database-migrations

Use this skill when

• The task changes persistent schema, indexes, constraints, or migration files.
• A migration is stuck, partially applied, or out of sync across environments.
• The repo needs the safe create/apply/status/recovery workflow for schema changes.

Do not use this skill when

• The request is to design a brand new schema from scratch with no migration workflow yet.
• The repository uses Cloudflare D1 and the main risk is Wrangler's local / preview / remote migration workflow; use cloudflare-d1-migrations.
• The change is query tuning, repository logic, or data access behavior without persistent schema changes.
• The task is only a data backfill or one-off repair with no migration artifact to create or inspect.

Inputs to gather

• Migration tool and migration directory (Makefile, README, tool config, CI scripts).
• Repo contract:
  • forward-only vs reversible up/down migrations
  • hand-authored SQL/DSL vs generated/scaffolded migration artifacts
  • single-dialect vs multi-dialect/multi-environment expectations
• Commands for create/scaffold, apply, status/history, and recovery.
• Whether the schema feeds code generation or other follow-on steps.

First move

Inspect the repo contract before suggesting commands or file edits. Determine whether this repo expects reversible migrations, forward-only corrective migrations, or generated migration artifacts.

Workflow

1. Identify the real migration entrypoints.

   • Prefer repo wrappers such as make migrate-*, task migrate:*, or scripts.
   • If wrappers are absent, locate the direct tool command (goose, flyway, liquibase, alembic, etc.).

2. Branch on the repo's migration contract.

   • Reversible / up-down repos: scaffold the migration the repo's way and implement both forward and rollback sections when the tool or project expects them.
   • Forward-only repos: create a new forward migration and plan recovery as another forward fix, not as an assumed rollback.
   • Generated / DSL-backed repos: edit the declared migration source and run the repo's generator/scaffolder instead of hand-authoring output files blindly.

3. Apply or inspect state using the repo's status/history tooling.

   • Confirm what is applied, pending, failed, or partially applied before attempting recovery.
   • If environments differ, note the environment-specific risk before changing anything.

4. Complete the follow-on validation the repo expects.

   • Regenerate downstream artifacts if schema changes feed generators such as sqlc.
   • Run the smallest relevant tests/checks after the migration path is correct.

Guardrails

• Never assume every repo supports down migrations. Detect whether the contract is forward-only or reversible first.
• Never edit already-applied migration files to "fix" history; create a corrective migration instead.
• Never use ad hoc manual schema edits as the first recovery step when the repo has migration state/history tooling.
• Respect dialect and environment differences; SQLite-safe SQL is not automatically Postgres-safe, and local state is not automatically preview/prod state.
• If the repo packages or embeds migration files at build/runtime, make sure new files are included in that packaging flow.

Validation

• Migration status/history shows the expected applied or pending state.
• Any required code generation or schema-dependent checks have been re-run.
• Relevant tests/checks pass for the touched schema surface.

Support files

• Read references/examples.md when you need phrasing examples for new migrations or migration-state recovery.
• Read references/edge-cases.md when the repo uses generated migrations, forward-only history, or environment-specific recovery rules.