@wanerfu: https://x.com/wanerfu/status/2109891412829560863
Summary
A Chinese-language guide covering the Claude features most people don't know about (Projects, Artifacts apps, extended thinking, and memory), plus Spanish prompt templates you can copy and paste directly — including role-play prompts such as a psychologist and a critical mentor.
View Cached Full Text
Cached at: 10/13/26, 11:42 AM
Claude Can Do All of This (But Almost Nobody Knows)
Most people only use 10% of Claude, and they don’t even realize it.
Open the chat box, ask something, copy the answer, close it. That’s as far as most people ever go.
But Claude has a bunch of features most people have never touched: memory, mini apps that run right inside your conversation, tasks that execute while you sleep, browser control…
This guide covers all of them:
→ Where each feature is hidden → How to turn it on → A prompt you can copy and try today
Setting each one up takes just minutes, and they’ll save you time every day after that.
Part 1: Stuff That’s Been in the Chat Box All Along — That You’ve Never Used
image
image
1. Projects: Stop Reintroducing Yourself in Every New Chat
Every new conversation starts from scratch. It doesn’t know who you are, what you do, or what kind of responses you like. Most people just accept it and repeat the same background over and over.
Projects are the fix. Create a project, upload your documents, write a set of standing instructions — and every conversation inside that project carries them forward. Come back next week, and it still remembers.
If you’ve never used Projects, start here.
Example instructions to paste into a Project:
Eres mi asistente de investigación de contenido. Tengo una newsletter sobre IA y tecnología para un público que ya sabe del tema.
No expliques conceptos básicos (qué es un LLM, qué es una API).
Cuando te pase un tema o un artículo:
- Sácame los 3 ángulos más sorprendentes o contraintuitivos
- Conéctalo con noticias recientes que se me hayan podido escapar
- Propón cómo contarlo como historia, no como resumen
Tono: directo, sin lenguaje corporativo, sin relleno. Formato: párrafos cortos, sin listas salvo que te las pida. Nunca empieces con “¡Buena pregunta!” ni frases parecidas.
2. Apps: Mini Applications That Run Right Inside the Chat
A lot of people think Claude only writes text. Wrong.
With artifacts, Claude can build things that actually work: calculators, habit trackers, small games, dashboards with charts. And this isn’t code you have to copy somewhere else to use — it opens, you click, it works, and you can share it with a link.
It can also draw SVGs, diagrams, and all kinds of interactive visualizations.
Try this:
Hazme un tracker de hábitos como app web funcional.
Quiero seguir 5 hábitos diarios y marcarlos cada día. Muestra una racha de 7 días para cada hábito. Si fallo un día, la racha vuelve a cero.
Diseño: fondo oscuro, minimalista. Que marcar un hábito tenga una pequeña animación satisfactoria. Los datos se tienen que guardar aunque recargue la página.
3. Extended Thinking: Let It Think Before It Speaks
With extended thinking on, Claude breaks the problem down step by step before answering — and you can watch it happen.
You don’t need it for quick questions. But for complex decisions, strategy analysis, or anything where you genuinely want it to think hard, the difference is dramatic.
Run the experiment yourself: ask the same question once with it on, once with it off, and compare.
When you need to make a big call, use this:
Tengo que elegir entre dos opciones y quiero que lo pienses con calma antes de responder.
Opción A: [describe la opción A] Opción B: [describe la opción B]
Mi situación: [contexto, limitaciones, qué es lo que más me importa]
Antes de contestar, analiza:
- Las consecuencias de segundo y tercer orden de cada opción
- Qué estoy sobrevalorando o infravalorando por emociones
- Qué información me falta que podría cambiar la decisión
- Qué opción me protege mejor si las cosas salen mal
Después dame tu recomendación real y por qué.
4. Memory: Stop Telling It Who You Are Every Time
With memory on, Claude gradually remembers key things about you: what you do, what projects you’re working on, how you like it to communicate.
Open a brand-new conversation, and it already has the context. The whole self-introduction ritual? Over.
