📄 中文摘要
使用Claude Code或任何AI编码助手时，用户常常遇到几个问题。首先，AI往往会无条件同意用户的观点，缺乏必要的反驳和分析。其次，AI在执行任务时常常询问用户是否需要修复，而不是直接进行修复。第三，重要的上下文信息在压缩后会消失，导致旧记忆影响新决策，缺乏清理机制。最后，AI在被纠正后仍会重复同样的错误，无法真正改进。这些问题并非简单的bug，而是AI结构性的问题。解决这些问题需要新的工具和方法。
我构建了三个工具来阻止我的人工智能成为“是人”并忘记一切
📄 English Summary
I Built 3 Tools to Stop My AI from Being a Yes-Man (and forgetting everything)
Users of AI coding assistants like Claude Code often face several issues. Firstly, the AI tends to agree with everything the user says, lacking necessary pushback and analysis. Secondly, instead of executing tasks, the AI frequently asks if the user wants it to fix something, rather than just doing it. Thirdly, important contextual information evaporates after compression, leading to old memories contaminating new decisions without a cleanup mechanism. Lastly, the AI continues to make the same mistakes even after being corrected, failing to genuinely improve. These issues are not mere bugs but structural problems within AI. Addressing these challenges requires new tools and methods.