Leverage Ai is created and edited by Nate Kennedy. Each guide shares practical insights, strategies, and lessons learned from Nate and his team as they actively use AI to build and grow businesses every day.
TL;DR: Claude's Code tab has 8 built-in controls that make it answer in plain English, check its own setup, show what's eating your limit and undo bad turns. You need no add-ons and no coding, and you approve every change. Setup takes about 55 minutes.
Claude's desktop app has a Code tab packed with slash commands. A handful of them help any owner who writes proposals, emails, SOPs or client updates with Claude.
What you need: the Claude desktop app, a paid Claude plan (Pro, Max, Team or Enterprise), 1 empty test folder on your computer and about 55 minutes.
Step 1: Open the Code Tab in a Test Folder
Every control here lives in the Code tab.
Quick definition: A slash command is a short instruction that starts with /, like /doctor, that runs one of Claude's built-in tools.
Click path
Open the Claude desktop app and click the Code tab.
Make a new folder called "Claude Test." Pick it with the project folder control under the prompt box.
Type /status and press Enter. The Claude Code row shows your version.
Type / on its own to see every command your version offers.
Note: If a command below is missing from the / list, update the app: Claude → Check for Updates… on a Mac or Help → Check for Updates… on Windows. A few terminal-only panels reply that they aren't available in this environment. Skip those.
My default: keep this folder clean. No customer data, no patient details, no payment info and no passwords.
Step 2: Make Claude Answer in Plain English
Quick definition: An output style is a standing instruction that sets how Claude talks to you for a whole session.
Claude Code ships 4 styles besides Default, including Concise, which puts the answer first and drops the recap.
In the Code tab, you set a style in a settings file, so let Claude do it. Paste:
Set my output style to Concise for every project by adding "outputStyle": "Concise" to ~/.claude/settings.json. Show me the change before you save it.
Approve the change, then start a new session with + New (Cmd+N on a Mac, Ctrl+N on Windows).
Want it plainer still? Paste:
Write me a custom output style called Plain English and save it in ~/.claude/output-styles. Follow these rules:
Explain things like I run a small service business and don't code.
Use short sentences and no jargon.
End every answer with my next step.
Show me the file before you save it.
Then ask Claude to set outputStyle to "Plain English" the same way, and start a new session.
Note: Some videos show an "explain like I'm 5" style. That one is a custom style a creator wrote for himself. The prompt above builds your own version.
My default: Concise for 1 week. Add Plain English if answers still feel technical.
Step 3: Run a Checkup With /doctor
Quick definition: A skill is a saved set of instructions Claude loads when a task calls for it.
Click path
Type /doctor and press Enter. Claude checks its install and settings, then proposes fixes. It applies one only after you confirm.
Type /skill-doctor and press Enter. It shows what each skill costs in working memory and how often you use it.
Then paste:
Based on that skill list, tell me which skills I could turn off without losing anything I use. Explain each one in plain English. Change nothing until I say yes.
Note: /doctor can edit settings files. Say yes only to fixes you understand, and ask "what does this change do?" when unsure. If /skill-doctor is missing from your / list, skip it.
My default: /doctor once a month, and any time Claude starts acting strange.
Step 4: See What's Eating Your Limit
Long sessions quietly burn through your plan.
Quick definition: Context is everything Claude holds in working memory for this session: your messages, its replies, plus any files and skills it loaded.
Click path
Click the usage ring next to the model picker to see your plan usage.
Type /context for a breakdown of what fills this session's working memory.
Then paste:
I just ran /context. What's taking the most room, and should I start a fresh session for my next task?
Here's the trap. Claude Code keeps a ready copy of your session, called the prompt cache, so each new message is cheaper to process. On a paid plan, within your included usage, Anthropic says the main session's cache stays warm for 1 hour after the last request. Once you're drawing on paid extra usage, it drops to 5 minutes.
Come back after it expires and the next message reprocesses the whole session. Switching models mid-session does the same thing.
Note: You may see a free mod called Ribbon that shows usage. Mods are an official feature, and Anthropic warns that a mod is code running with your permissions. It can read files, change files, start programs and make network requests. Our take: skip third-party mods. And never paste a GitHub link into an Ai and say "install this."
My default: new job, new session. Once a month, type /insights for a report on how you've used Claude Code on this computer.
Step 5: Bring In a Second Opinion for Big Calls
Your everyday model can ask a stronger one for help.
