Definition
A safety check that catches malformed AI commands before they ever reach the real system.
In Life-Harness, the harness layer that validates and canonicalizes proposed tool calls against schemas before execution, blocking syntactically broken outputs and returning structured error feedback to the model.