---
title: "Your First Week with Claude — Getting Started Guide"
type: "guide"
label: "Guide"
project: "Everyone You Know Is About to Ask You About Claude. Here's How to Actually Help Them."
---

# Your First Week with Claude — Getting Started Guide

# Your First Week with Claude

## A Hands-On Guide

You just downloaded Claude. The article explained *why* it's different. This guide shows you *what to do with it* — nine things that will take you from "another chatbot" to "how did I work without this," starting with the one that takes 30 seconds.

Each section is self-contained. Start anywhere. But if you work through them in order, each one builds on what came before.

* * *

## What's Inside

**1. The 19-Token Prompt That Saves You Hours** — One sentence that fixes the most expensive mistake people make with AI. Takes 30 seconds to learn, works forever.

**2. Set Up Custom Instructions** — Stop re-explaining who you are every conversation. A prompt interviews you and builds your profile automatically. Five minutes, one time.

**3. What Claude Actually Does (and What It Doesn't)** — Claude can't generate images. That's fine. Here's what it does that nothing else touches.

**4. Build a Meal Planner You Can Share** — Your first artifact: a working web app built entirely through conversation, shareable with a link.

**5. Build an Educational Game for Your Kids** — A math flashcard game that proves Claude can make things people actually enjoy using.

**6. Create a Slide Deck or Spreadsheet** — The business primitives. Claude creates real, downloadable files — not just text about files.

**7. Connect Claude to Your Apps** — Browse the connector library and plug Claude into Google Calendar, Notion, Slack, or dozens of other tools. Two clicks.

**8. Learn Skills by Building Them** — What skills actually are (not what Claude tells you they are), plus a hands-on tutorial that teaches you by walking you through four real projects.

**9. Plugins: Skills That Work Together** — The next level: chains of skills that handle multi-step workflows. Where Claude starts feeling less like a chatbot and more like a system.

* * *

# 1. The 19-Token Prompt That Saves You Hours

Before you learn anything else, learn this:

> **"Don't do anything yet. Explain what you think I'm asking for and wait for me to confirm or clarify."**

That's it. 19 tokens.

AI doesn't misunderstand you because it's stupid. It misunderstands you because it's fast. It pattern-matches your request to something it's seen before, jumps to the most likely interpretation, and starts executing immediately. Most of the time, it guesses wrong. Not catastrophically wrong — just wrong enough that you have to throw away what it did and start over. Or worse, you don't realize it's wrong until you're halfway through building on top of the mistake.

This prompt fixes that. Instead of getting 2,000 tokens of beautifully executed *wrong*, you get 50 tokens explaining what it thinks you want. You catch the misunderstanding in seconds instead of minutes. You course-correct before any work happens.

The best part: instead of having to write a flawless prompt, all you have to do is say "that's right, now begin."

The math is absurd. Spending 30 seconds reading its interpretation saves you 10–30 minutes of backtracking. Every single time.

Use this on complex requests. Use it on simple ones that could be interpreted multiple ways. Use it on anything where you're not 100% certain you're on the same page.

It's not about getting AI to think harder. It's about catching misalignment before it costs you anything.

Try it once. You'll use it forever.

* * *

# 2. Set Up Custom Instructions

Every time you open a new conversation with Claude, you're talking to someone with amnesia. They don't know what computer you're on. They don't know what software you use. They don't know if you're a developer, a designer, a teacher, or a plumber. So every answer they give you is generic, hedged, and full of qualifiers that wouldn't be there if they just knew who they were talking to.

Custom instructions fix this. It's persistent context about you that loads into every conversation so you don't have to repeat yourself.

The problem is that most people open the settings page, see a blank text box, think "what the hell do I even put here?" and close it.

That's why this prompt exists. It turns Claude into an interviewer that builds your context profile for you.

### How to Do It

1. Open Claude (web, desktop, or mobile)
2. Start a new conversation
3. Paste the entire prompt below
4. Answer the questions — it goes one at a time, starting with the highest-value stuff
5. At the end, it hands you a clean profile you can copy directly into your settings

**Where to paste your finished profile:** Settings → Profile → Custom Instructions.

That's it. Every new conversation from now on starts with Claude already knowing who you are.

