# Starter prompts for Claude Copy any of these into Claude. They are starting points, not magic words. Change anything in [square brackets] to fit you, and reword the rest freely. Keep this file open in a window beside Claude as you work. These prompts ask Claude to go slowly, explain as it goes, and check with you before it creates or changes anything. You are always the one who says go ahead. --- ## Getting set up (your first hour) Set me up from scratch, and teach me as you go: I have just installed Claude Code and I am brand new to it. I want you to help me set up this project, and teach me what you are doing as we go. Work in small steps and stop for me after each one. Start by explaining what a CLAUDE.md file is, ask me a few short questions about me and my work, and draft one from my answers. Do not create any files until I have seen the draft and said go ahead. Build my folder structure: Help me set up a tidy folder structure for this project: numbered top-level folders so they stay in order, and a short README in each. Ask me what kind of work this covers, suggest folders, let me adjust them, and only create the folders and READMEs once I confirm. Explain the naming convention as you go. Interview me to write my CLAUDE.md: Interview me so you can write a good CLAUDE.md for this project. Ask me one question at a time and keep each one short. Cover: who I am and what this project is; how I like you to work; my hard rules and never-dos; where my important files live; and what I am focused on right now. Then show me a draft, let me edit it, and only save it once I am happy. Walk me through connecting a tool: I want to connect [Shopify / Gmail / Google Drive / Notion] to Claude. Walk me through it one step at a time, tell me what I am agreeing to when I sign in, and once it is connected, suggest two safe, read-only things I could ask for to check that it is working. Check my setup is healthy: Check whether my Claude setup is healthy and explain what you find in plain English: what is installed and working, what is missing or needs signing in, and what you would do next. If there is a command I should run myself, give it to me and tell me what a good result looks like. --- ## Learning as you go Teach me as we work: While we work, briefly explain what you are doing and why, especially anything I could reuse another time. If there is a simpler or more standard way to do what I asked, tell me before you start. Treat this as me learning the tool, not just getting the job done. What can you actually do here: Look at this project folder and tell me, in plain English, the kinds of jobs you could help me with here. Give me five concrete examples based on what you can actually see, not generic ones, and mark which are quick wins. Explain that like I am new: Explain [the last thing you did / this term / this command] as if I am new to all of this. Keep it short and plain, and tell me when it matters in practice. Build me a glossary: Make me a short glossary of the words I keep seeing here, for example CLAUDE.md, plugin, skill, subagent, MCP, connector, terminal. One plain-English line each, in the order a beginner would meet them. --- ## Everyday working Start a task well, brief first: Before you start, tell me back what you understand the goal to be, and your plan in a few steps. Flag anything you are unsure about. Wait for me to confirm before you do the work. Plan first, then do: Plan this before you change anything. List the steps, what each one touches, and anything risky. Once I approve the plan, carry it out and tell me what you actually did at the end. Review my work: Review [this file / this draft / what we just did] as a careful second pair of eyes. Tell me what is strong in a line, then what is weak or unclear with specifics, then the single most useful change to make. Tidy this folder: This folder has got messy. Suggest a tidier structure and a consistent way of naming the files. Show me the before and after first. Do not move or rename anything until I approve, and never delete anything without asking. Write it in my voice: Write [the thing] in my voice. Match how I write in [point to a file, or paste an example]: the same plainness, the same length of sentence, the same words I would and would not use. No hype, no filler. --- ## Keeping Claude sharp (memory) Save this as memory: That is worth remembering for next time. Save it as a short, clear note so you recall it in future sessions, and tell me where you saved it. Update my CLAUDE.md: We have changed how this project works. Read my CLAUDE.md, tell me which parts are now out of date, and propose edits. Show me the changes before you save them. End-of-session recap: We are wrapping up. Give me a short recap: what we did, what is left, and the first thing to pick up next time. If anything from today should go into my CLAUDE.md or memory, say so. --- ## Connecting your tools Shopify, read-only to start: My Shopify store is connected. To start, only read, do not change anything. Show me my orders from the last seven days and my five best-selling products, and tell me where the numbers came from. Shopify, a careful change: I want to [describe the change] in my Shopify store. Before you touch anything, tell me exactly what you are about to change and show me the current values. Make the change only after I confirm, then read it back so I can see it worked. --- ## Working safely Ask before anything risky: For the rest of this session, check with me before anything that sends, publishes, changes or deletes data. Reading and drafting are fine to do on your own. When you do want to make a change, show me what and where first. Dry run first: Show me what this would do without doing it. List every file or setting it would change and how. I will tell you which parts to actually carry out. --- If you find yourself pasting the same prompt often, ask Claude to turn it into a custom skill or a saved command for your project, so it is one short word next time.