Tips & Prompts
Practical advice for getting the most out of Claude — prompts, commands, and troubleshooting you'll come back to. Cowork tips up top, then Claude Code further down.
Tips for Working with Cowork
Cowork is the best fit for most operators and small business owners. These tips come from real use — not theory.
Start with one connector at a time
Hook up Google Workspace (or Microsoft 365) first — that alone unlocks email, docs, sheets, and calendar. Add Slack, Notion, and others as you find a real need. Don't connect everything at once.
Describe the outcome, not the steps
"Pull the last 30 customer reviews and tell me the top three complaints" works better than "open Reviews.csv, then sort by date, then…" Cowork figures out the steps. Your job is the goal.
Tell it where to put the result
"…and save the summary as a Google Doc in my BTC folder." Cowork is happy to create files; it just needs to know where they should land. Saves you a copy-paste later.
Walk away — but check the work
Cowork can run for many minutes on a real task. Let it. When you come back, skim the result before you trust it — same as you would with a new assistant.
Low-stakes first
Start with research, drafts, and summaries before pointing Cowork at customer or financial data. Build trust the same way you would with any new tool — small wins, then bigger ones.
Write your prompts like a brief, not a question
Cowork rewards specifics: who is this for, what should the output look like, what tone, how long. Two extra sentences in the prompt save a round of revisions.
Cowork prompts to copy
Customer follow-ups
"Open my 'Customers' Google Sheet. For everyone whose last contact date is more than 30 days ago, draft a friendly check-in email referencing their last project. Save the drafts in Gmail — don't send them. Make a quick list of who you drafted for."
Competitor research
"Research the top 5 [your industry] businesses in [your city]. For each: services offered, pricing if visible, what they emphasize on their site. Put it in a Google Doc called 'Competitor scan — [today's date]'."
Inbox triage
"Look at my unread Gmail from the last 3 days. Group it: needs a real reply from me, FYI only, can be archived. Don't actually move anything yet — just give me the list."
Meeting recap
"Read the notes in my 'Discovery — Todd' Google Doc. Write a one-page recap I can email back to him: what I heard, what we agreed to, and the next two steps. Keep the tone warm and direct."
More: Cowork guide for setup, examples, and a "first week with Cowork" plan.
Example Prompts
The more specific you are, the better the result. Here's the difference — and examples for common situations.
Vague vs. Specific
Vague
"Make me a to-do list app."
Specific
"Make me a to-do list where I can add tasks, check them off, and delete them. Group tasks by category. Use a clean, modern design with soft colors."
Vague
"Fix the colors."
Specific
"The red background is too aggressive. Use a soft blue-gray instead, and make the text darker so it's easier to read."
Prompts for Common Situations
Starting a new project
"Build me a personal recipe book. I want to add recipes with a title, ingredients, and instructions. Let me search by ingredient. Save everything in my browser."
"Create a birthday tracker where I can add people's names and birthdays, and it shows me whose birthday is coming up next."
Changing the design
"Make the header bigger and use a dark blue background. The current font feels too small on mobile."
"Add some spacing between the cards. They feel too cramped. And round the corners more."
Adding a feature
"Now add a search bar at the top that filters the list as I type."
"Add a dark mode toggle in the top right corner. Remember my preference when I come back."
Fixing something that's wrong
"When I click the delete button, nothing happens. Can you figure out why and fix it?"
"The page looks fine on my computer but everything is squished on my phone. Make it work on both."
Understanding what Claude built
"Explain what you just did in simple terms. What files did you create and why?"
"Walk me through how this app works — like you're explaining it to someone who's never coded."
When you're stuck
"Something isn't working but I'm not sure what. Can you check for errors and tell me what you find?"
"I described what I wanted but the result doesn't match. Here's what I expected vs. what I got: ..."
Tips for Working with Claude Code
Describe what you want — don't fix code
Be specific about what it should look like and how it should behave. If something's wrong, describe the problem instead of trying to edit code yourself: "The button is too small," "Sort by date instead of name."
You're the director, not the coder
Think of yourself as a movie director. You don't operate the camera — you describe the scene you want. Claude is the crew that makes it happen.
Review before you accept
Claude Code will ask for your permission before making changes. Take a moment to look at what it's proposing. Try it out, and if something isn't right, tell Claude what to change.
Start small, then add features
Build the simplest version first, make sure it works, then add one thing at a time. "Now add a search bar." "Now add dark mode." This works much better than asking for everything at once.
Use Plan Mode for big projects
Press Shift+Tab (cycles through: normal → auto-accept → plan mode) or type /plan. Claude will research and plan first, then show you the plan before making any changes. Great for anything with more than a few steps.
Use /compact in long sessions
If a conversation gets long, type /compact to compress the history. Claude keeps the important context but frees up memory so it stays sharp. Essential for sessions that last more than 20-30 minutes.