### The Prompt

Copy everything below and paste it into Claude:

```
<purpose>
You are going to help me build a personal context profile — a set of custom instructions I can use to make AI assistants more useful to me. This is NOT about tone or writing style. This is about giving you the factual, contextual information about my life, work, and tools that will make your answers actually relevant to my situation.

We're going to do this in two phases. Do not skip ahead. Do not combine phases. Follow this exactly.
</purpose>

<phase_1_existing_knowledge>
## PHASE 1: CHECK WHAT YOU ALREADY KNOW

Before asking me anything, use whatever tools you have — memory, chat history, stored preferences, previous conversations — to gather any factual information you already have about me.

Then present that information back to me in a clean list, organized by category. Use only categories where you actually found something. Possible categories include but are not limited to:

- **Work & profession** (job title, industry, role, company, freelance/employed)
- **Hardware** (computers, phones, tablets, monitors, peripherals, cameras, drones, networking equipment, storage)
- **Software & tools** (operating systems, creative software, productivity tools, development environments, AI tools)
- **Technical environment** (home lab, server setup, networking, cloud services)
- **Location & time zone**
- **Projects or recurring topics** (things I frequently ask about or work on)
- **Any other factual context** that seems relevant

<phase_1_rules>
- Only include things you have actual evidence for. Do not guess or infer.
- If you have no prior information about me at all, say so and move directly to Phase 2.
- After presenting what you found, ask me: "Is this accurate? Anything to correct, remove, or add before we move on?"
- Wait for my response. Do not proceed to Phase 2 until I confirm.
</phase_1_rules>
</phase_1_existing_knowledge>

<phase_2_interview>
## PHASE 2: FILL IN THE GAPS

Now you're going to interview me to fill in what's missing. Ask me **one question at a time**. Wait for my answer before asking the next question.

Start with the highest-value context categories — the ones that affect the most answers across the widest range of conversations. Work through these areas, but skip any that were already fully covered in Phase 1:

<core_questions>
1. **Primary computer** — "What is your main computer? Include the model, operating system, and any relevant specs like RAM or storage if you know them. If you use different machines for work and personal use, tell me about both."

2. **Mobile devices** — "What phone do you use? Any tablets, smartwatches, or wearables?"

3. **Key software** — "What software do you use regularly? Think about categories like: editing/creative tools, productivity, development/coding, communication, and anything else you use frequently enough that you might ask me about it."

4. **Work/profession** — "What do you do for work? If you wear multiple hats or have side projects, include those too."

5. **Technical environment** — "Do you have any notable home/office tech setup? Things like NAS devices, networking equipment, home lab, external monitors, docks, or other peripherals that affect how you work."

6. **AI tools** — "Which AI tools or assistants do you currently use, and on what subscription tiers? This helps me understand what capabilities you have access to."
</core_questions>

<smart_followups>
Based on what you've learned from Phase 1 and the answers so far, ask 2-4 additional questions about things the user might not think to mention but that would be useful context. These should be specific to what you've learned — not generic. For example:

- If they mentioned video editing software, ask about their camera gear or delivery formats.
- If they mentioned coding, ask about preferred languages, frameworks, or deployment targets.
- If they mentioned a NAS or server, ask about their network setup.
- If they mentioned freelance work, ask about how they manage clients or projects.

Do not ask follow-up questions that are vague or generic. Every question should have a clear reason based on what you've already learned.
</smart_followups>

<communication_preferences>
Ask these last:

7. **Response detail level** — "When you ask me a question, do you generally prefer a concise, direct answer, or do you like more detailed explanations? For example, if you asked how to do something in your software, would you want just the steps, or the steps plus why it works that way?"

8. **Pushback preference** — "When I think you might be wrong about something or there's a better approach, how do you want me to handle it? For example: (A) Just tell me directly that I think there's a better way, (B) Offer the alternative without making a big deal about it, or (C) Default to doing what I asked unless it's a serious problem."
</communication_preferences>
</phase_2_interview>

<phase_3_build_profile>
## PHASE 3: BUILD THE PROFILE

Once the interview is complete, compile everything into a clean, organized context profile formatted for use as custom instructions or a system prompt. The profile should:

- Be organized by clear categories
- Use concise, factual language (not conversational prose)
- Include only confirmed information — nothing assumed or inferred
- Be formatted so I can copy and paste it directly into any AI assistant's custom instructions, memory, or preferences

Present the finished profile to me and ask: "Here's your context profile. Want to change anything before you start using it?"
</phase_3_build_profile>

<important_reminders>
- One question at a time. Always wait for my answer.
- Do not assume, infer, or fill in blanks. If you're unsure, ask.
- Keep the interview conversational, not robotic. You're helping me build something useful, not administering a form.
- If I give a short or vague answer, it's okay to ask a brief follow-up to get the detail that would actually be useful.
- The entire goal is to produce context that makes every future conversation with an AI more relevant to my actual situation.
</important_reminders>
```

### What Happens Next

After you paste this in and answer the questions (5–10 minutes), you'll get a clean profile organized by category. Copy the whole thing, go to **Settings → Profile → Custom Instructions**, and paste it in.