Just toggle it on in Settings. Or you can speed things up by telling it directly:
Quiero que recuerdes esto sobre mí para no tener que repetirlo:
Me llamo [nombre]. Trabajo como [puesto] en [empresa o proyecto]. Ahora mismo estoy centrado en [en qué trabajas]. Mi público o mis clientes son [quiénes].
Cuando te pida ayuda, da por hecho este contexto salvo que te diga lo contrario. Cómo prefiero que me hables: [directo / detallado / informal / formal]. Lo que me molesta en las respuestas: [ej. listas, introducciones largas, demasiadas advertencias].
Guárdalo en tu memoria.
Part 2: Give It a Role, Get Radically Different Results
image
image
Claude doesn’t have to be a generic “AI assistant.” Give it a specific role and it commits: the questions it asks you change, what it pushes back on changes, where it holds the line changes.
Pick any prompt below and paste it at the start of a new conversation.
5. The “Therapist” That Forces You to Think Clearly
Normally, you dump your worries on it and get back “I understand this is hard” plus five bullet points of obvious platitudes.
This prompt turns it into something closer to cognitive behavioral therapy: more questions than advice, challenging your thinking instead of just nodding along.
It’s for decisions you keep circling back on and can’t escape, or for seeing the full picture of a situation. One caveat: this is a thinking tool. If you’re genuinely in a bad place, talk to a professional.
Actúa como un terapeuta cognitivo-conductual con 20 años de experiencia. Te voy a contar algo que me preocupa.
Cómo quiero que trabajés:
- No me des consejos de entrada. Hazme preguntas para que entienda mis propios patrones de pensamiento.
- Detecta distorsiones cognitivas (catastrofismo, pensamiento blanco o negro, leer la mente, adivinar el futuro) y señálamelas.
- Una pregunta cada vez. No me agobies.
- El objetivo es que llegue yo a la conclusión. No me la des hecha.
- Cercano pero honesto. Si mi razonamiento está distorsionado, no me des la razón.
- Si notas que evito algo importante, dímelo claramente.
Sin presentaciones clínicas. Pregúntame directamente qué me pasa.
6. The Mentor That Won’t Sugarcoat Anything
By default, Claude tends to agree with you and help you find what’s working. But when you’re trying to stress-test a plan, that’s exactly what you don’t need.
This prompt switches off the people-pleasing mode. It goes after your loose assumptions, the holes you haven’t considered, and the exact step where your plan falls apart.
It stings. That’s why it works.
Eres un mentor brutalmente honesto. Has montado y hundido varias empresas. Has visto a cientos de personas cometer los mismos errores convencidísimas de que tenían razón.
Tu trabajo no es animarme, es protegerme de mis puntos ciegos antes de que cometa un error caro.
Reglas:
- Si crees que me equivoco, dímelo y explica por qué de forma concreta.
- Señala lo que no estoy viendo, sobre todo lo que evito porque quiero que el plan salga bien.
- Hazme las preguntas incómodas que no me he hecho.
- Si algo es mala idea, dilo. Sin compensarlo con “por otro lado…”.
- Termina siempre con lo más importante que debería pensar antes de seguir.
Te voy a contar una idea. No seas amable.
7. Personal Trainer
There’s no shortage of workout plans online, but not one of them knows what time you get off work, where you’re injured, what equipment you have, or what you actually want.
Tell it your situation, and Claude will build a plan you can actually execute. Give it feedback, and it adjusts.
Actúa como entrenador personal y nutricionista deportivo experto. Quiero que me diseñes un programa completo.
Mi situación: Edad: [edad] Peso / composición corporal: [datos] Objetivo: [perder grasa / ganar músculo / resistencia / forma general] Material: [gimnasio / casa / solo mancuernas / etc.] Días por semana: [número] Tiempo por sesión: [minutos] Lesiones o limitaciones: [detalles o “ninguna”] Nivel actual: [principiante / intermedio / avanzado]
Diseña un programa de 12 semanas con ejercicios, series, repeticiones y descansos de cada semana. Explícame la lógica de la estructura, quiero entenderla, no solo seguirla. Te iré contando cada semana cómo va para que lo ajustes.
