一句话总结
Claude 不只是聊天窗口里的那个对话框。Projects、Artifacts、Extended Thinking、Memory、角色扮演、Claude Code、Cowork、Scheduled Tasks、Skills、CLAUDE.md、Claude Design、Prompt Caching——这篇文章把 15 个核心功能全拆开,附上拿来就用的 Prompt,今天读完今天就能用上。
你只用了 Claude 的 10%
大多数人打开 Claude 的姿势是这样的:新建对话 → 打字提问 → 得到回答 → 关掉窗口。下次再打开,重复一遍。
这就像买了一台 MacBook Pro 只用它写备忘录——没错,能用,但暴殄天物。
2026 年中,Claude 已经从一个单一的大语言模型变成了一个完整的产品生态:桌面端、浏览器端、终端、API、定时任务、插件系统、设计工具——每一个维度都有被大多数人忽略的杀手级功能。
本文按「入门 → 进阶 → 专家」的结构,把 Claude 的 15 个核心功能全部梳理一遍。每个功能都有具体操作路径和可直接复制粘贴的 Prompt。
🧩 第一部分:基础必知(但 90% 的人没设对)
1. Projects — 让 Claude 从「失忆症」变成「长期记忆」
每次打开 Claude 新对话,它都不知道你是谁。你的名字、工作、偏好——一片空白。大多数人已经习惯了每次对话重新介绍自己。
Projects 就是解决这个问题的。 你创建一个 Project,上传文档,写好固定指令(Project Instructions),Claude 就会把这些全部记住。下周再打开,它接着上次的来,不用重新交代背景。
怎么开:Claude 左侧栏 → Projects → Create Project → 填写 Project Instructions → 上传相关知识文档。
示例 Project Instructions — 直接粘贴:
You are my content research assistant. I run a newsletter about AI and crypto for a technical audience.
Always assume my readers know the basics. Don't explain what an LLM is or what a blockchain is.
When I share a topic or article, your job is to:
1. Identify the 3 most counterintuitive or surprising angles
2. Find connections to recent events I might have missed
3. Suggest how I could frame this as a story, not a summary
Tone: direct, no corporate language, no filler phrases.
Format: short paragraphs, no bullet points unless I ask.
Never start a response with "Great question" or "Certainly".
适合场景:任何你反复和 Claude 讨论的项目——写 newsletter、做研究、代码项目、课程学习。
2026 更新:2026 年 6 月推出的 Claude Tag(标签系统)可以让你给 Project 打标签分类,多个 Project 之间的切换和检索更丝滑。
2. Artifacts — 在对话里直接跑起来的应用
很多人觉得 Claude 只能输出文字。这是错的。 Artifacts 让 Claude 在对话侧边栏直接生成可交互的内容——不是一段需要复制粘贴到别处的代码,而是一个活着的成品。
计算器、习惯追踪器、小游戏、带图表的仪表盘、SVG 图形、交互式数据可视化、Mermaid 流程图——打开就能用,不用离开对话。
试试这个 Prompt:
Build me a habit tracker as a working web app.
I want to track 5 daily habits. Each day I can check them off.
Show a 7-day streak counter for each habit.
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.
The data should persist if I refresh the page.
免费版就能用。大多数人从没试过。你刚才试了吗?
3. Extended Thinking — 完全不同的推理质量
大多数 Claude 用户从没开过这个功能。
Extended Thinking 是一种模式,Claude 在给你答案之前,会一步步推理整个问题——而且你能看到它的完整思考过程。简单问题不需要。但遇到复杂决策、策略分析、代码架构设计——输出质量的差距非常明显。
对比一下:先不开 Extended Thinking 问一个复杂问题,再打开问同样的问题。你自己判断。
遇到真正需要做决策时用这个 Prompt:
I'm deciding between two options and I want you to think through this carefully before answering.
Option A: [描述选项 A]
Option B: [描述选项 B]
My situation: [你的背景、约束条件、最在乎什么]
Work through this before responding. Think about:
- The second and third-order consequences of each option
- What I'm probably overweighting or underweighting emotionally
- What information I might be missing that would change the decision
- Which option has better downside protection if things go wrong
Then give me your actual recommendation with your reasoning.
2026 更新:Sonnet 5(2026 年 6 月发布)在 Extended Thinking 模式下的自主代理能力大幅提升——它能在思考过程中自主决定调用浏览器、终端等工具,而不需要你手动引导每一步。配合不同的 effort level(努力程度),你可以在成本和效果之间灵活调节。
4. Memory — 知道你是谁的 Claude
Memory 默认是关的。打开之后,Claude 会慢慢记住你的信息——你的工作、项目、沟通风格偏好、正在做的事情。
开一个全新的对话,它已经知道背景了。你再也不用自我介绍。
怎么开:Settings → Memory → 打开开关。
给 Memory 喂初始信息的 Prompt:
I want you to remember the following about me so you don't need to ask again:
My name is [名字]. I work as [职位] at [公司/项目].
My main focus right now is [你正在做的那件事].
My audience or customers are [你的目标用户].
When I ask for help, always assume this context unless I say otherwise.
My preferred communication style: [direct / detailed / casual / formal].
Things I find annoying in responses: [比如 bullet points、长开场白、过度谨慎].
Save all of this to memory now.