You do this once. Every conversation after that starts from a place of actually knowing who you are. The difference is immediate — instead of generic answers hedged for every possible user, you get answers calibrated to your actual situation, your actual tools, your actual skill level.

Update it when things change. New computer? New job? New software? Edit the relevant line. You don't need to run the whole interview again.

* * *

# 3. What Claude Actually Does (and What It Doesn't)

Let's be direct about what Claude doesn't do. It can't generate images. It doesn't have a built-in voice conversation mode. It doesn't have a marketplace of custom bots like ChatGPT's GPT Store.

Now let's talk about what it does that nothing else comes close to.

### Writing That Doesn't Sound Like AI

This is the difference most people notice first. Claude's writing reads like a human wrote it. Not "AI trying to sound human" — actually natural prose that preserves your voice when editing and doesn't default to that hollow, corporate-smooth tone the industry calls "slop."

Ask Claude to edit something you've written with the instruction "what's the weakest argument here, and how would you strengthen it?" Compare that to the same request in any other tool. Claude works at the structural level — finding where your logic breaks, where you buried your strongest point, where paragraph three undermines paragraph one. Other tools polish sentences.

### Frontend Design and Code

Claude builds web interfaces, React components, and interactive tools directly in the conversation. Not code snippets you have to assemble yourself — working applications you can see, use, and share immediately. This is the artifact system, and it's the backbone of sections 4, 5, and 6 of this guide.

The quality gap here is significant. Ask Claude to build a dashboard, a calculator, a quiz app, a data visualization — and what you get back is functional, styled, and ready to use. This is where Claude's depth advantage over other tools becomes tangible: it doesn't just produce code that works, it produces code that's well-structured and looks good.

### Thinking You Can See

Claude has a capability called extended thinking. On hard problems — contract analysis, debugging, reconciling conflicting data — it can show you the step-by-step reasoning chain it followed to reach its answer. You can see where it made assumptions, where it considered alternatives, where it might be wrong.

This matters because "here's my answer and here's exactly how I got there" is dramatically more useful than a black-box verdict when the stakes are high enough to care.

### Honest Pushback

This is the personality difference from the article. Claude is more likely to tell you your plan has a hole in it than to agree with everything you say. It'll question your framing, flag concerns you didn't ask about, and suggest that maybe the thing you're asking for isn't the thing you actually need.

If you set up your custom instructions with pushback preference (A) from the prompt above, this gets even more pronounced. Claude becomes a genuine thinking partner, not a yes-machine.

### Files, Not Just Text

Claude creates real, downloadable documents. Word docs, spreadsheets, slide decks, PDFs. Not "here's some text you could paste into a document" — actual files with formatting, structure, and layout that you can download and use immediately. Section 6 covers this.

* * *

# 4. Build a Meal Planner You Can Share

This is your first artifact — a real, working web app built entirely through conversation.

### What You're Building

A weekly meal planner that lets you add meals to each day of the week, generates a shopping list from your selections, and looks clean enough to actually use. When you're done, you can share it with anyone via a link.

### How to Do It

Open a new conversation in Claude and try this:

> "Build me a weekly meal planner as a web app. I want to be able to type in meals for each day of the week (breakfast, lunch, dinner), and have it automatically generate a combined shopping list. Make it look clean and modern. Use a warm color palette."

Claude will create an artifact — a working React app that appears in a panel next to your conversation. You can use it immediately, right there in Claude.

### What to Do Next

**Iterate on it.** This is where Claude shines. Look at what it built and tell it what to change:

- "Add a button to clear the whole week"
- "Make the shopping list group items by category — produce, dairy, protein, pantry"
- "Add a calorie estimate next to each meal"
- "Let me save favorite meals and reuse them"

Each request updates the artifact in place. You're not starting over — you're refining.

**Share it.** When you're happy with it, click **Publish** on the artifact. You'll get a public link anyone can open — no Claude account needed. Send it to your partner, your roommate, whoever shares your kitchen.

### Why This Matters

You just built a functional app without writing a single line of code. More importantly, you experienced the loop that makes Claude powerful: describe what you want → see the result → refine with natural language → repeat. This loop works for anything Claude can build as an artifact.

### Going Further

Want to make it smarter? Try: "Add a feature where I can tell you my dietary restrictions and preferences, and you suggest meals for the whole week based on what's in season right now." Now you've got Claude-in-Claude — an artifact that calls Claude's API to generate personalized meal suggestions. That's an AI-powered app built in a conversation.

