What this covers. You have Claude Code installed (Mac or Windows). This lesson is the part everyone wonders about next: what do I actually say, and how do I work with it day to day. The short answer is treat it like a capable teammate who can read and write the files in your folder.
Claude works best pointed at a folder. It reads what is there and writes new files into it. An empty folder is fine for a first try.
In VS Code, use File → Open Folder and pick any folder (or make a new one). Then start a Claude Code session from the Command Palette (Cmd+Shift+P on Mac, Ctrl+Shift+P on Windows) and type Claude Code.
Prefer the terminal? cd into a folder and run:
Without a folder, Claude has no project to read from. Even an empty one gives it a place to create files. This is the single most common beginner snag.
You do not need special syntax. Describe what you want in plain language, the way you would brief a colleague. Try one of these:
Claude will think, then do the work: create files, run commands, or answer. Watch what it does. You are always in the loop.
That last prompt asks for two file types, and it is worth knowing the difference. Markdown (a .md file) is for quick, readable notes: you write plain text with a few simple symbols, like # for a heading or - for a list item, and the raw file is still easy to read on its own. HTML (an .html file) is what a web browser displays: it can have colors, layout, and clickable boxes, but the raw file is busier and meant to be viewed in a browser, not read as-is. Rule of thumb: ask for Markdown when you just want clean notes, and HTML when you want something that looks good in the browser. Claude is happy to make either, or both.
When Claude edits files, it shows you a diff: green for added lines, red for removed. Nothing is final until you accept it.
Changes appear inline. Review them and click to accept or reject. If something is off, just say so in the chat and Claude revises.
Claude shows the change and asks for confirmation before writing. Read it, then approve or decline.
Ask, watch, read the diff, accept or course-correct. That loop is the whole job. You do not need to know the code to steer it. Tell Claude what looks wrong in plain words and let it fix it.
After the first result, just keep talking. Claude remembers the conversation, so build on it:
If you want to undo a direction, say "revert that" or "go back to the previous version." If you are not sure what to do next, ask "what would you suggest improving here?"
Small things that make day-to-day work smoother.
Made by AI Service Engine. Next up: give Claude a memory with a CLAUDE.md file so it remembers your project every session.