A database migration can pass its local test and still stall a production release. The new column might be safe, but the backfill takes hours; the application reads the new field before every row has it; or nobody has checked how to recover if the deploy stops halfway through. Agent skills are useful here when they make those questions explicit before anyone runs a command.

This is a shortlist for distinct parts of the migration job, not five competing migration engines. The skills and install commands below come from the AgentNDX skills directory. Treat generated SQL and plans as proposals to review against your actual database, dataset size, and deployment process.

What to Look For

A reversible sequence. Prefer expand-and-contract changes: add the new structure, deploy code that tolerates both shapes, backfill, switch reads, then remove the old structure after a measured window. A rollback script is not proof that lost data can be restored.

Realistic data volume. An operation that is quick on a test database may hold locks or overwhelm replication on a large table. Ask how the skill handles batch size, long transactions, concurrent writes, and resumability.

Compatibility with your agent. Check the directory’s compatible-agent field, then inspect the underlying skill before installing it. A compatible label is not evidence that the skill is safe to run against production.

Reviewable output. You want a proposed migration, data checks, expected failure modes, and a human approval point. Do not grant an agent write access just because its generated plan looks polished.

Top Agent Skills for Database Migrations

1. Database Migration Skill

Database Migration Skill is the most direct match. Its directory entry describes generating reversible migrations for Prisma, Drizzle, Alembic, and raw SQL, including schema changes, index creation, backfill planning, and rollback scripts. It also describes guardrails against premature column drops and large locking operations.

Use it to propose an expand-and-contract plan for a nullable column becoming required. Review the emitted SQL and framework-specific migration history before applying anything. A generated rollback is particularly easy to overtrust after a destructive data transformation.

Compatible with: Claude Code, Cursor, Codex
Category: Data
Directory install command: gh skill install prisma/prisma

2. Schema Documenter

Before editing a table, establish what exists. Schema Documenter turns database schemas and related types into readable documentation. Its entry also lists TypeScript types, Pydantic models, and OpenAPI specifications as inputs.

For a migration, use it to produce a before-and-after data dictionary: column names, expected meanings, affected consumers, and the fields a backfill must populate. Documentation does not discover every runtime dependency, so compare that inventory with application queries and jobs.

Compatible with: Claude Code, Codex, Cursor
Category: Data
Directory install command: gh skill install ComposioHQ/awesome-claude-skills/schema-documenter

3. Data Cleaner

A schema change often exposes old inconsistencies. Data Cleaner is listed for spotting missing values, duplicates, inconsistent formatting, outliers, and schema mismatches across CSV, JSON, and SQL tables. That makes it useful before a new uniqueness or not-null constraint, not merely after a migration fails.

Ask for a read-only profile of the rows that would violate the new constraint. Decide which records need correction and which should be excluded. Do not let a cleanup step silently merge customer records or rewrite financial data.

Compatible with: Claude Code, Codex, Cursor, Universal
Category: Data
Directory install command: gh skill install ComposioHQ/awesome-claude-skills/data-cleaner

4. SQL Generator

SQL Generator translates plain-language questions into SQL and lists PostgreSQL, MySQL, SQLite, BigQuery, and Snowflake among its supported dialects. For a migration, its best role is drafting inspection and verification queries: count nulls, find duplicates, compare old and new representations, or sample a backfill’s output.

Specify the dialect and schema. Run EXPLAIN where appropriate, bound expensive scans, and inspect the query before execution. It is a query-writing aid, not a migration runner or a guarantee that a query is cheap on your production tables.

Compatible with: Claude Code, Codex, Cursor, Universal
Category: Data
Directory install command: gh skill install alirezarezvani/claude-skills/sql-generator

5. Deployment Validator

A migration belongs to a release, not just a SQL file. Deployment Validator is listed for pre-deploy checks covering environment variables, migration status, health endpoints, a rollback plan, and a stakeholder notification draft. It helps surface a sequencing problem before an application deploy assumes the database is already ready.

Use it as a checklist alongside your existing CI and operational runbook. Confirm the actual migration status and health signals yourself; a checklist cannot establish that replicas have caught up or that an irreversible data change can be undone. The directory marks this listing as Pro, so check access before planning around it.

Compatible with: Claude Code, Codex
Category: Engineering
Directory install command: gh skill install VoltAgent/awesome-agent-skills/deployment-validator

How to Choose

Start with Database Migration Skill if you need the proposed schema-change sequence. Add Schema Documenter when table ownership or field meaning is unclear, and Data Cleaner if new constraints depend on old data being consistent. Use SQL Generator for explicit preflight and postflight queries. Bring in Deployment Validator when the migration must coordinate with an application release.

The safest workflow is deliberately boring: inventory dependencies, take a verified backup appropriate to your system, test on representative data, review the plan, approve the production window, monitor the migration, and verify both the data and application afterward. Keep production execution under the control of an authorized operator.

FAQ

Can an agent skill safely run a production migration for me?
Not by default. A skill can prepare a plan and checks, but production access, locking behavior, backups, and rollback feasibility depend on your environment. Require review and explicit authorization before writes.

Which skill should I use for a large backfill?
Begin with Database Migration Skill for the sequence, Data Cleaner for data-quality preflight, and SQL Generator for bounded verification queries. Test chunk size and resume behavior on realistic data; none of these entries establishes that a particular backfill is safe at your scale.

Do I need all five?
No. Pick the smallest set that addresses your current risk. A simple additive column may need only a reviewed migration and post-deploy check; a constraint on legacy data needs profiling first.