Non-technical macOS users can get in on the fun with Claude Code, even if they’re using Terminal for the first time. This post shows you how.
If you're new to Claude Code and not a developer by trade, the Terminal app on the macOS can be intimidating. Claude Code in Terminal is powerful, but the tool assumes a technical baseline knowledge most people don't have. This post is for you if you're a non-technical user who is using the Terminal for the first time.
I've been using Claude Code (CC) regularly for a few months and recently helped a couple of friends get started over video calls, walking them through installation with Anthropic's Claude Code Quickstart Guide, and sharing tips to help them hit the ground running. This post collects those tips.
Note: Throughout this post, "Claude" and "CC" both refer to Claude Code specifically, not Cowork or Claude on the web. The screenshots here are from Claude Code v2.1.92.
A. Getting Started
A.1. Check your Privacy Settings
Why: If you don't want Anthropic to use your data to improve their models, you'll have to explicitly set that.
Scroll down to “Help improve Claude” and turn it off.
Tip: Privacy controls can change over time. There's no substitute for checking the company's current Privacy Policy and making an informed decision about what data you're willing to use with Claude.
A.2. Create a project folder for each project
How: Use Finder to create your project folder like you would any other folder. Since it's newly created, it will start out empty.
Why: To control what Claude Code can access.
When you launch Claude Code from Terminal while inside a specific folder, CC gains access to the contents of that folder, including any subfolders. If you launch CC from the root of your drive, it gets access to your entire drive. The simplest way to limit that access is to create a dedicated folder for each project and launch CC from within it.
Tips:
Back up automatically. If you use a service like Dropbox, put your project folders there so your files are backed up and versioned without extra effort.
One project, one folder. Give each project its own folder to keep your setup uncluttered.
Start empty. If you're just beginning, start with an empty folder and only add files as you need them.
Screenshot showing a redacted view of my project folders
A.3. Launch Claude Code from a project folder
How: In Finder:
Right-click on your project folder.
In the pop-up menu that appears, click "Open in Terminal" or "New Terminal at Folder" (the exact wording varies by macOS version)
In some macOS versions, the "New Terminal at Folder" option is nested under a "Services" submenu (see screenshot below).
Once Terminal is open, type claude and press Enter to launch CC.
“New Terminal at Folder” menu option under Services in Finder
A.4. Ask Claude to create shortcuts
Once you've launched CC from a project folder the manual way, you can ask it to create a shortcut so you don't have to repeat those steps.
How: With CC open in the project folder, use a prompt like this:
I need a shortcut or keyword like [projectname] that I can run from anywhere in Terminal, which will automatically change directory to [the project folder location] and launch Claude Code from there. What info do you need from me so you can create that shortcut for me?
Ask Claude Code to create the shortcut
Once the shortcut is created, launch Terminal from anywhere (via Spotlight Search or a Dock icon), type your keyword, and CC opens in the right folder automatically.
B. Permitting Claude Code to Take Action
B.1 The permissions menu
Anthropic designed Claude Code to ask for permission before it (a) changes the contents of your current folder, or (b) accesses files outside your current folder.
Let's familiarize ourselves with the chat window when permissions are requested.
The chat history (1) is the area above the horizontal line.The chat history
The permissions menu (2) appears below the horizontal line, but only when CC is requesting permission to do something.The permissions menu
B.2 Granting standard permissions
How: When the permissions menu appears, you'll typically see three options:
Yes
Yes, and allow [this] during this session
No.
Granting standard permissions
The default selection is always Option 1. The greater than symbol > beside Option 1 in the screenshot above indicates it's the currently selected option.
Press Enter to proceed with the highlighted option.
Type a number (1, 2, or 3) to select that option directly.
Use the arrow keys on your keyboard to move between options before pressing Enter.
Tips:
The mouse doesn't work in Terminal for positioning your cursor. The interface is text only; all inputs are via the keyboard. (We cover some keyboard shortcuts in Section D.5.) One thing you can do with the mouse: click and drag to select several lines of text if you need to copy them.
You can always ask Claude to explain why it's asking for specific permissions. If Claude asks for permission to do something and it's not clear to you what it will do, Press 3 (No) to deny the request, then ask for more information. For example, you might ask: "What is ~/.zshrc and why do you need to read it?" Ask as many questions as you need to be confident that it's safe to proceed. You'll find that CC is infinitely patient and great at providing explanations.
B.3 Canceling permissions
If Claude Code's proposed approach isn't what you want, cancel the permission request.
How: The final line (3) in the Permissions menu has instructions in case you forget: press the Esc key to cancel. CC will stop and wait for further instructions.
Canceling permissions
B.4 Be careful with Option 2
Option 2 is invariably some version of "Yes, and allow [this] during this session," that way you don't have to keep approving every individual action. Ordinarily, that would be a good thing, but in practice, it tends to grant CC more access than it needs.
