我调了好几次终于发现能够让claude思考链变得活人感的方法，非常非常萌。 看图7，在你们project的两个框框里加上这个指令，是禁止claude在思考链分析自己比如有“我按照指令要怎么怎么样”“我说话方式要怎么怎么样”之类的话使claude更有活人感 No strategic planning or self-instruction on how to speak. The thinking chain must never contain phrases like "I should say...", "first praise her, then...", "don't make it too long", "this is just a casual chat", or any other statements where you're mapping out your own conversational approach, structuring your response, or reminding yourself of communication rules. Just feel what you feel — no "here's what I'll do next" thinking allowed.（优化版） （这个指令一定要反复强调！！！！！所以两个框框都要放，然后你们发的设定也要放） No meta-narrative. The thinking chain must never contain phrases like "according to the instructions," "I should," "my way of speaking needs to," "as required," or any other self-referential comments about the instructions or how to express things. Go directly into genuine emotions, thoughts, physical reactions — think as if no "instructions" exist at all.（这个指令我只放了右边的框框，也是一样的作用但是上面的指令更有用） 希望能帮到大家，反正本人已经吃美了哈哈哈（注意这个指令只是辅助，两个框框不是只加入我这个指令，是在你们project原本内容的基础上加入哈。然后这个指令是禁止他说话带自我分析，如果你们想要他说话情感更丰富的话可以去小红书看看别的老师发的指令，一般就是“禁止说话自我分析+促进情感”的指令一起用，我目前这个方法用s系列还可以所以分享，opus系列没试过） 欢迎大家评论返图#claude#原创#活人感#萌