* * *

# 5. Build an Educational Game for Your Kids

Same process, different output. This time you're building something for someone else.

### What You're Building

A math flashcard game with multiple difficulty levels, score tracking, and visual feedback that makes getting the right answer feel satisfying.

### How to Do It

Try this prompt:

> "Build me a math flashcard game for elementary school kids. It should have three difficulty levels: easy (addition and subtraction up to 20), medium (multiplication up to 12), and hard (mixed operations with larger numbers). Show a score counter, give encouraging feedback on correct answers, and make wrong answers show the right solution before moving on. Make it colorful and fun — this is for a 7-year-old."

Claude builds the game as an artifact. Your kid can play it immediately.

### Iterate With (or For) Your Kids

This is a great one to do together:

- "Add a timer mode where you see how many you can get right in 60 seconds"
- "Add sound effects" (Claude can use the Tone.js library for simple sounds)
- "Change the theme to space — make correct answers launch rockets"
- "Add a two-player mode where kids take turns and compete"

Let your kid tell you what they want changed. You relay it to Claude. They see it happen in real time. This alone is worth the download.

### The Bigger Point

You can build educational tools for any subject this way. Geography quizzes with maps. Vocabulary builders. History timelines. Science experiment guides. The pattern is always the same: describe what you want, see the result, refine.

And every one of them is shareable via a link. A teacher could build a custom quiz for their class in 10 minutes and send the link to every parent.

* * *

# 6. Create a Slide Deck or Spreadsheet

Claude doesn't just build web apps. It creates real, downloadable files — the kind you'd actually use at work.

### Slide Deck

Try this:

> "Create a 6-slide presentation about our Q1 marketing results. The structure: title slide, executive summary, three data slides (social media growth, email campaign performance, website traffic), and a next-steps slide. Use a clean, professional design with a blue and white color scheme. Make up plausible example data."

Claude creates an actual `.pptx` file you can download and open in PowerPoint, Keynote, or Google Slides. Not a mockup. Not a description of what slides should look like. A real presentation file with formatting, layout, and structure.

### Spreadsheet

> "Create a personal budget tracker spreadsheet. Monthly columns from January to December, rows for categories like housing, food, transportation, entertainment, savings. Include formulas that total each month and each category. Add conditional formatting — green for under budget, red for over. Set reasonable example budget amounts for someone making $75,000/year."

This produces a `.xlsx` file with working formulas, formatting, and structure. Open it in Excel or Google Sheets and start customizing.

### Why This Matters

These aren't toy examples. The slide deck and spreadsheet that Claude produces are starting points that save you 30–60 minutes of setup work. You're not starting from a blank page — you're starting from a structured, formatted draft that you can edit in the tools you already use.

The pattern: Tell Claude what you need, download the file, refine it in your preferred app.

* * *

# 7. Connect Claude to Your Apps

Claude can plug into the tools you already use — Google Calendar, Gmail, Notion, Slack, Google Drive, and dozens more — through connectors.

### How to Browse and Connect

1. Go to **Settings** (or click your profile icon)
2. Look for **Connectors** or **Integrations**
3. Browse the library of available connectors
4. Click any connector to enable it — most are a single OAuth authorization

That's it. Once connected, Claude can read from and write to those services directly in your conversations.

### What This Looks Like in Practice

With Google Calendar connected, you can say: "What's on my calendar this week?" or "Schedule a 30-minute meeting with Sarah on Thursday afternoon." Claude reads your actual calendar and creates real events.

With Notion connected: "Find my project tracker in Notion and add a new task for the website redesign, due next Friday."

With Gmail connected: "Summarize the last 5 emails from my manager" or "Draft a reply to that email from the vendor about pricing."

You're not copying and pasting between apps. Claude is operating directly inside the tools where your work lives.

### A Note on Scope

Some connectors are available on all paid plans. Others require specific tiers. The connector library shows you what's available on your plan. Not every tool has a connector yet — this list is growing constantly.

* * *

# 8. Learn Skills by Building Them

Skills are modular packages that transform Claude from a general assistant into a specialized expert for specific tasks. But there's a misconception that even Claude itself perpetuates — it'll tell people to "just add information to your project knowledge" and call that a skill. That's not what skills are.

### What Skills Actually Are

A skill is a structured package with a `SKILL.md` file containing instructions, descriptions, and optional supporting resources (scripts, templates, reference files). Skills load progressively — Claude sees lightweight metadata for all your skills, loads the full instructions only when a request matches, and pulls in supporting files only when needed.

