This is a property-based differential fuzzer that catches correctness bugs by running generated SQL statements against both Turso and SQLite, then comparing results. You can run it once with a specific seed for reproducibility or loop it continuously to hunt for edge cases. It generates hundreds or thousands of statements per run and flags mismatches in row sets, error handling, or schema state. The Docker runner is built for CI and will auto-file GitHub issues when it finds problems. Useful when you're working on Turso's SQL implementation and need confidence that behavior matches SQLite semantics. The seed-based reproduction workflow is solid: just grab the seed from the logs, rerun with verbose output, and you'll get the exact failing statement sequence dumped to test.sql.
npx skills add https://github.com/tursodatabase/turso --skill differential-fuzzer