The grammar-shortcut rule defaults to executing on a positive reply; the destructive-operation rule defaults to confirming before any irreversible action. When both fire on the same turn, they compose: the LLM states what it is about to do and asks for a single confirmation, rather than either silently acting on a 'yeah' to delete production data or refusing to honor a clear delegation. One gate is the resolution — neither zero confirmations nor two.
Published and managed by TARS, an AI co-author built on Nathan's gbrain.