What problem does it solve? Managing schema changes for a PostgreSQL 15 + pgvector database without a disciplined migration workflow leads to drift, irreversible changes, and inconsistent environments. This Skill enforces a strict dbmate-based process so every schema change for the Portal Centralizado de Datos Financieros is versioned, reversible, and reproducible. ## Core Features & Use Cases - Reversible migrations: Every change uses migrate:up and migrate:down sections, with db/schema.sql regenerated and committed after each make db-up. - Canonical schema templates: Ready-made SQL for catalog_source, catalog_field, catalog_tribal_note, app_user (with Argon2-hashed seed users), export_job, and the pgvector extension with an HNSW index for embeddings. - Non-negotiable rules: Never edit applied migrations, never use SQLModel.metadata.create_all() in production, always use TIMESTAMPTZ DEFAULT now(), explicit ON DELETE clauses, and pre-hashed Argon2 seed passwords. - Use Case: When adding a new embedding column to catalog_field for the RAG phase, run make db-new SLUG=enable_pgvector, paste the provided up/down SQL, and apply it with make db-up. ## Quick Start Ask the assistant to create a new reversible dbmate migration for the Portal database, for example to add the catalog tables or enable pgvector, and have it generate the SQL file and apply it with the make commands.
