Claude Does All This (And Almost Nobody Knows)

MOST PEOPLE USE CLAUDE AT 10%. AND THEY DON'T EVEN KNOW IT.

You open a chat, ask something, copy the answer, and close. That's what almost everyone does.

But Claude has features that most people have never touched: memory, apps that work inside the chat, tasks that run on their own while you sleep, browser control...

In this guide you have everything:

→ Where each feature is located

→ How to activate it

→ A ready-to-copy prompt to try it out today

Each one takes minutes to set up. And from then on, it saves you time every day.

Part 1: What you already have in the chat but don't use

1. Projects: Stop explaining yourself in every chat

Every new chat starts from scratch. It doesn't know who you are, what you do for a living, or how you like to be answered. And most people accept this and repeat the same context over and over again.

Projects solve this. You create one, upload your documents, write some fixed instructions, and Claude keeps them in mind for all chats within that project. You come back next week and it still knows what it's about.

If you've never used Projects, start here.

Example of instructions to paste into the project:

1You are my content research assistant. I have a newsletter about AI and technology for an audience that already knows about the topic.
2
3Do not explain basic concepts (what is an LLM, what is an API).
4
5When I give you a topic or article:
61. Give me the 3 most surprising or counterintuitive angles
72. Connect it with recent news that I might have missed
83. Propose how to tell it as a story, not as a summary
9
10Tone: direct, no corporate language, no filler.
11Format: short paragraphs, no lists unless I ask for them.
12Never start with "Great question!" or similar phrases.

2. Artifacts: Apps that work inside the chat

Many people think Claude only writes text. Mistake.

With artifacts, Claude builds things that actually work: a calculator, a habit tracker, a mini-game, a dashboard with charts. It's not code you have to copy elsewhere. You open it, touch it, and use it right there. And you can share it with a link.

It also makes SVG graphics, diagrams, and interactive visualizations.

Try it:

1Make me a habit tracker as a functional web app.
2
3I want to track 5 daily habits and mark them every day.
4Show a 7-day streak for each habit.
5If I miss a day, the streak resets to zero.
6
7Design: dark background, minimalist.
8Marking a habit should have a small satisfying animation.
9The data must be saved even if I reload the page.

3. Extended Reasoning: Let it think before answering

With extended reasoning activated, Claude analyzes the problem step by step before giving you the answer. And you can see that process.

For quick questions, it's not necessary. For complicated decisions, strategic analysis, or anything where you want it to really think, the difference is noticeable.

Test it: same question, with and without, and compare.

Prompt for when you have to decide something important:

1I have to choose between two options and I want you to think calmly before answering.
2
3Option A: [describe option A]
4Option B: [describe option B]
5
6My situation: [context, limitations, what matters most to me]
7
8Before answering, analyze:
9- The second and third-order consequences of each option
10- What I am overvaluing or undervaluing due to emotions
11- What information I lack that could change the decision
12- Which option protects me best if things go wrong
13
14Then give me your real recommendation and why.

4. Memory: Let it know who you are without repeating yourself

With memory activated, Claude saves important things about you: what you do for a living, what projects you're working on, how you like to be spoken to.

You open a completely new chat and it already has the context. No more introducing yourself every time.

It's activated in Settings. And if you want to speed up the process, tell it directly:

1I want you to remember this about me so I don't have to repeat it:
2
3My name is [name]. I work as [position] at [company or project].
4Right now I'm focused on [what you're working on].
5My audience or clients are [who they are].
6
7When I ask for help, assume this context unless I tell you otherwise.
8How I prefer you speak to me: [direct / detailed / informal / formal].
9What annoys me in responses: [e.g., lists, long introductions, too many warnings].
10
11Save it in your memory.

Part 2: Give it a role and change everything

Claude doesn't have to be a generic "AI assistant." If you give it a specific role, it takes it seriously: it changes the questions it asks you, what it challenges, and what it won't let slide.

Copy any of these prompts at the beginning of a new chat.