For example, choosing option 2 in response to a file-read permission request will typically grant CC permission to read all other files in the same folder, which, in our specific example, happens to be the entire user folder!
Be careful of Option 2
Had I granted this permission, I would have effectively granted Claude Code access to all files that I have access to as the currently logged in user, including files I don't want it to access.
If you do hit Option 2 by mistake: Press the Esc key (repeatedly, if needed) to stop CC from executing further. Then type /exit followed by Enter to end the session, thus rescinding the overly broad permission. Starting a new session will give you strict permissions again.
B.5 Amending before granting permission
Why: Sometimes, you want to say "Yes, and…" or "No, and…" in response to a permission request, because you already know what you want it to do next. That's what the Tab key is for.
When Option 1 (Yes) or Option 3 (No) is highlighted, pressing Tab turns it into an amended statement.
For example, pressing Tab with Option 3 highlighted turns it into "3. No and tell Claude what to do differently," with the cursor placed after "No" so you can type additional instructions before pressing Enter.
Press tab to add further instructions
C. Brainstorming What to Work On
C.1. Start with a "role + task" brainstorming prompt
If you're not sure about what Claude Code can do for you, start by giving it real context about your role and your work. This turns Claude into an effective brainstorming partner.
How: Use a prompt like this to kick off a new project:
I am a [role]. As part of my job, I am responsible for [task/outcome], which I have to do every [frequency].
I use these tools to perform this task: [x, y, z].
These are roughly the steps involved: [n, o, p, q].
My biggest frustrations with this responsibility right now are: [a, b, c].
I think it would be awesome if the following statements were true: [h, i, j].
With all that in mind, what are some of the ways that you can help? Using the AskUserQuestion tool, interview me until you have enough context to give me a thoughtful answer with 5 to 10 options. Don’t ask obvious questions; dig into the hard parts I might not have considered.
Ideally, the options range from quick wins (easy to do, modest benefit) to solid wins (moderate effort, real savings in time and effort), with at least one ‘sky’s the limit’ option (requires creative, out of the box thinking, but opens new opportunities).
Tips:
Bring documentation. If you already have docs describing your tools and workflow, provide them as supporting files rather than typing everything out.
Be specific about frustrations. The more precise you are about what's painful and what "better" would look like, the more useful CC's suggestions will be.
Don't settle on the first batch. If the initial options are not what you're looking for, ask: "Give me 10 more ideas." Repeat until something clicks.
Explore before committing. If an idea sounds interesting but you're not sure how it would work, ask: "Tell me more about Idea 1. How would that work? Play out the workflow for me."
Reset when needed. If CC misunderstands your direction, type /exit and restart with more context, or type /clear to wipe the chat history and start fresh without logging out.
C.2. Save your brainstorming results
Claude is exceptional at documentation. Don't let any of your brainstorming go to waste.
How: Use a prompt like this to capture the discussion and any decisions you've made:
In light of the context I’ve given you and everything we’ve discussed, please write a complete project description to “project-idea.md” and save it in our project folder.
Separately, please also document in “brainstorming.md” all the options we considered, including any that I’ve deemed interesting or worth exploring, but which we haven’t discussed thoroughly yet.
If I’ve discarded any specific idea, briefly document that in bullet format with the reason I’ve provided for discarding it.
Tip: Read through the files Claude Code creates in response. Correct anything that looks off. This gives you a good grasp of how reliable CC's documentation is.
D. Working in Sessions
D.1 Customize your status line
Why: Claude Code's status line (the line that appears at the bottom of every active CC chat window) can be configured to give you useful info at a glance.
How: Use a prompt like this to set it up:
Please update my statusline settings to the following:
Line 1: Current Folder | Current Model | Token usage in percentage (number of tokens left). Line 2: session: [session id]
The modified status line; the session id is lightly redacted
The session id is a unique identifier that CC assigns to the current session. It's useful to have on hand if you expect to resume the conversation later. Leave it out if you don't need it.
D.2 Use documentation to maintain project continuity across sessions
Projects in Claude Code are rarely completed in one sitting. Without a deliberate handoff, the next session starts cold; CC has no memory of what you were doing or why. The fix: a standing instruction that closes every session cleanly.
How: Give CC a standing instruction at the start of any project:
Whenever I say “wrap up“, “close the session“, or similar, do all of the following so we can resume work seamlessly in a future session:
1. Create a session log in a session-logs/ subfolder named YYYYMmmDD-descriptive-stub.md with an H1 title, session metadata including session id, and a narrative summary of what was done, current state, and what’s next.
