克劳德代码的/洞察
Claude Code's /Insights

原始链接: https://www.natemeyvis.com/claude-codes-insights/

Claude 代码现在提供了一个 `/insights` 命令,提供使用报告——而且体验出人意料地像人类。该报告会分析你的互动,提供关于你工作流程的反馈,就像一位经理一样。 作者发现 Claude 将他们归类为“浏览器自动化专家”,因为他们进行了资源密集型的脚本编写,并因放弃对话而受到温和的批评。虽然一些反馈是有效的,但其他建议与其探索性工作方式不太一致,强调了向 AI *解释* 你的方法的重要性。 尽管如此,该报告提供了有价值的、具体的改进 AI 互动建议——例如创建技能和代理——并鼓励重复/封装成功的工作流程。报告并非静态的,似乎更侧重于最近的活动。总的来说,作者认为 `/insights` 是对未来工作方式的引人注目的展望,并强烈建议使用它来完善你的 AI 协作技能。

黑客新闻 新 | 过去 | 评论 | 提问 | 展示 | 招聘 | 提交 登录 Claude 代码的 /Insights (natemeyvis.com) 7 分,由 ingve 发表于 42 分钟前 | 隐藏 | 过去 | 收藏 | 讨论 指南 | 常见问题 | 列表 | API | 安全 | 法律 | 申请 YC | 联系 搜索:
相关文章

原文
Claude Code's /insights | Nate Meyvis

Claude Code now has an /insights command that generates a report about your Claude usage. I ran it for the first time today. Here are some notes:

  1. At times, it is almost eerily like getting feedback from a very well-informed, highly trained human manager.
  2. That's not all good. Claude views me as a browser automation specialist: I've been scripting some user flows and asking Claude to run through them in Chrome. This generates some long-running, resource-intensive sessions that skew my metrics (or so I conjecture).
  3. Claude gave me some gentle negative feedback for abandoning too many conversations before I got concrete results. In some cases I certainly did set up projects wrong and fail to give Claude the tooling it needed. In other cases I was just exploring a bit, and abandoning a lot of these threads seems pretty optimal to me. Especially since I went through a stretch of using Codex to "hit singles", I'm pretty comfortable with my approach here. Again, though, this felt exactly like something a human manager would say, however limited it might be as an evaluation of me.
  4. I'm chatting with Claude about the report, trying both to learn more about why it made those recommendations and describing why I think my work style has benefits Claude is under-valuing. Using AI feedback to get better is going to be an essential skill. (Justifying your work's value to AI, too.) Start practicing now.
  5. The report gives concrete examples of things to try (making skills, agents, and hooks, with copy-and-pastable suggestions); the suggestions are very good. Most of us could profitably do more to repeat and encapsulate our AI interactions, and these reports should help with that.
  6. I ran /insights twice and got broadly similar reports, but with different emphasis. I had done some work with Claude between runs, so I don't know whether this reflects random elements, the effect of the differential sessions, or both. My best conjecture is that /insights weights recent work heavily.

Now that I'm getting at least somewhat used to the amazing things AI can do to build, /insights was one of the most striking, "feels like the future" experiences I've had recently. Recommended.


#Claude #future of work #generative AI

联系我们 contact @ memedata.com