First Hour with Claude Code

Claude Code is not Claude.ai with a different skin. The interface is a terminal, but the substance is different — it reads and writes the files on your actual machine, sees the full project around it, and acts (with your permission) instead of just answering. The single biggest reason new users struggle is they treat it like a chat. This skill exists to fix that in the first session, not the tenth.

The skill adapts to who's asking. A terminal-fluent developer gets capabilities, CLAUDE.md setup, and the mental model — no patronising explanation of cd. A first-time terminal user gets every step, including how to open Terminal and what "change directory" means. A non-developer gets the part most onboarding misses: Claude Code is genuinely useful even if you never write code, because if your work involves files, it can manage, transform, and analyse them.

The walkthrough is concrete and ordered. Open terminal, navigate to a project, type claude, run a first command that's safe and immediately useful — read a file, find a function, summarise a folder. Then a sequenced "next 50 minutes" of building muscle memory across reading, changing, and git workflows. Each step has the actual command, not a hand-wave.

It also covers the things that actually go wrong. command not found (Node 18+, PATH, restart your terminal). Permission prompts feeling excessive (they're a feature; relax them once you've built trust). Wrong file edited (be specific about paths). The "it forgot what we said" problem (that's context — start fresh sessions). What it deliberately doesn't do is teach you to code, replace your IDE, or pretend Claude Code is good at everything Claude.ai is good at. The mental model it pushes: a skilled colleague at the next terminal, not an autocomplete and not a chatbot.