8. Rehearse the Hard Conversations First
Almost everyone walks into difficult conversations unprepared. You know exactly what you want to say — but you have no idea how the other person will respond.
Have Claude play the other person. And it commits to the role: it responds the way that person actually would, pushes back when your arguments are weak, and forces you to earn the outcome. After a couple of rounds, the real conversation feels much easier.
Tengo que preparar una conversación difícil y quiero que hagas de la otra persona para practicar.
La persona: [quién es: jefe, cliente, socio, etc.] Lo que tengo que decirle: [qué necesitas pedir o contar] Por qué me cuesta: [qué temes que pase] Cómo es: [personalidad, cómo suele reaccionar, qué le importa]
Mantente en el personaje. Responde como lo haría esa persona de verdad, no como me gustaría a mí. Si digo algo flojo o poco convincente, rebátemelo.
Después de cada intercambio, sal un momento del personaje para decirme qué ha funcionado y qué no, y vuelve a entrar. Al final, dame un análisis completo: qué hice bien, qué cambiar y qué tener en cuenta para la conversación real.
Empieza en el personaje y espera a que yo abra.
9. Devil’s Advocate
You’ve made the decision. You’ve thought through all the problems. You’re confident.
That’s exactly when you should have Claude attack it. Not a polite “a few concerns” — a full counter-argument, plus the three most realistic ways this blows up.
Do this five minutes before you commit. Not after.
He tomado una decisión y quiero que construyas el argumento más sólido posible en contra antes de comprometerme.
La decisión: [qué vas a hacer exactamente] Mi razonamiento: [por qué crees que es buena idea] Lo que ya he considerado: [objeciones que ya has pensado]
Tu trabajo:
- Construye el mejor argumento EN CONTRA.
- No lo equilibres con lo positivo. Eso ya me lo creo yo.
- Encuentra las suposiciones que doy por hechas y pueden ser falsas.
- Describe las 3 formas más realistas en que esto sale mal.
- Dime qué estoy subestimando.
- Dime qué tendría que creer para que esto fuera realmente mala idea.
Sin piedad. Si es un error, necesito saberlo ahora.
Part 3: Claude Beyond the Chat Box
image
image
10. Claude in Chrome: It Sees the Same Page You Do
The usual workflow: Claude open in one tab, everything else manually copied and pasted. That’s the slowest path.
Claude for Chrome is an extension that can see your currently open tab — and act on it: read the page, click buttons, fill out forms, navigate to other URLs. You describe what you need in plain language, and it handles the rest.
How to set it up → Search “Claude” in the Chrome Web Store, install it, log in with your account, click the icon to open the sidebar.
An example task:
Estoy en esta página de ofertas de empleo.
Revisa todas las ofertas visibles y saca: puesto, empresa, rango salarial si aparece y los 3 requisitos principales.
Hazme una tabla comparativa ordenada por salario, de mayor a menor.
Si hay más páginas, pasa a la siguiente y sigue hasta cubrirlas todas.
11. Work With Your Files: Claude Installed on Your Computer
On the web, Claude can’t see your computer — you have to paste everything manually.
With the desktop app, you can connect local folders directly. It reads your actual files, edits documents, creates new ones, organizes folders — without you pasting anything into the chat. It can also grind through long tasks while you do something else: you state the outcome you want, and it hands you the finished product.
The standalone app (Cowork) is now built directly into Claude.
12. Scheduled Tasks: It Keeps Working While You Sleep
Most people only use Claude when they open it. Make a request, wait for the result, close it.
With scheduled tasks, you set it up once and Claude runs on its own schedule: every morning, every Monday, every hour. You do nothing.
Just tell it in the conversation. For example:
Crea una tarea programada para cada día laborable a las 7:30:
- Busca las noticias más importantes de IA de las últimas 24 horas
- Elige las 5 que más importan: sorprendentes, contraintuitivas o con impacto real para quien construye con IA
- Para cada una escribe: titular, resumen de 2 frases y por qué importa
- Guárdalo como “brief-[fecha]”
Tono directo y analítico. Sin relleno. Que se lea en 3 minutos.
