# 大模型防护 这次工具调用是否需要先拦下来? Input: 调用参数 + 约束规则 Judgment: 标记疑似越界或泄露线索 Action: 权限校验后放行、拦截或送审 Editorial illustration, not a deployed product. Source: https://docs.typesafe.ai/concepts/use-case-map