Claude Code: what I wish I had known from the start

I've been exploring Claude Code for a few weeks and, as with everything I've tried, I've realized that the hardest part isn't using it, but organizing the work well.

Claude Code acts as a programming agent: it reads code, edits files, and runs commands. But since every session starts with a blank memory, if the project you're working on doesn't include the documentation it expects to find, in the files it expects to find and with the names it expects to find, it's like having a brilliant colleague you have to introduce the project to every morning.

I started out by intuition: a /docs folder here, an SQL schema file over there. It worked, but not quite. Until I understood how Claude Code expects to find a project's documentation and how to give it context between sessions.

I've put together what I've learned in these two guides, which I'm sharing in case they're useful to you:

Getting started with Claude Code — For starting from scratch: seeing that claude.ai is not the same as Claude Platform (which wasn't obvious to me at first), how to install it, authenticate, and run your first session.

Best practices with Claude Code — How to structure project folders, what to put in CLAUDE.md, how to create custom commands and subagents, and the mistakes worth avoiding.

I wrote them with Claude's help (yes, meta-recursive), so it's been like taking an accelerated course with it on how to work with it :-D.

I'm still learning, so to those of you who are ahead and see things that could be improved or have your own practices, suggestions are very welcome.