hs:databases

Follow the OLTP workflow: identify entities and query patterns, explore any existing schema, then propose DDL using snake_case plural table names, FK indexes, audit columns, and composite indexes matching your main queries. The skill requires user approval before executing any DDL.

Build pipelines by ordering stages efficiently: $match and $project early to reduce documents, then $group, $sort, and $limit. The references cover arithmetic, string, date, array, and conditional operators plus patterns like faceted search and window functions.

PostgreSQL fits relational data with complex joins, transactions, and strict integrity, while MongoDB fits document-oriented data with flexible schemas and embedded structures. The skill covers both so you can choose based on your data model and query patterns.

Run the db_performance_check.py script with your connection URI and a millisecond threshold. It reads pg_stat_statements for slow queries, detects tables with high sequential scans, lists unused indexes, and outputs index recommendations.

No. For PostgreSQL the password is stripped from the URI and passed via the PGPASSWORD environment variable, and for MongoDB the URI is written to a temporary 0600-permission config file, keeping credentials out of the process argument list.

D1 is SQLite-based with a 10 GB size limit and single-threaded queries, so every query needs an index and batches must stay under 100k rows. It also lacks native comments, so metadata tables are required to document tables and columns.