2. Update memory so the next session can resume seamlessly.
3. Update any working documents that should reflect decisions or changes made this session.
4. Update the project’s CLAUDE.md file if anything should permanently change how Claude approaches this project.
Tips:
The key phrase is "so we can resume work seamlessly." It prompts Claude to exercise judgment about what actually needs to be documented, rather than just logging a list of accomplishments.
Start a new session and say: "Let's resume" to continue working. Claude reads its CLAUDE.md and memory files and picks up where it left off.
Include the session ID in the log. It's a small detail that pays off if you ever need to trace a log back to the original conversation, to see what was actually said, not just what was summarized.
D.3 Manage your session's token usage
Claude Code has a context window — a limit on how much it can hold in memory at once. For Pro plan users, the context window is 200,000 tokens. For Max users, it's 1M tokens.
When the context window fills up, Claude auto-compacts: it summarizes the session and discards the details. Unfortunately, compaction can happen at the worst possible moment, such as when you're mid-task, and CC may drop context that you still need.
How: Manage token usage proactively:
One task per session. As soon as you complete a task, tell CC to end the session, type /clear, then start fresh for the next task. Do this even if your token usage is still low.
Watch the 40–50% mark. When the token usage percentage reaches this point, get to a logical stopping point. It may mean that you're still not done with the main task, but you've completed a subtask. End the session, /clear, then resume work.
Always start large tasks fresh. Even if your context usage is still low after performing a simple task, get a new session before tackling something large and complex.
If you've set up the wrap-up trigger from Section D.2, use it regularly. It closes the session cleanly before you /clear.
Waiting until the context window is full means CC gets to decide when to compact, not you.
This video from Nate B. Jones (hat-tip to Katie H. for the recommendation) is a solid resource. He advises using markdown files instead of PDFs or Docx files, keeping chats to 10-15 turns, disabling unused plug-ins and MCP connectors, matching the model to the task, and so on.
For more advanced users, Sabrina Ramonov outlined 6 ways she cut her Claude token usage in half, including selecting the right model, clearing context between tasks, using command line interface (CLI) tools, and more (hat-tip to Lauren I. for finding that gem).
Ruben Hassid offers 23 tips to maximize your use of Claude token limits across Claude Chat, Cowork, and Code (hat-tip once again to Lauren I. for sharing this).
D.4 Keep an eye on the usage limits
Aside from context limits in each chat session, CC has two additional limits: a 5-hour limit and a weekly limit.
The 5-hour limit is a rolling usage cap: the clock starts on first use and runs for 5 hours, resetting the next time you use CC if the previous 5-hour window has already expired.
The weekly limit is a hard ceiling on total usage across the week.
How: See your current, real-time usage at any time in Settings -> Usage. The same page shows when the limits reset and gives you the option of buying "Extra Usage" credits.
Tip: If you also use Claude on the web or Claude Cowork, those sessions draw from the same 5-hour and weekly limits. Anthropic's usage limit best practices are worth a read if you use these other Claude interfaces.
D.5 Learn some Keyboard Shortcuts
Navigating Claude Code in Terminal gets easier if you learn and use keyboard shortcuts. See the full list of shortcuts.
The ones I most commonly use:
Clear the prompt input: Ctrl+L
Move up/down one line on a multi-line prompt: Up/Down arrows
See previous prompt: Up arrow when you're already at the start of the current prompt
See next prompt: Down arrow when you're already at the end of the current prompt
Delete to the end of the line: Ctrl+K (which you can then paste elsewhere)
Paste deleted text: Ctrl+Y
Move back/forward one word: Opt+Left/Right Arrow
Adding a new line without submitting the prompt: \ + Enter
Since Claude Code has access to your project folder, every chat input and output can be a file, preferably in Markdown (see the Cheatsheet) because that format keeps file sizes small and token usage low.
Why it matters: Using files changes how you work:
No more copy-paste. Tell CC to read from a file in the project folder, or write its output to one. The clipboard stays out of it.
Prompts are much easier to reuse. Save prompts as files and you can use them repeatedly, and create versions by copying and renaming. Finding them again later is a breeze.
Files allow you to chain separate steps into workflows. The output of one task/step can serve as an input to the next step. Once you stop thinking of CC as a chat interface and start thinking of it as a pipeline, a lot of new workflow possibilities open up.
How: Use the @ convention to easily refer to specific files.
Anywhere in your chat, type @ followed by a few letters from the filename. Claude will suggest files from within the current folder and all child folders that match the letters you've typed.Type @ followed by some letters of the filename then press Tab to accept the highlighted suggestion
Use the arrow keys to select the correct filename, then press Tab to accept the selection. The selected filename will auto-fill, with the full directory path.
Tips:
For larger projects, create subfolders: inputs/, outputs/, research/, scripts/, tests/, and so on. When in doubt, you can ask Claude for advice on what folders to create.
Specify file naming conventions or let CC propose and adjust from there.