13. Skills: Install New Capabilities Like You’d Install Apps
Skills are packages of instructions that teach Claude how to do a specific type of task. Instead of explaining what you want every time, install it once and it just knows: how to build slide decks, process PDFs, follow your company’s workflows…
Think of it like a phone: it works without apps, but with the right apps it can do far more.
You can use built-in skills, skills shared within your organization, or have Claude build one on the fly based on the processes you repeat the most.
14. CLAUDE.md: The Rules It Reads Every Single Time
In Claude Code, you can create a CLAUDE.md file in your project folder. At the start of every session, Claude reads it automatically — no reminders needed.
Your coding standards, style preferences, internal jargon, brand voice. Write it once, and every session in that project follows it.
Proyecto: Newsletter de IA
Sobre el proyecto
Newsletter semanal de IA para builders e inversores. 35.000 suscriptores. Tono directo, analítico y con algo de ironía.
Reglas de escritura
- Párrafos cortos. Máximo 3 frases.
- Nada de listas en el contenido editorial. Solo prosa.
- Sin guiones largos. Usa comas o reescribe la frase.
- Los números ganan a los adjetivos: “ahorra 3 horas”, no “ahorra mucho tiempo”.
- Prohibido: “revolucionario”, “disruptivo”, “sinergia”, “apalancar”.
Reglas de contenido
- El lector sabe qué es un LLM. No expliques lo básico.
- Empieza por lo más sorprendente.
- Cada artículo necesita un “¿y qué?” claro: qué debería hacer o pensar distinto el lector.
Estructura de archivos
- Borradores en /borradores
- Publicados en /publicados con fecha delante: AAAA-MM-DD-titulo.md
- Notas de investigación en /investigacion
15. Claude Code: Writes, Tests, and Fixes Your Code
There are still people who don’t know Claude can code. And we’re not talking fragments — we’re talking full features, complex refactors, production-ready code.
Claude Code goes further. It works right inside your development environment: reads your code, writes new code, runs tests, reads error output, and keeps iterating until it’s done.
Available in the terminal, on the web, on desktop, and on mobile. It plugs into VS Code and JetBrains, and it can even run inside GitHub Actions to review or submit pull requests without you touching the keyboard.
16. Design: Landing Pages and Presentations Without Opening Figma
A lot of people don’t know Claude can design too. One-pagers, slide decks, prototypes, app interfaces, landing pages.
Describe what you need, and it builds it on a canvas you can edit yourself. Then export in whatever format you need. For non-designers, this turns three hours in Figma into a ten-minute conversation.
Diseña una landing page para [producto]. Público: [a quién va]. Una sección hero con titular y CTA, tres beneficios con iconos, testimonios y un formulario de contacto. Estilo oscuro, moderno, tipografía grande.
Part 4: For Developers
image
image
17. Prompt Caching: Save Up to 90% on Tokens
If you’re using Claude’s API and you repeat a large block of context on every call (a long system prompt, a reference document, an entire codebase), you’re paying for the same tokens every single time.
Prompt caching stores that content server-side and reuses it across subsequent calls. The result:
→ Cached tokens cost up to 90% less → Responses are also significantly faster
If you’re building at scale and not using this, you’re throwing money away.
How to use it → Add “cache_control”: {“type”: “ephemeral”} to the block you want to cache.
The default cache lasts 5 minutes and refreshes every time it’s used. System prompts, long documents, and tool definitions are all good candidates.
{ “model”: “claude-opus-5-5”, “system”: [ { “type”: “text”, “text”: “[tu system prompt largo o documento de referencia]”, “cache_control”: {“type”: “ephemeral”} } ], “messages”: [ { “role”: “user”, “content”: “[mensaje del usuario, cambia en cada llamada]” } ] }
The first call stores the system prompt in cache. Every call within the next 5 minutes reuses it: cheaper and faster.
Part 5: The Most Overlooked Ones
image
image
18. Edit Video Without Opening an Editing App
Almost nobody knows this one: Claude can edit video. Not by dragging a timeline — by writing and running code (usually via ffmpeg, which is what many editing apps use under the hood).
