If you've installed Claude Code but froze up the moment you saw the blinking cursor, this is for you. There's no special syntax to memorize, no flags to look up — just plain English and a terminal. Here's exactly how the first conversation goes, what to expect while Claude is working, and how to keep the thread alive across multiple questions.
The Problem: CLI Tools Train You to Be Rigid
Every CLI tool you've used before had a grammar you had to internalize. git commit -m "...", npm install --save-dev, docker run -it --rm — each one is a mini language with its own flags and order rules. Break the syntax and you get an error, not help.
That muscle memory makes the first Claude Code session feel weird. Your brain is looking for the right flag. There is no flag. The tool literally just wants you to talk to it.
The mental shift is the whole thing. Once it clicks that Claude Code takes natural language and not memorized incantations, everything else follows.
Section 1: How to Actually Send Your First Message
Open a terminal in any project directory and run this:
claude 'What files are in this folder and what do they do?'
That's it. No subcommands, no --query, no JSON payload. Just claude followed by a quoted string containing whatever you want to know.
A few variants that all work equally well:
# Ask about your project structure
claude 'Show me the folder structure of this project'
# Ask for a code explanation
claude 'Explain the hardest part of this codebase to understand'
# Ask for a plain-language summary
claude 'Summarize the README in simple terms'
The gotcha I hit early on: quotes matter. If your question contains a single quote (like "what's"), use double quotes to wrap the whole thing, or escape it.
# This breaks
claude 'What's in the src folder?'
# This works
claude "What's in the src folder?"
# This also works
claude 'What'\''s in the src folder?'
On a modern M-series Mac, simple questions like these come back in 2–5 seconds. You won't be staring at a frozen screen.
Section 2: Reading the Response — Follow the Flow
Claude Code doesn't dump a wall of text the moment you press Enter. It works through the problem step by step and narrates what it's doing as it goes. You'll see something like:
> Analyzing project structure...
> Reading src/index.js...
> Reading package.json...
Here's what I found:
- src/index.js — main entry point, sets up Express server
- src/routes/ — three route files (auth, users, posts)
- package.json — Node 18, Express 4.18, no TypeScript
...
The processing lines at the top tell you what Claude is touching. If you're asking about a big codebase and it takes 10 seconds, those lines show you it's actually working through your files, not hung. The actual answer is always at the bottom — if the output is long, skip to the last third and you'll find the meat of it.
What worked for me: don't try to read every processing line. Glance at them to confirm Claude is looking at the right files, then scroll to the conclusion.
Section 3: Keeping the Conversation Going
This is where Claude Code pulls ahead of a one-shot grep or a Stack Overflow search. Context persists within a session. After you get a first answer, you can narrow in without re-explaining:
# First question — broad
claude 'Show me the overall folder structure'
# After reviewing the output, follow up — specific
claude 'Which of those files was modified most recently?'
# Narrow further
claude 'What changed in that file compared to the one in /backup?'
Each follow-up carries the full context of everything before it. You don't re-paste file paths or re-explain the project. Claude already has it.
The pattern that works best: start wide, then zoom. Ask for the overview first, then drill into the specific file, function, or error you actually care about. Think of it like a 20-questions game where you control the narrowing.
Section 4: Three Starter Questions That Require Zero Coding Knowledge
If you're staring at the cursor wondering what to type first, here are three questions I tested on a real project. All three work even if you're not a developer — they're the kind of thing you'd ask a colleague on your first day.
| Question | What it gives you | Response time (M4 Mac) |
|---|---|---|
claude 'Where might this project throw errors?' |
Risk map of the codebase | ~5 sec |
claude 'Summarize the README in plain English' |
Jargon-free project description | ~3 sec |
claude 'What's the hardest part of this code to understand?' |
Honest complexity map | ~4 sec |
None of these require you to know anything about the code ahead of time. That's the point — you're asking Claude to be the expert, not proving that you already are one.
The meta-lesson here: being explicit about what you don't know produces better answers. "I'm not familiar with this codebase — where would I start?" is a better first prompt than "explain the architecture," because Claude calibrates the depth of its answer to your stated starting point.
Section 5: Variations and Gotchas
Interactive mode vs single-shot mode
The examples above use single-shot mode: one command, one response. Claude Code also has an interactive REPL mode you can enter by just running claude with no arguments:
claude
# You're now in an interactive session
# Type your question and hit Enter
# Type /exit or Ctrl+C to quit
Interactive mode is better for back-and-forth debugging sessions. Single-shot mode (with the quoted string) is better when you want to pipe output or script Claude into a larger workflow.
Context resets between separate terminal commands
This is the gotcha I see trip people up most. When you run two separate claude '...' commands in single-shot mode, Claude does not carry context between them by default. The context threading I described in Section 3 works within a single interactive session, or when you explicitly pass --continue to pick up a previous session.
# These two share NO context (separate invocations)
claude 'show me the folder structure'
claude 'which file was modified most recently?' # ← Claude doesn't remember the first answer
# To continue: use interactive mode or --continue flag
claude # start interactive session, then type both questions
Differences across environments
| Environment | Notes |
|---|---|
| macOS (M-series) | Fastest local performance, ~2-5s for simple queries |
| macOS (Intel) | Slightly slower, same behavior |
| Linux | Identical behavior, may need export PATH adjustment after install |
| Docker container | Works, but mount your project directory so Claude can read files |
| Windows WSL2 | Fully supported, treat it like Linux |
Long responses in small terminals
If your terminal is narrow or short, Claude's output can be hard to read. Pipe it through less to scroll:
claude 'explain every file in this project' | less
Or redirect to a file for bigger responses:
claude 'audit this codebase for security issues' > audit.txt
Closing
The first Claude Code session isn't about knowing what to ask — it's about trusting that you can ask anything. Natural language input, streamed output, persistent context within a session: once you've seen it work once, the mental model snaps into place. 🖥️
Next up: feeding Claude a specific file to analyze, and how to ask it to make changes rather than just describe them.
🐦 Faster updates on X: @baegseungh7061
📚 More in this series: Code Intro
💌 Subscribe: Follow on X or grab the RSS
댓글
댓글 쓰기