snowflake-connector

Bring your own account. The plugin ships no account identifier, no client id and no role name, because every one of those is specific to one organisation and wrong for everyone else.

The setup skill is a wizard: how to discover your own account identifier, which authentication method to choose between key-pair, OAuth and SSO and what each costs you, why a read-only role is the right default for an agent, and how to verify the connection actually works before you depend on it.

The querying skill treats cost as the hazard, the way an observability plugin treats an unbounded log read. A careless query against a large warehouse is a real bill, and the damage is invisible until it arrives. So: bound every query by time and LIMIT, size the warehouse deliberately and let it auto-suspend, understand why SELECT * on a wide columnar table is expensive rather than merely untidy, read a query profile, and treat a query whose expected row count you cannot state as one you have not finished thinking about.