“The OpenClaw platform has a built-in security feature that differentiates between "authenticated command channels" and "information channels" to prevent prompt injection.”