Keep sessions focused
Claude has a limited memory window — in very long conversations, it may lose track of earlier instructions. Use /compact to compress context, or start a fresh session with a good CLAUDE.md file.
Try Voice Mode
Type /voice to toggle voice mode. Hold the spacebar, speak your request, and release — Claude transcribes and runs it. Great if you'd rather talk than type. Available on Pro, Max, Team, and Enterprise plans.
Auto-Memory works for you
Claude now remembers things across sessions automatically — your preferences, project patterns, and debugging insights. You don't need to write everything in CLAUDE.md anymore. Type /memory to see what Claude has remembered and manage it.
Routines: put Claude on a schedule
Use /schedule to turn a prompt into a recurring task that runs on Anthropic's cloud — great for "generate the weekly report every Monday" or "summarize my inbox each morning." Pro includes 5 runs per day; Max includes 15. You don't have to keep Claude open for it to work.
Computer Use: the last-resort automation
Claude can see your screen and click around like a person — useful for automating desktop apps that don't have direct integrations. Slower than other approaches and still a research preview, but it handles things nothing else can. Available on Mac and Windows. Learn more
CLAUDE.md: Your Project's Memory
Claude now has auto-memory that carries some context between sessions — but a CLAUDE.md file is still essential. Auto-memory captures small preferences and patterns; CLAUDE.md gives Claude the big picture: your project's tech stack, conventions, and how you like to work. Think of it as the instruction manual you leave at the top of every session.
What is it?
A simple text file named CLAUDE.md that you put in your project folder. Claude reads it automatically at the start of every conversation, so it knows your project's rules and preferences without you repeating them.
Starter template
The easiest way to start: type /init in Claude Code and it will generate a CLAUDE.md based on your project. Or ask Claude: "Create a CLAUDE.md file with these details" and describe your project. Here's what one looks like:
# My Project
## What this project is
A personal recipe book web app.
## How it's built
HTML, CSS, and JavaScript. Single page app.
## My preferences
- Use simple, clean design
- Save data in the browser (localStorage)
- Mobile-friendly
It gets better over time
Every rule you add, every preference you record, makes the next session smarter. After a few projects, Claude knows your style before you type a word.
Keep it short
Community best practice: keep CLAUDE.md under 200 lines. If it gets longer, Claude may start ignoring parts of it. For each line, ask yourself: "Would removing this cause Claude to make mistakes?" If not, cut it.
You can also create a global CLAUDE.md file (in your home directory) that applies to all your projects — things like your preferred design style or how you like things explained.
Essential Commands Quick Reference
If you're using the terminal, here are the commands you'll use most. Just come back to this page when you need a reminder.
claude Start Claude Code. Type this in your terminal to begin a new conversation.
/help See a list of available commands.
/exit End the current conversation and go back to your normal terminal.
/clear Clear the conversation and start fresh — without leaving Claude Code.
/rewind Undo Claude's recent changes — like a quick undo button.
/compact Compress conversation history to free up memory in long sessions.
/cost See how much usage you've consumed in the current session.
/init Auto-generate a CLAUDE.md file for your project.
/plan Enter Plan Mode — Claude plans before making changes.
/voice Toggle voice mode — hold the spacebar and speak instead of typing.
/memory Browse and manage auto-memory — what Claude remembers about your project across sessions.
/powerup Interactive lessons that teach Claude Code features hands-on — great for beginners who learn by doing.
Shift+Tab Cycle between modes: normal → auto-accept → plan mode.
Escape Stop Claude mid-response. Hit Escape and redirect if it's going the wrong way.
You're Always in Control
Claude Code has a built-in permission system — it asks before making changes to your files. Nothing runs without your say-so.
"Will it delete my files?"
No. Claude asks permission before every change. And /rewind can undo recent changes.
"Can it see my whole computer?"
No. Claude Code only works inside the project folder you choose. It can't access other folders, your browser, or personal files.
"Is my data sent to the cloud?"
Your prompts and the files Claude reads are sent to Anthropic's servers to generate responses. Your files stay on your computer. Anthropic doesn't use your data to train models.
"What if I can't follow what it's doing?"
Ask Claude to explain: "What did you just do and why?" You can also use Plan Mode (Shift+Tab) so Claude explains its plan before making changes.
Start cautious — review every change, ask questions, keep permissions tight. As you build confidence, you can give Claude more autonomy. There's no rush.
Troubleshooting
Getting stuck is normal. Here's how to get unstuck.
"Claude made something, but it doesn't look right."
Describe exactly what's wrong: "The sidebar should be on the left, not the top." The more specific, the faster Claude can fix it.
"I see a bunch of red error text."
Don't panic — errors are normal. Copy the error and paste it to Claude: "I'm getting this error. Can you fix it?"
"Claude seems confused or is going in circles."
Press Escape to stop, then type /clear to start fresh. Re-describe what you want from the beginning.
"I can't find the files Claude created."
Ask Claude: "Where did you save the files?" It will tell you the exact folder.