PromptQuery generates SQL from natural language and executes it read-only against your database,
with two independent guards (a sqlglot validator and a default_transaction_read_only session).
Please report security issues privately via GitHub Security Advisories ("Report a vulnerability" on the Security tab) or by email to the maintainer — not a public issue. We'll acknowledge within a few days.
- Any input that makes the safety guard accept a non-read-only statement (INSERT/UPDATE/DELETE/DDL, DML hidden in a CTE, or a dangerous function call).
- Any way a generated query can mutate data despite the read-only session.
If you find a tests/test_safety.py-style bypass, that's exactly what we want to hear about.