CLAUDE.md explained — vibe-coding primer

What is cloud.md and how to use it? This is day five of building with cloud code. So if you don't know what cloud.md is, it's basically a memory file for cloud code. Similar to how you can set up your iPhone to bring an alarm every morning or set a specific wallpaper, you can set rules and preferences in this file. Cloud reads it every time you start a session so it already knows how to work with you before you even say anything. But what should you actually put in it? Well, there's no single right answer. It really depends on your project and how you work. Boris Cherny, the creator of cloud code at Anthropic, shares internal bad practices and workflows on X that he and his team use with cloud code. And someone turned those threads into a structured markdown file that you can use in your own project. So let's walk through it. First, plan before executing. This tells cloud to think through a plan before diving into any non-trivial task. Instead of immediately writing code, cloud stops, outlines its approach and only starts building once the plan looks right. And if something goes wrong, it replans instead of pushing forward with a bad approach. Second is sub-agent strategy. This is like hiring kitchen staff instead of trying to cook the entire meal yourself. It tells cloud to use sub-agent to keep the main context window clean. So for research or exploration tasks, cloud offloads that to a sub-agent instead of clearing up your main conversation. Third, self-improvement loop.…