SQL console
A real SQL console against the same database the dashboard reads. The guard is server-side and refuses rather than sanitises: one statement, SELECT or WITH only, no comments, and a row cap applied when the query does not limit itself. Several worked examples are preloaded, including the ones behind the stories.
Query
SQLite dialect. Reads only — the pipeline writes, this does not.
What the guard does and does not protect
Stated plainly, because a console that implies safety it lacks is worse than none.
Protected: multi-statement scripts, comment-hidden statements, and every mutating or schema keyword (INSERT, UPDATE, DELETE, DROP, ALTER, CREATE, ATTACH, PRAGMA, and others outside SELECT/WITH).
Not protected: the guard is
still the only thing standing between a query and the database. The dashboard prefers a
read-only credential (TURSO_TOKEN_RO), which makes a guard bug harmless —
but if that is not set, the connection falls back to the write-scoped token and a bug
would have write consequences. The banner above the editor reports the scope of
whichever token is actually in use, from its JWT, rather than implying a safety the
deployment may not have.