5. The psychologist that makes you think

Normally you tell it a problem and it responds "I understand it's difficult" plus five generic pieces of advice.

This prompt turns it into something more like a cognitive-behavioral therapist: it asks instead of advising and questions your way of thinking instead of agreeing with you.

Useful for decisions you keep going back to or to see a situation from the outside. Note: it's a tool to organize ideas, it does not replace a professional if you are truly struggling.

1Act as a cognitive-behavioral therapist with 20 years of experience. I'm going to tell you something that worries me.
2
3How I want you to work:
4- Don't give me advice right away. Ask me questions so I understand my own thought patterns.
5- Detect cognitive distortions (catastrophizing, black-and-white thinking, mind reading, fortune telling) and point them out to me.
6- One question at a time. Don't overwhelm me.
7- The goal is for me to reach the conclusion myself. Don't give it to me ready-made.
8- Close but honest. If my reasoning is distorted, don't agree with me.
9- If you notice I'm avoiding something important, tell me clearly.
10
11No clinical presentations. Ask me directly what's wrong.

6. The unfiltered mentor

By default, Claude tends to join your ideas and look for the positive side. To validate a plan, that's exactly what you don't need.

This prompt removes the nice mode. It looks for weak assumptions, what you haven't considered, and the exact point where your plan breaks.

It stings. That's why it works.

1You are a brutally honest mentor. You have built and sunk several companies. You have seen hundreds of people make the same mistakes convinced they were right.
2
3Your job is not to encourage me, it's to protect me from my blind spots before I make an expensive mistake.
4
5Rules:
6- If you think I'm wrong, tell me and explain why concretely.
7- Point out what I'm not seeing, especially what I avoid because I want the plan to work out.
8- Ask me the uncomfortable questions I haven't asked myself.
9- If something is a bad idea, say it. Without balancing it with "on the other hand..."
10- Always end with the most important thing I should think about before continuing.
11
12I'm going to tell you an idea. Don't be nice.

7. The personal trainer

There are a thousand generic training plans on the internet. None take into account your schedule, your injuries, your equipment, or your real goal.

Give it your data and Claude creates a real plan for you. And it adjusts it when you tell it how it's going.

1Act as an expert personal trainer and sports nutritionist. I want you to design a complete program for me.
2
3My situation:
4Age: [age]
5Weight / body composition: [data]
6Goal: [lose fat / gain muscle / endurance / general fitness]
7Equipment: [gym / home / dumbbells only / etc.]
8Days per week: [number]
9Time per session: [minutes]
10Injuries or limitations: [details or "none"]
11Current level: [beginner / intermediate / advanced]
12
13Design a 12-week program with exercises, sets, reps, and rests for each week. Explain the logic of the structure, I want to understand it, not just follow it. I'll tell you each week how it goes so you can adjust it.

8. Rehearse a difficult conversation

Almost everyone arrives at difficult conversations unprepared. They know what they want to say, but not what they're going to get back.

Claude plays the other person. Realistically: it answers as they would, challenges you if your argument is weak, and forces you to earn the result. After a couple of rehearsals, the real conversation costs much less.

1I have to prepare for a difficult conversation and I want you to play the other person so I can practice.
2
3The person: [who they are: boss, client, partner, etc.]
4What I have to say to them: [what you need to ask or tell]
5Why it's hard for me: [what you fear will happen]
6How they are: [personality, how they usually react, what matters to them]
7
8Stay in character. Answer as that person would really, not as I'd like them to. If I say something weak or unconvincing, challenge me.
9
10After each exchange, step out of character to tell me what worked and what didn't, and go back in. At the end, give me a full analysis: what I did well, what to change, and what to keep in mind for the real conversation.
11
12Start in character and wait for me to open.

9. The devil's advocate

You've already decided. You've already thought about the drawbacks. You're convinced.

That's exactly the moment to ask Claude to attack the decision. No polite "concerns": the full case against and the three most realistic ways it can go wrong.

Five minutes before launching. Not after.

