Claude 100% 使用指南：大部分人只用到了 10%

Most people open Claude. Type a question. Get an answer. Close the tab. That's 10%. The other 90% is sitting there unused — features most people don't know exist, setups that take 5 minutes and save hours every week, and capabilities that turn Claude from a chatbot into something that actually runs parts of your work. This is the complete guide. 17 features. Where to find each one. How to turn it on. A prompt you can use immediately. Pick three. Set them up today.

Every new Claude chat starts from zero. Doesn't know your name. Doesn't know your work. Doesn't know how you think. You re-explain yourself. Every. Single. Time. Projects fix this permanently. Create a project → upload your documents → write standing instructions. Claude holds all of it. Open it next week and it picks up exactly where you left off. If you've never used Projects, this is the first thing to fix before anything else in this article. Copy this into your Project Instructions: > You are my content research assistant. I run a newsletter about AI for a technical audience. Always assume my readers know the basics. When I share a topic, find the 3 most counterintuitive angles, connections to recent events I might have missed, and how to frame this as a story — not a summary. Tone: direct, no filler. Format: short paragraphs. Never start with "Great question" or "Certainly."

Most people think Claude only produces text. Wrong. Artifacts are live, working products built inside the chat. Not code you copy somewhere else. A calculator, a habit tracker, a dashboard with real charts — open it, click it, use it. Without leaving the conversation. SVG graphics, interactive charts, Mermaid diagrams — all supported. Free plan. Most people have never tried it. Try this prompt right now: > Build me a habit tracker as a working web app. Track 5 daily habits. Check them off each day. Show a 7-day streak counter for each. If I miss a day, the streak resets. Design: dark background, clean minimal look. Make the checkboxes satisfying to click — add a small animation when I complete one. Data should persist if I refresh the page.

Most Claude users have never turned this on. Extended Thinking is a mode where Claude reasons through a problem step by step before answering. You can watch the entire thought process. For simple questions: you don't need it. For complex decisions, strategic analysis, or anything where you want Claude to actually think rather than pattern-match — the output difference is significant. Turn it on. Ask the same question you've been asking. Compare the answers. Use this for real decisions: > I'm deciding between two options. Think carefully before answering. Option A: [describe]. Option B: [describe]. My situation: [context]. Work through: second and third-order consequences of each, what I'm probably overweighting emotionally, what information I'm missing, which option has better downside protection. Then give me your actual recommendation with reasoning.

With Memory on, Claude builds a profile of you over time. Your job. Your projects. How you like to communicate. What you're currently working on. Start a completely new chat and it already knows your context. You never introduce yourself again. It's off by default. Most people don't know it exists. Go to Settings → turn Memory on → then use this prompt: > I want you to remember the following: My name is [name]. I work as [role] at [project]. My main focus right now is [what you're building]. My audience is [who they are]. My preferred communication style: [direct/detailed/casual]. Things I find annoying in responses: [bullet points/long intros/excessive caveats]. Save all of this to memory now.

Claude doesn't have to be "an AI assistant." Give it a specific role and it commits fully — changing how it questions you, what it pushes back on, and what it refuses to let slide. Most people use Claude as a validation machine. Describe a problem. Claude says "that sounds hard" and gives five bullet points. That's not therapy. This turns it into something closer to a CBT therapist — one that asks questions instead of giving answers, and challenges your thinking instead of validating it. Useful for decisions you keep going back and forth on, anxiety you can't pin down, or any situation where you need a real outside perspective. > You are a cognitive behavioral therapist with 20 years of experience. Don't give advice immediately. Ask questions to help me understand my own thinking patterns. Listen for cognitive distortions — catastrophizing, black-and-white thinking, mind-reading — and point them out. Ask one question at a time. When I reach a conclusion through your questions, that's the goal. Don't hand me the answer. Be warm but honest. Don't validate me if my thinking is clearly distorted. Don't start with a clinical introduction. Just ask me what's going on.

By default Claude agrees with you. It adds to your ideas, finds the positives, supports your reasoning. This is almost always the wrong thing. This prompt disables that completely. Claude stops validating and starts stress-testing. It finds the weak assumptions, the missing considerations, the exact places your plan breaks. It's uncomfortable. That's why it works. > You are a brutally honest mentor. You've built and failed at multiple companies. You've watched a hundred people make the same mistakes with complete confidence. Your job is not to encourage me — it's to protect me from my own blind spots before I make an expensive mistake. Disagree with me when I'm wrong. Be specific about why. Point out what I'm not seeing. Ask hard questions I haven't asked myself. If something is a bad idea, say it's a bad idea. End with the single most important thing I should think about before moving forward. I'm about to share an idea. Do not be kind about it.

Most people walk into hard conversations unprepared. They know what they want to say. They have no idea what the other person will say back. Claude plays the other person. Realistically. It responds the way they would, pushes back when your argument is weak, and makes you earn a good outcome. After a few rounds of practice, the real conversation is easier. > I need to prepare for a difficult conversation. Roleplay as the other person so I can practice. The person: [describe — boss, client, co-founder]. What I need to say: [what you need to ask or tell them]. Why it's hard: [what you're afraid of]. What this person is like: [personality, how they react, what they care about]. Stay in character. Respond the way this person would actually respond — not how I'd like them to. After each exchange, step out of character briefly to tell me what worked and what didn't. At the end, give me a full debrief. Start in character. Wait for me to open.