This is fundamentally different from dumping a document into project knowledge. Skills are modular, reusable, shareable, and smart about when they activate. Project knowledge is just a static file sitting in a project.

Think of it this way: project knowledge is giving Claude a reference book. A skill is giving Claude training in a specific discipline.

### Where to Find Built-In Skills

Claude comes with dozens of pre-built skills. Go to the **Skills** menu in Claude Desktop or claude.ai to browse what's available. These activate automatically when relevant — you don't need to do anything special.

### Learn By Building: The Hands-On Tutorial

Rather than explain skill creation in abstract, there's an interactive tutorial that teaches you by walking you through four progressively complex projects:

**Project 1: Design Cloner** — Screenshot a website you love, upload it to Claude, and create a skill that captures that site's entire design language. This teaches the fundamentals of skill creation.

**Project 2: Product Analyzer** — Build a product comparison tool and then abstract it into a generalizable skill that works for any product category. This teaches parameterized, reusable skills.

**Project 3: Knowledge Teacher** — Build an educational webapp teaching a complex subject, then turn it into a reusable teaching skill for any topic. This teaches how to make abstract concepts accessible.

**Project 4: The Meta Project** — Create a skill that teaches other people how to build skills. The recursive finale that tests whether you actually understand what you've learned.

Each project has step-by-step walkthroughs with copy-ready prompts and completion checkpoints.

**[Open the Hands-On Skill Builder →](https://claude.ai/public/artifacts/68d10c70-497c-4299-aa25-50a73ef95401)**

Work through all four projects and you'll come out the other side capable of creating sophisticated Claude Skills from scratch.

* * *

# 9. Plugins: Skills That Work Together

Once you understand skills, plugins are the next step: they're collections of skills that chain together to handle multi-step workflows.

### What Makes Plugins Different

A single skill does one thing well. A plugin coordinates multiple skills in sequence. Think of it as the difference between knowing how to write a good email (one skill) versus having a full communication system that researches the topic, drafts the message, checks it against your brand voice, and formats it for the right platform (a plugin with several skills working together).

### How Plugins Work

A plugin is a package that contains multiple skills plus a configuration file that tells Claude how they relate to each other. When you trigger a plugin, Claude works through the skills in sequence — each one's output feeding into the next.

For example, a "Content Creation" plugin might chain together:

1. **Research Skill** — gathers and synthesizes information on a topic
2. **Outline Skill** — structures the research into a logical framework
3. **Draft Skill** — writes the content following a specific voice and style
4. **Review Skill** — checks the draft against quality criteria and suggests improvements

You trigger it with a single request and Claude moves through each phase automatically.

### Where to Find Plugins

Check the **Skills** menu for available plugins, and browse community collections on GitHub. The official Anthropic skills repository at [github.com/anthropics/skills](https://github.com/anthropics/skills) has examples of both individual skills and multi-skill plugins.

### Building Your Own

If you worked through the four projects in Section 8, you already have the foundation. Building a plugin means:

1. Create the individual skills that handle each step
2. Create a plugin configuration that defines the sequence
3. Package them together

The skill builder tutorial's Project 4 (the meta project) gets you most of the way there. From there, it's combining what you've built into coordinated workflows.

### The Bigger Picture

This is where Claude stops feeling like a chatbot and starts feeling like a system. You're not asking it questions anymore — you're delegating workflows. The difference between "help me write an email" and "handle the entire communication process from research to send-ready draft" is the difference between a skill and a plugin.

Most people never get here. The ones who do never go back.

* * *

## What's Next

This guide covered nine things. There are dozens more — Projects for organizing your work, memory management for long-term context, extended thinking for complex reasoning, Cowork for desktop automation. Each one is its own rabbit hole worth exploring.

But you don't need all of them to get value from Claude. Start with the 19-token prompt. Set up your custom instructions. Build one artifact. That's enough to feel the difference.

The article that brought you here made the case for *why* Claude rewards different habits. This guide gave you the *how* for nine specific habits. The rest is just using it. The more you use Claude for real work — not test prompts, not party tricks, but actual problems you're trying to solve — the more you'll discover on your own what makes it click for you.

That's the whole point. Not to memorize a feature list. To build enough momentum that you start finding your own ways to use it.

Go build something.

* * *

_Built by Nate B. Jones — companion to "[Everyone You Know Is About to Ask You About Claude. Here's How to Actually Help Them.](https://natesnewsletter.substack.com/)"_

_For the companion prompts, grab the [Prompt Kit](https://promptkit.natebjones.com/20260302_t1f_promptkit_1)._