusercontent.com/vibe-md/vibe.md/7cb0620fbd80f9829155d8866328b58cd18d5aa7/scripts/install.sh \
| bash -s -- --skill safe-migrations \
--agent claude-code --ref 7cb0620fbd80f9829155d8866328b58cd18d5aa7Safe Migrations
AvailablePlans and reviews database schema changes so they are reversible, verified, and never silently destructive. Use when the user asks to add, change, or remove tables or columns, when a migration fails partway, or before running pending migrations in a production or shared environment.
Turns "change the database" requests into reviewed, reversible migrations. Before anything runs, your agent classifies the risk, writes a rollback, shows you the exact SQL and plans how to check that it worked.
DatabaseSecurity
- Claude Code
- Codex CLI
- GitHub Copilot
- Cursor
- Gemini CLI
Maintained by
The problem
Schema changes are among the easiest ways to lose data. An AI agent asked to "just add a column" or "run the migrations" can drop a column, fail halfway and leave the database in a mixed state, or hit production without a backup. None of that is easy to undo.
What it does
Classifies every change
Labels each change as additive, transform or destructive, and plans accordingly. Destructive steps always need your explicit approval.
One change, one reversible file
Keeps one logical change per migration, each with a rollback.
Shows the SQL before it runs
Presents the exact SQL, the risk, what happens if it fails partway and the rollback plan, and waits for approval on risky changes.
Verifies the result
Checks the schema, row counts and sample rows after applying, with a verification query per change.
Careful with production
Never touches production without approval, and takes a backup or snapshot before risky changes.
Installation
One command
Install directly into your project.
Open Terminal in your project folder, then run this command. No ZIP, moving files, or prompt needed.
- 1Open your project folder in Terminal
- 2Run the command above
- 3Reload your agent and ask to use Safe Migrations
Works on macOS, Linux, and Windows through WSL or Git Bash. The installer checks destinations and stops if existing files conflict.
Usage examples
Type these to your agent in plain language. Here's what to expect.
you › Add a status field to orders and backfill it to 'pending'.
Expected workflow
- 01Inspects the orders table.
- 02Proposes adding a nullable column, a batched backfill, then NOT NULL once the counts match.
- 03Applies it in development and verifies with row counts.
- 04Presents the rollback plan before staging.
you › Just run the pending migrations, production needs them now.
Expected workflow
- 01Lists what is pending and classifies each change.
- 02Holds destructive steps for your explicit approval.
- 03Snapshots the database before risky changes, applies the rest in order and verifies.
How it works
The skill works in a fixed order: read the current state, classify the change, write migrations that can be re-run and rolled back, show them to you, then apply and verify in development before anything shared or live.
If a migration fails halfway, your agent works out which statements actually ran and restores a consistent state before trying again. It does not just run it a second time.
Safety and limitations
Required permissions
- Read your migration files and database schema.
- Run migrations against the environment you approve.
Potentially destructive operations
Read what your agent is about to run before approving it. These can lose work:
- Dropping tables or columns, truncation and deletions need your explicit approval, a backup and a stated rollback plan every time.
- Production is never touched without your approval.
Limitations
- It makes a schema change safe. It does not design your schema.
- It is not tied to one framework, and you may be asked which migration system your project uses.