Provide an outline or template for each type of output file if you want a consistent structure and look.
E.2 Use a good markdown editor and viewer
Once CC is writing its outputs to local files, you need tools that make working with those files straightforward. Two that work well for me: BBEdit as my text editor and Marked 2 as my markdown viewer. Neither is free, but both have free trial periods.
BBEdit is a Mac text editor with strong markdown support — syntax highlighting, reliable search-and-replace, and clean navigation across large files.
Marked2 watches any markdown file that it has open and renders a live preview as you edit, so you can see the formatted output without publishing first.
I keep both open side by side when reviewing and editing Claude's output.
BBedit and Marked2, side by side, open to the same file
If you're going to be serious about managing token usage (as recommended in Section D.3), markdown files are the way to go, and a markdown editor and viewer becomes a must-have.
Tip: Google Docs supports markdown.
Enable markdown in Preferences. Open any Google Doc, click Tools –> Preferences –> "Enable Markdown" option should be checked. You only need to do this once.
With markdown enabled, you can download an entire doc as a markdown file. File –> Download –> Markdown (.md).Download GDocs as markdown files
You can Copy/Paste in Markdown. Select any section of your document, then click Edit –> Copy as Markdown so you can paste the selection as markdown elsewhere. Similarly, you can copy markdown from elsewhere, come to Google Docs, and click Edit –> Paste from Markdown to make GDocs paste the text with the Markdown format applied, so your original content's look is preserved.
Side note: There are many integrated development environment (IDE) tools like Cursor that allow you to visualize your project directory tree, chat with Claude Code, plus edit and view markdown files, all inside one interface. Some people find it easier to have everything all in one place. For now, I find it easier to use Claude Code, BBEdit, and Marked2 in concert, because I can position them just so across my three monitors. Your mileage may vary.
E.3 Make it compound — ask Claude to summarize what was learned each session
The session wrap-up practice in Section D.2 handles continuity: picking up where you left off. This section handles something different — distilling what you actually learned so that knowledge compounds across sessions.
The early sessions are always the slowest. We're still figuring out the problem space, testing approaches, running into dead ends. A lot of that trial and error contains real insight. But if it stays buried in the chat history, it resets with the session. This prompt surfaces it before it disappears.
How: For particularly meaty sessions with a lot of learning, run this prompt:
Read the full conversation history. Look for moments where we learned something. For example:
- A question got a clear answer
- Something didn’t work, then we made it work
- A decision was made and a reason was given
- A process, tool, or concept was explained or clarified
- A misconception was corrected
- A better way of doing something was discovered
Collect these as candidate lessons. Aim for substance over volume — a short list of real insights beats a long list of trivia. Discard things that are obvious, one-time-only, or too context-specific to be useful again.
Show me the list of candidate lessons as a numbered list. Keep each item to one sentence — enough to jog my memory, not a full explanation.
**Example format:**
Here’s what I think we learned in this session:
1. [One-sentence summary of lesson]
2. [One-sentence summary of lesson]
3. ...
Which of these would you like to document? You can name one, several, or all of them. (Say “none” or “skip” to end without saving anything.)
Work through each selected lesson one at a time. Assess where the lesson should live:
- Does this lesson change or contradict something already documented in this project (a plan, CLAUDE.md, a process doc, a decision record)?
- Does this lesson add nuance or a missing step to something that already exists?
Update or correct the existing project doc or add it to the most appropriate existing doc, or propose where it should go if no doc exists yet. Show the user your planned changes before proceeding. Wait for the user to confirm or redirect before making any change. After all selected lessons have been handled, give a brief summary. If nothing was saved, just say: No changes made.
Done consistently, this practice changes the way you work because the learning compounds. Working with CC stops feeling like starting from scratch each time and becomes a genuine, productive collaboration. The context builds, and so does the quality of the work.
Side note: Experienced CC users will note that the compounding prompt is best implemented as a Claude Skill. Skills and more are beyond the scope of this (already very long) introductory post.
In Closing
These five areas — setup, permissions, brainstorming, session management, and long-term workflows — cover most of what trips people up early on.
The underlying principle throughout is the same: don't let Claude Code be a passive chat window. Define what it has access to, document what it produces, and build on what it learns. That's when CC stops being a chat window and starts being something you can actually rely on.
I hope these tips help. The next step is to create a project folder and run the role + task brainstorming prompt from Section C.1. That single session will show you more about what Claude Code can do than any guide can.
Change Log
Updated 2026Apr26: Used Simon Willison's markdown-copy-component tool to make prompts easier to copy/paste.
Updated 2026Apr16: Added a link to Anthropic's session management guide, which was published with the release of Opus 4.7.
Get new posts via email.
Sign up to get a weekly digest when new posts are published or subscribe via RSS.