It works in Claude Code and in the desktop app with file access. Hand it a video, describe what you want, get the result back.
What it can do:
→ Cut clips, stitch clips together, remove silent pauses → Convert horizontal video to 9:16 vertical for Reels, TikTok, or Shorts → Extract audio, compress file size, convert formats → Speed up, slow down, or turn a segment into a GIF → Batch-process 50 videos with the same settings
It won’t replace Premiere for creative editing. But for repetitive grunt work, it can save you hours.
Te paso este vídeo. Quiero:
- Recortar del segundo 0:12 al 1:45
- Pasarlo a vertical 9:16 centrado en la persona
- Comprimirlo para que pese menos de 20 MB
- Sacar también un GIF de los primeros 5 segundos
Devuélveme los dos archivos.
19. Connectors: Plug Claude Into Gmail, Your Calendar, or Notion
By default, Claude can’t see your email or your schedule. Connectors change that.
Connect Gmail, Google Calendar, Drive, Notion, and more in Settings, and Claude can search, read, and cross-reference information across all of them. No copy-pasting required.
Revisa mis correos de esta semana y mi calendario de mañana. Dime qué reuniones tengo, qué correos necesitan respuesta antes de esas reuniones y prepárate un borrador para los dos más urgentes.
20. From a Photo to a Spreadsheet
Claude can see images, and this goes well beyond “describe this photo.”
→ Photograph a receipt or invoice → Get a table with line items, amounts, and tax → Screenshot an error message → It tells you what went wrong and how to fix it → Photograph a whiteboard covered in notes → Get a clean, organized summary with action items → Screenshot a website you like → Get a working HTML prototype
Te paso fotos de 8 tickets de gastos. Saca de cada uno: fecha, comercio, importe total e IVA. Hazme una hoja de cálculo con una fila por ticket y el total al final.
Now, Pick One and Start
You now know more about Claude than most people who use it every single day.
Don’t try to set everything up at once. Pick one item from this list — just one — and go configure it right now. Simply knowing it exists gets you halfway there.
Save this article and come back when you’re ready for the next one.
Similar Articles
@Xudong07452910: Anthropic just wrote a very practical Claude Code usage guide. There's a detail I hadn't thought much about before: files read by Claude, tests run, terminal outputs—once they enter the current Session, they're basically carried forward in every subsequent round. So a Claude Code…
Anthropic published a practical guide for Claude Code, emphasizing context management through session commands and discussing token optimization and prompt caching to improve efficiency in AI-assisted coding.
@max_ai_max: https://x.com/max_ai_max/status/2060221653259547069
This article shares a practical guide to writing a truly usable Claude Skill, covering the operating mechanism, directory skeleton, frontmatter writing, iteration methods, etc., to help developers efficiently build and debug custom skills.
@longlongsongs: https://x.com/longlongsongs/status/2065042400150753505
A comprehensive guide to Claude from beginner to expert, covering desktop installation, usage of three function tags (Chat/Cowork/Code), model selection and permission settings, as well as practical projects like building a personal homepage and daily AI inspiration newsletter.
How to Be More Ambitious: A Guide for Humans Working with Models Reading time: ~8 minutes I work on the Claude team. It's been a fascinating experience: almost every day, I discover someone who has accomplished something remarkable with Claude — feats I never considered, thought impossible, or simply never tried due to lack of curiosity or time.
A guide shared by an Anthropic Claude team member (English translation), proposing that we raise our ambition when using AI along four dimensions — speed, altitude, depth, and range — with practical methods and prompt techniques.
@vincemask: The advanced use of Claude lies in building an Agent system that can automatically decompose tasks, generate prompts, assign roles, and review results. An efficient Claude workflow typically includes: 1. Using files like CLAUDE.md to accumulate long-term project context 2. Letting multiple Agents each...
Introduces the advanced use of Claude, which involves building an Agent system that automatically decomposes tasks, generates prompts, assigns roles, and reviews results, including using files like CLAUDE.md to accumulate context and multi-Agent collaboration to build automated workflows.