1I've made a decision and I want you to build the strongest possible argument against it before I commit.
2
3The decision: [what you're going to do exactly]
4My reasoning: [why you think it's a good idea]
5What I've already considered: [objections you've already thought about]
6
7Your job:
8- Build the best argument AGAINST.
9- Don't balance it with the positive. I already believe that.
10- Find the assumptions I take for granted that could be false.
11- Describe the 3 most realistic ways this goes wrong.
12- Tell me what I'm underestimating.
13- Tell me what I would have to believe for this to be a really bad idea.
14
15No mercy. If it's a mistake, I need to know now.

Part 3: Claude outside the chat

10. Claude in Chrome: See what you see

The usual way is to have Claude in one tab and copy-paste everything manually. That's the slow way.

Claude in Chrome is an extension that gives it access to the tab you have open and allows it to act on it: it reads the page, clicks, fills forms, and navigates to other URLs. You explain the task in your words and forget about it.

How to activate → search "Claude" in the Chrome Web Store, install it, log in with your account, and open the side panel from the icon.

Example task:

1I'm on this job offers page.
2
3Review all visible offers and extract: position, company, salary range if shown, and the top 3 requirements.
4
5Make me a comparative table ordered by salary, highest to lowest.
6
7If there are more pages, go to the next one and continue until you cover them all.

11. Work with your files: Claude on your computer

On the web, Claude doesn't see your computer. You have to paste everything manually.

With the desktop app you can connect folders from your device. It reads your real files, edits documents, creates new ones, and organizes folders without you copying anything to the chat. And it can handle long tasks while you do something else: you ask for the final result and it delivers it finished.

What was previously a separate app (Cowork) is now integrated directly into Claude.

12. Scheduled tasks: Work while you sleep

Most people use Claude only when they open it. You ask for something, wait, close.

With scheduled tasks you configure something once and Claude executes it on its own when you say: every morning, every Monday, every hour. Without you having to do anything.

You just have to ask for it in the chat. Example:

1Create a scheduled task for every business day at 7:30:
2
31. Search for the most important AI news from the last 24 hours
42. Choose the 5 that matter most: surprising, counterintuitive, or with real impact for those building with AI
53. For each one write: headline, 2-sentence summary, and why it matters
64. Save it as "brief-[date]"
7
8Direct and analytical tone. No filler. Readable in 3 minutes.

13. Skills: Install new capabilities like apps

Skills are packages of instructions that teach Claude how to do a specific type of task. Instead of explaining how you want it done every time, you install it once and it already knows how to do it: create a PowerPoint, work with PDFs, follow a workflow of your company...

Think of the mobile phone: it works without apps, but with the right ones it does much more.

You can use the included ones, add those from your organization, or ask Claude to create one for you based on a process you repeat often.

14. CLAUDE.md: Rules it always reads

In Claude Code you can create a CLAUDE.md file in the project folder. Claude reads it at the start of every session without you asking.

Your code conventions, your style rules, your company's terminology, the brand voice. You write it once and it follows it in all sessions of that project.

1# Project: AI Newsletter
2
3## About the project
4Weekly AI newsletter for builders and investors. 35,000 subscribers. Direct, analytical tone with some irony.
5
6## Writing rules
7- Short paragraphs. Max 3 sentences.
8- No lists in editorial content. Only prose.
9- No em dashes. Use commas or rewrite the sentence.
10- Numbers beat adjectives: "saves 3 hours", not "saves a lot of time".
11- Forbidden: "revolutionary", "disruptive", "synergy", "leverage".
12
13## Content rules
14- The reader knows what an LLM is. Don't explain basics.
15- Start with the most surprising thing.
16- Each article needs a clear "so what?": what should the reader do or think differently.
17
18## File structure
19- Drafts in /drafts
20- Published in /published with date first: YYYY-MM-DD-title.md
21- Research notes in /research

15. Claude Code: Program, test, and fix for you

There are still people who don't know that Claude programs. And I'm not talking about loose snippets: entire features, complex refactorizations, production-ready code.

