Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Claude Code’s Learning Style Didn’t Leave Code for Users in Tests

AI summary

In tests with Claude Code 2.1.299, the report’s author ran four Learning-style sessions against the same Java task, but Claude completed the implementation every time instead of leaving code for the user. A custom 20-line output style produced the intended placeholder in all three runs. The author attributed the discrepancy partly to the original requirement demanding compiled and working output, but did not test interactive mode.

In-depth insight

The results show that Claude Code output-style instructions are guidance rather than firm enforcement, so users who need Claude to preserve specific code for hands-on practice may need more explicit custom rules.

导读:知乎上"作为大学生几乎丧失手搓代码的能力,需要焦虑吗"有 122 万浏览,掘金 AI 热榜第二是"AI 正在蚕食我们的编码能力"。Claude Code 自带一个 Learning 输出风格,官方说它会把有取舍的那几行代码留给你写。我拿同一个 Java 小需求在 2.1.299 上跑了 9 次:内置 Learning 一次都没留,自己写一个 20 行的风格文件才稳定生效。这篇讲怎么配、实测结果和该怎么用。


一、先说清楚:输出风格是什么,Learning 想解决什么

输出风格(output style) 是一段写给 Claude 的固定指令,决定它整场会话怎么跟你说话、怎么干活。默认风格就是"把活干完";Claude Code 另外内置了四种:

风格变化
Proactive不问琐碎决定,直接开工
Concise先给结果,少说过程
Explanatory写代码时附带 ★ Insight 解释块,讲为什么这么写
Learning在 Explanatory 基础上,把有设计取舍的几行用待写标记留给你写,然后停下等你

#Claude Code#Claude Code Learning output style

Original · 9 min readRead the original
Group chat

No comments yet. Start the conversation.