Database Conventions

Engine-specific routing

PostgreSQL - supabase-postgres-best-practices for query patterns, indexing, partitioning. Ignore its RLS / Supabase-tenancy rules unless the project actually uses DB-enforced tenancy.
SQL Server / T-SQL - query-optimization (rewrites, SARGability, plan reading), index-strategies (clustered / nonclustered / filtered / columnstore), tsql-functions (function catalog).
SQLite - no dedicated skill; treat as Postgres-lite: WAL mode by default, single-writer per file, pragmas at connection open, INTEGER PRIMARY KEY for autoincrement, no concurrent writers across processes.
MongoDB / document stores - no dedicated skill; apply document-modeling care: embed vs reference per access pattern, index every queried field path, bound array growth, no unbounded $lookup.

Never write raw SQL strings in application code. Use parameterized queries or ORM. Concatenation is forbidden even for "safe" inputs.
Never log query text containing parameter values that may carry PII / secrets.
Default reads to read-only intent (transaction isolation READ COMMITTED minimum; SNAPSHOT / REPEATABLE READ only when needed and justified).
Bound every result set: LIMIT / TOP on any query that could grow unboundedly. No SELECT *.