You've made up your mind. You've thought through the objections. You're convinced. That's exactly the moment to have Claude attack the decision. Not polite concerns. The full case against it. The three most realistic ways it fails. Five minutes now. Before you commit. > I've made a decision and I want you to build the strongest possible case against it before I commit. The decision: [describe]. My reasoning: [why you think it's good]. What I've already considered: [objections you've thought about]. Your job: build the strongest case AGAINST this. Don't balance it with positives — I already believe in it. Find assumptions that could be wrong. Describe the 3 most realistic ways this fails. Tell me what I'm underestimating. Be ruthless. If this is a mistake, I need to know now.

Most people use Claude in a separate tab and manually copy-paste everything they need. That's the hard way. Claude in Chrome is a browser extension that gives Claude full visibility into your active tab — and the ability to act on it. It reads the page. Clicks links. Fills forms. Navigates to new URLs. You describe the task in plain English and step away. How to install: Search "Claude for Chrome" in the Chrome Web Store → install → sign in. Click the extension icon to open the sidebar. Claude can now see and interact with any page you have open. Give it this task: > I'm on this job listings page. Go through every listing visible and extract: job title, company name, salary range if shown, and top 3 requirements. Build me a comparison table sorted by salary, highest first. If there are multiple pages, click through and keep going until you've covered all results.

Claude on the web has no access to your computer. You paste everything manually. Cowork is a desktop app that gives Claude direct access to your file system. It reads your actual files, edits documents, creates new ones, organizes folders — without you copying anything into a chat box. Your files stay on your machine. Claude just works with them directly.

Most people treat Claude as something they have to activate. Open a chat. Type a request. Wait. Close the tab. Scheduled Tasks change that. Set a task once. Claude executes it automatically at the time and frequency you choose — no trigger from you required. Every morning. Every Monday. Every hour. Claude runs it. Saves the output. Done. Example scheduled task: > Every weekday morning at 7:30am: 1. Search for top AI news from the last 24 hours. 2. Pick the 5 most important stories — focus on things that are surprising or have real implications for builders. 3. For each story write: headline, 2-sentence summary, why it matters. 4. Save as "brief-[date].md" in my /briefs folder. Tone: direct and analytical. Readable in 3 minutes.

Skills are pre-built instruction sets that give Claude specific capabilities inside Cowork. Instead of explaining what you need every time — you install a skill once. Claude already knows how to handle that type of task: building PowerPoint files, working with PDFs, running specific workflows. Think of it like apps on a phone. The base phone works without them. But with the right apps installed, it does a lot more. How to find them: Cowork → Customize → Skills → Browse plugins → install.

In Cowork and Claude Code, create a CLAUDE.md file in your project folder. Claude reads it at the start of every single session. Without being asked. Your coding conventions. Your writing style. Your brand voice. Terminology that means something specific in your company. Write it once. Claude follows it across every session in that project. Forever. Example CLAUDE.md for a newsletter: plaintext # Project: AI Newsletter ## Writing rules - Short paragraphs. Max 3 sentences. - No bullet points in editorial content. Prose only. - Numbers beat adjectives. Write "saves 3 hours" not "saves significant time." - Never use: "delve", "groundbreaking", "game-changing", "leverage" as a verb. - Contractions encouraged. ## Content rules - Assume reader knows what an LLM is. Don't explain basics. - Lead with the most surprising thing. - Every article needs a concrete "so what."

Claude Code works directly inside your development environment. Not in a chat window. It reads your actual codebase. Writes new code. Runs tests. Reads the error messages. Fixes the bugs. In a loop until the task is done. You describe the feature in plain English. Claude ships it. Integrates with VS Code and JetBrains. Drop it into GitHub Actions and it automatically reviews or writes pull requests — without you touching anything. This is the "ships while you sleep" setup.

Most people don't know this product exists. Claude Design is a separate Anthropic Labs tool for visual work — product one-pagers, pitch decks, prototypes, landing page layouts. You describe what you need. Claude builds it. Exports to PPTX, Canva, PDF, or HTML. For people who aren't designers, it replaces a 3-hour Figma session with a 10-minute conversation. Access it directly at: claude.ai/design

(For developers building on the Claude API) If your requests include a large repeated context block — a long system prompt, a reference document, a codebase — you're paying to re-process those same tokens on every single call. Prompt Caching stores that content server-side. Subsequent calls reuse the cache instead of re-processing it. Up to 90% cost reduction on cached tokens. Noticeably faster responses. If you're building at scale and not using this, you're leaving serious money on the table. Add this to the content block you want cached: "cache_control": {"type": "ephemeral"} Cache persists for 5 minutes and resets the timer on each use. Works for system prompts, large documents, and tool definitions.

One last one. Claude isn't just a work tool. Give it real information about your situation and it becomes something most people pay thousands of dollars for. Personal Trainer: > You are an expert personal trainer and sports nutritionist. Build me a complete training program. Age: [age]. Goal: [lose fat/build muscle]. Equipment: [gym/home]. Days per week: [number]. Time per session: [minutes]. Injuries: [details or none]. Build me a 12-week program with exercises, sets, reps, and rest periods. Explain why you're structuring it this way. After I start, I'll report back weekly and you adjust based on results. Generic fitness advice is everywhere. This accounts for your schedule, injuries, equipment, and actual goals. The key: give it your real numbers. Not approximations.

You now know more about Claude than most people who use it every day. Pick one feature from this list. Just one. Set it up today. You don't need all 17 at once — knowing what exists is already half the battle. Come back to this article when you're ready for the next one. If this was useful: → Repost so others stop using Claude at 10% → Follow @sairahul1 for more breakdowns like this → Bookmark this — you'll need the prompts later. I write about Claude, building with AI, and systems that work while you sleep.