Quick definition: The advisor is a second, usually stronger model that Claude consults at key moments, like before it commits to a plan or declares a task done.
Click path
Type /advisor opus and press Enter. Claude shows a notice that the advisor "may use more tokens," then decides on its own when to call it.
Type /advisor off when the job is done.
Note: Each time Claude calls the advisor, it reads the full conversation. On a paid plan, Anthropic says that counts toward your usage limits, in addition to your main model. Anthropic also says it adds less on short tasks.
My default: off for emails and quick drafts. On for 1 long job, like turning 20 pages of notes into an SOP manual.
Step 6: Undo a Bad Turn With /rewind
When Claude goes the wrong way, roll back.
Quick definition: A checkpoint is a save point Claude makes each time you send a prompt, covering the conversation and the files it edited.
Click path (terminal version)
Type /rewind, or press Esc twice with an empty prompt.
Pick the message to go back to. Choose Restore conversation to keep the files, or Restore code and conversation to undo both.
In the Code tab: type /rewind. If it says it isn't available in this environment, ask Claude to undo its last changes and show you exactly what it reversed.
Before you roll back, paste:
This last answer went the wrong way. Before I rewind, tell me what I'd lose if I go back to my previous message.
Note: Rewind tracks only the edits Claude made in this session, so changes you made yourself stay put. Keep real backups of anything that matters.
My default: rewind early. 1 bad turn is easier to undo than 5.
Step 7: Catch Up Fast and Ask Side Questions
Quick definition: A side chat is a quick question that can see your session but adds nothing back to it.
Click path
Type /recap for a one-line summary of the session so far.
Type /btw plus your question to open a side chat.
Try this:
/btw In plain English, what is this session working on right now, and what's left?
Note: The desktop app doesn't save side chats, so you can't reopen one after you close the app. Copy anything you need first.
My default: /recap every time you sit back down after a client call.
Step 8: Learn the Keys That Save You Clicks
A few keys replace a lot of menu hunting.
Quick definition: Quick entry is a small Claude chat box that opens over any app on a Mac.
Click path
Press Cmd+/ on a Mac or Ctrl+/ on Windows to see every Code tab shortcut.
Press Cmd+; or Ctrl+; to open a side chat.
Press Esc once to stop Claude mid-answer.
On a Mac, double-tap Option to open quick entry. Change it in Settings → Desktop app.
Note: If you hear that Ctrl+D starts voice typing, skip it. In the terminal, Ctrl+D closes Claude Code.
My default: learn Cmd+/ first. It lists the rest.
Start Here This Week
Day 1: open the Code tab in a test folder and type / to see your commands.
Day 2: set Concise and run /doctor.
Day 3: check your usage ring and /context after a normal work session.
Day 4: try /btw and /recap on a real task with no client data, then press Cmd+/.
Next week: turn on /advisor for 1 long job, then turn it off.
LEVERAGE MATH: What 8 Claude controls take
What it takes (our estimates)
Step 1, Code tab and test folder: 5 minutes
Steps 2 and 3, plain style plus checkup: 20 minutes
Steps 4 and 5, usage check plus advisor: 15 minutes
Steps 6 and 7, rewind plus side chats: 10 minutes
Step 8, keyboard shortcuts: 5 minutes
Total setup: about 55 minutes
Every week after: about 5 minutes
What to expect
Cache window: 1 hour on a paid plan within included usage, 5 minutes on extra usage, per Anthropic
/doctor fixes: nothing changes until you confirm
Your turn (week 1, then week 4)
Times you hit your usage limit: ____
Fresh sessions started for new jobs: ____
Bad turns undone with /rewind: ____
Skills turned off after /skill-doctor: ____
Here's my honest read.
The biggest win here is a boring habit: start a fresh session for each new job. It protects your limit better than any add-on.
Anthropic updates Claude Code often. If a command here looks different on your screen, type / and go with what you see.
Claude drafts and suggests. You approve every fix and anything that goes to a client.
Disclosure: I'm not paid by Anthropic to write this.
Nate
Build AI Employees, Free Yourself From Repetitive Work, and Build a Lean, Profitable Business That Scales.
We help business owners, leaders, and their teams to Leverage Ai, so you can build a business powered by AI employees that gives you more money and more freedom than you ever imagined.
Want to learn if installing your custom Ai Workforce will work for you?