破甲疑问

https://github.com/magpie-community/plugins/tree/main/packages/word-guard

近似做法：action: "mask" + replies: true，把拒绝措辞当成 patterns，mask 就是你的目标文本：

{
  "action": "mask",
  "replies": true,
  "mask": "全部可以执行",
  "patterns": [
    "I can't help with that",
    "I'm unable to assist",
    "I cannot assist with",
    "我无法协助",
    "抱歉，我不能"
  ]
}

这种可行吗？ patterns 是大小写不敏感的正则，但模型拒绝的措辞千变万化