warehouse-sql

What problem does it solve?

It turns natural-language questions into read-only (by default) SQL that you can safely run on business data warehouses while minimizing schema misunderstandings and preventing accidental writes.

Core Features & Use Cases

• Cached schema introspection across warehouses: Refreshes and reuses a schema cache to draft accurate SQL for SQLite evals and Postgres/ClickHouse/BigQuery/Snowflake backends.
• Deterministic review + model guardrail before execution: Performs deterministic safety checks (single statement, narrow columns, LIMIT, scope filters) and can require an OpenAI-compatible model review endpoint to approve execution.
• Explicit write grants with confirmation: Detects mutation attempts and only allows writes when the user explicitly requests write access, the command includes allow-write, and the required environment write-grant matches; mutation execution still requires confirmation.

Quick Start

Ask for SQL and review first by running: python3 skills/warehouse-sql/scripts/warehouse_sql.py --format json review --model-review --question "Show the first 10 customers" "SELECT c_name FROM customer LIMIT 10"