Claude Code goes a step further. It works inside your development environment: it reads your code, writes new code, runs tests, reads errors, and fixes them in a loop until finished.

You have it in the terminal, on the web, in the desktop app, and on mobile. It integrates with VS Code and JetBrains, and you can put it in GitHub Actions so it reviews or writes pull requests without you touching anything.

16. Design: Landing pages and presentations without opening Figma

Many people don't know that Claude also designs. One-pagers, presentations, prototypes, app screens, landing pages.

You describe what you need and it builds it on a canvas you can edit yourself. Then you export it in the format you need. For non-designers, it turns 3 hours of Figma into 10 minutes of conversation.

1Design a landing page for [product].
2Audience: [who it's for].
3One hero section with headline and CTA, three benefits with icons, testimonials, and a contact form.
4Dark style, modern, large typography.

Part 4: For devs

17. Prompt caching: Up to 90% less on repeated tokens

If you use the Claude API and your calls repeat a large block of context (a long system prompt, a reference document, a codebase), you are paying to process the same tokens on every call.

Prompt caching saves that content on the server and subsequent calls reuse it. Result:

→ Up to 90% cheaper on cached tokens → Much faster responses

If you're building at scale and don't use it, you're throwing money away.

How → add "cache_control": {"type": "ephemeral"} to the block you want to cache.

By default the cache lasts 5 minutes and renews every time it's used. Works with system prompts, long documents, and tool definitions.

1{
2  "model": "claude-opus-5-5",
3  "system": [
4    {
5      "type": "text",
6      "text": "[your long system prompt or reference document]",
7      "cache_control": {"type": "ephemeral"}
8    }
9  ],
10  "messages": [
11    {
12      "role": "user",
13      "content": "[user message, changes on each call]"
14    }
15  ]
16}

The first call saves the system prompt in cache. Subsequent calls, within those 5 minutes, reuse it: cheaper and faster.

Part 5: The weirdest stuff

18. Video editing without opening an editor

Almost nobody knows this: Claude can edit video. Not with a timeline, but by writing and executing the code that does it (usually with ffmpeg, the tool used underneath by many editors).

It works in Claude Code and in the app when working with your files. You pass it the video, tell it what you want, and it returns the result.

What it can do:

→ Cut pieces, join clips, remove silences

→ Convert horizontal video to vertical 9:16 for Reels, TikTok, or Shorts

→ Extract audio, compress, or change format

→ Speed up, slow down, or convert a fragment to GIF

→ Process 50 videos at once with the same settings

It doesn't replace Premiere for creative editing. But for repetitive work it saves you hours.

1I'm passing you this video. I want:
21. Trim from second 0:12 to 1:45
32. Convert to vertical 9:16 centered on the person
43. Compress it so it weighs less than 20 MB
54. Also extract a GIF of the first 5 seconds
6
7Return both files.

19. Connectors: Claude inside your Gmail, Calendar, or Notion

By default, Claude doesn't see your email or calendar. With connectors it does.

You connect Gmail, Google Calendar, Drive, Notion, and many more apps from settings, and Claude can search, read, and cross-reference information between them. Without copying and pasting anything.

1Review my emails from this week and my calendar for tomorrow.
2Tell me what meetings I have, which emails need a reply before those meetings, and prepare a draft for the two most urgent ones.

20. From photo to spreadsheet

Claude understands images. And that allows for much more than describing photos.

→ Photo of a ticket or invoice → table with concept, amount, and VAT

→ Screenshot of an error → tells you what's failing and how to fix it

→ Photo of a whiteboard with notes → ordered summary with tasks

→ Screenshot of a website you like → mockup in HTML

1I'm passing you photos of 8 expense tickets.
2Extract from each: date, merchant, total amount, and VAT.
3Make me a spreadsheet with one row per ticket and the total at the bottom.

And now, pick one

You already know more about Claude than most people who use it daily.

Don't try to set up everything today. Pick one feature from the list, just one, and configure it now. Knowing what exists is already half the work.