Claude & Claude Code Quick-Start Guide
20 practical tips for using Claude and Claude Code as a non-developer — better prompts, cleaner workflows, smarter model choices, and safer file changes.
“Use Claude to think clearly. Use Claude Code to change files safely.”
Claude is where you plan, explain, and craft prompts. Claude Code is where those prompts become real changes inside your project files. Keep the two roles clear and both tools become dramatically more useful.
PART 1 — CLAUDE
Claude (claude.ai)
Get more out of every conversation — and stop burning tokens without knowing it.
Use Projects — stop repeating yourself
Go to Projects, create one, and upload your files there once. Then start all new chats inside that Project. Claude can reference your documents without you re-uploading the same file every time.
Choose the lightest model that fits the job
Opus — use when the task needs deeper reasoning or architecture thinking
Sonnet — the default fit for most writing, analysis, and coding work
Haiku — use for quick answers, lightweight checks, and simple tasks
Do not memorize one fixed default. Claude model defaults vary by account and can change over time, so pick the lightest model that still does the job well.
Edit your message — don't send a correction
When Claude gets it wrong, do not send a follow-up message. Click Edit on your original message and regenerate instead. The wrong answer gets replaced, not added to your history.
Saves up to 90% of tokens in long sessions.
Start a fresh chat every 15–20 messages
Every message re-reads your entire chat history. After 20 messages, ask Claude to summarize your progress, copy that summary, open a new chat, and paste it in. You will drop thousands of wasted tokens instantly.
Force Claude to ask questions first
Stop writing 1,000-word prompts trying to cover every case. Use this instead and let Claude figure out what it needs:
Copy-paste prompt
I want to [TASK]. Ask me questions one at a time before you start. Do not ask more than one question at a time.
Turn off tools you don't need
Every enabled tool uses tokens, even when sitting idle. Open the tools panel and turn off anything you are not using for that specific task — web search, file tools, code execution. Only keep what you need active.
Set up your user preferences once
Go to Settings — Profile and store your background, writing style, and technical level permanently. Claude will know who you are from message one ? no wasted tokens re-introducing yourself every session.
Upload Markdown files, not PDFs
PDFs are token-heavy. Instead: open a Google Doc, paste your PDF text in, download as a .md file, and upload that. Same information — a fraction of the token cost.
Plan first, then build in a fresh chat
Never ask Claude to build something complex in one shot. First ask it to plan and clarify. Once you have a solid plan, paste it into a new chat and say “Build this.” Claude stops guessing and starts executing.
Check your usage dashboard
Bookmark your usage page and check it regularly to see exactly where your tokens are going before waste gets out of hand.
URL to bookmark
claude.ai/settings/usage
PART 2 — CLAUDE CODE
Claude Code
Claude's power tool — it reads files, writes code, and automates your entire workflow from the terminal. Use it carefully.
Set up CLAUDE.md first — your project rulebook
CLAUDE.md loads before everything else. It tells Claude your project rules, key folders, preferred style, and what to never touch — without you repeating it every session.
Copy-paste prompt
Create a CLAUDE.md file for this project. Include: what this project is, key folders to know, things you should never touch, my preferred output style (short and direct), and any commands you'll need regularly.
Master the essential slash commands
Type / in Claude Code to see all commands. Here are the ones you will use every day:
Use /compact, not /clear
When your session gets long, /compact squishes the history into a dense summary — keeping all the important context but using a fraction of the tokens.
Use /clear only when switching to a completely unrelated task. Clearing throws away everything, including context you might still need.
Use Plan Mode before building anything big
Use /plan for the clearest path into plan mode, or cycle modes with Shift+Tab until plan mode is active. Claude will think it through and show you a plan before touching a file. Approve the plan first, then let it build.
Copy-paste prompt
Enter plan mode. Before writing any code, ask me everything you need to fully understand the task. Then show me a clear plan for my approval before doing anything.
Switch models mid-session with /model
Use /model when the task changes. Reach for a stronger model when you need deeper reasoning, and a lighter one when you are doing quick reads, small fixes, or cleanup work.
Use /resume to pick up where you left off
Accidentally closed Claude Code? Type /resume to see a list of all past sessions and jump right back in — no need to re-explain your entire project from scratch.
Build custom slash commands for repeated tasks
If you type the same instructions every week, turn them into a one-word command. Ask Claude to build one for you:
Copy-paste prompt
Create a custom slash command called /weekly-report that summarises all files worked on this week and lists key decisions made. Create the skill file and tell me exactly where to save it.
Stop Claude with Escape — not Ctrl+C
Ctrl+C exits Claude entirely. To stop Claude mid-task without closing it, press Escape. Press Escape twice to see a list of all previous messages you can jump back to.
Be specific — vague prompts burn tokens
Vague prompts force Claude to scan everything. Specific prompts go straight to the answer.
Vague (expensive — avoid this)
Fix the login bug
Specific (cheap — use this)
In auth/login.js, the password validation on line 42 fails for users with special characters. Fix only that function.
Set up GitHub review workflow when you are ready
Run /install-github-app to set up Claude's GitHub review workflow for a repository. Treat it as an extra review layer for pull requests, not as a replacement for reading diffs and checking important changes yourself.
The golden rule
Give Claude a clean desk, not a messy room.
The best way to reduce wasted tokens is not to babysit every prompt — it is to design your workflow so Claude only sees what it genuinely needs. Clean Projects, tight CLAUDE.md, one task per chat.
“Design the workflow. Then trust the tool.”
Free Prompt Pack
25 ready-to-use prompts.
Apply these tips immediately. The free prompt pack gives you copy-paste prompts for every stage of building an app with AI.
No payment. No signup. Just practical prompts.
Save this guide
Want this guide delivered?
Enter your email and we will send it when delivery is connected.
No spam. We will only send the guide.
Also available:
ChatGPT & Codex Quick-Start Guide