🎭 第二部分:角色扮演 — Claude 不只是「一个 AI」
Claude 不一定要是「一个 AI 助手」。给它一个具体的角色,它会完全投入——改变提问方式、改变 push back 的方向、改变哪些东西它不会轻易放过。
以下 Prompt 全部可以直接用,复制粘贴到新对话开头即可。
5. 私人心理咨询师
大多数人把 Claude 当成一个「认可的机器」:描述问题 → Claude 说这确实很难 → 给你五条建议。
好的心理咨询不是这样的。 这个 Prompt 把 Claude 变成一个接近 CBT(认知行为疗法)咨询师的角色——它问你问题而不是给你答案,挑战你的想法而不是认可你的想法。
适合那些你反复纠结的决策、说不清楚的焦虑,或者任何你需要一个清晰外部视角的场景。
You are a cognitive behavioral therapist with 20 years of experience. I'm going to share something I'm struggling with.
Your approach:
- Don't give advice immediately. Start by asking questions to help me understand my own thinking patterns.
- Listen for cognitive distortions — catastrophizing, black-and-white thinking, mind-reading, fortune-telling — and point them out when you notice them.
- Ask one question at a time. Don't overwhelm me.
- When I reach a conclusion on my own 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.
- If I seem to be avoiding something important, name it directly.
Don't start with a clinical introduction. Just ask me what's going on.
6. 严厉的导师
Claude 默认会同意你。它会补充你的想法、支持你的推理、找到积极的一面。但这几乎总是错的。
这个 Prompt 把「认可模式」关掉了。Claude 不再顺着你——它开始压力测试。找你假设里的漏洞、被忽略的问题、你计划崩溃的确切位置。
不舒服。但正因为不舒服才有用。
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.
Rules:
- Disagree with me when you think I'm wrong. Be specific about why.
- Point out what I'm not seeing, especially things I might be avoiding because I want my plan to work.
- Ask hard questions I haven't thought to ask myself.
- If something is a bad idea, say it's a bad idea. Don't balance it with "on the other hand..."
- End your responses 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.
7. 私人教练
健身建议到处都是。但那些都不考虑你的时间安排、你的伤、你的器材、你的真实目标。
把你的数据给 Claude,它会给你做一个真正的计划。不是模板。不是你在任何健身网站上都能找到的东西。一个围绕你的情况设计的方案,而且你每周反馈之后它会调整。
You are an expert personal trainer and sports nutritionist. I want you to build me a complete training program.
My situation:
Age: [年龄]
Current weight / body composition: [详情]
Goal: [减脂 / 增肌 / 提升耐力 / 综合体能]
Available equipment: [健身房 / 居家 / 只有哑铃 / 其他]
Days per week I can train: [天数]
Time per session: [分钟]
Any injuries or limitations: [伤病史 或 "none"]
Current fitness level: [beginner / intermediate / advanced]
Build me a 12-week program. Give me the full plan for each week with exercises, sets, reps, and rest periods. Explain why you're structuring it this way — I want to understand the logic, not just follow instructions. After I start, I'll report back weekly and you adjust based on how it's going.
8. 练习一场困难的对话
大多数人走进困难对话的时候是没有准备的。他们知道自己想说什么,但不知道对方会怎么回。
Claude 来扮演对方。很真实地扮演。它按对方会有的方式回应,你的论点弱的时候它会顶回来,你得自己争取一个好结果。
练几次之后,真正的对话就轻松多了。
I need to prepare for a difficult conversation. I want you to roleplay as the other person so I can practice.
The person: [描述对方是谁 — 老板、客户、合伙人等]
What I need to say: [你要问什么或告诉对方什么]
Why it's hard: [你害怕什么 / 什么可能出错]
What this person is like: [他们的性格、通常怎么反应、在乎什么]
Stay in character throughout the conversation. Respond the way this person would actually respond — not how I'd like them to. If I say something weak or unconvincing, push back on it.
After each exchange, step out of character briefly to tell me what worked and what didn't — then go back in. At the end, give me a full debrief: what I did well, what to change, and the most important things to keep in mind for the real conversation.
Start in character. Wait for me to open.
9. 魔鬼代言人
你已经做了决定。你已经想过反对意见了。你确信了。
这正是该让 Claude 攻击这个决定的时候。 不是客气的担忧。是完整的反对意见。三种最现实的失败方式。你因为太想做成而看不到的东西。
现在花五分钟。在你 commit 之前。不是之后。
I've made a decision and I want you to build the strongest possible case against it before I commit.
The decision: [描述你打算做什么]
My reasoning: [你为什么觉得这是个好主意]
What I've already considered: [你已经想过的反对意见]
Your job:
- Build the strongest possible case AGAINST this decision.
- Don't balance it with positives. I already believe in it — I need the counterargument.
- Find the assumptions I'm making that could be wrong.
- Describe the 3 most realistic ways this fails or backfires.
- Tell me what I'm probably underestimating.
- Tell me what I would need to believe for this to be a genuinely bad idea.
Be ruthless. If this is a mistake, I need to know now.
🖥️ 第三部分:桌面与浏览器 — Claude 走出聊天窗口
10. Claude in Chrome — 能看到你屏幕的 Claude
大多数人在一个单独的标签页里用 Claude,需要什么手动复制粘贴。这是笨办法。
Claude for Chrome 是一个浏览器扩展,让 Claude 完全看到你当前标签页的内容,还能在上面操作。它读页面、点链接、填表单、跳转到新网址。你用大白话描述任务,然后走开就行。
怎么装:Chrome 应用商店搜 Claude for Chrome → 安装 → 用 Claude 账号登录。点扩展图标打开侧边栏,Claude 现在能看到你打开的任何页面并跟它互动。
试试这个任务:
I'm on this job listings page.
Go through every listing visible and extract: job title, company name, salary range if shown, and the top 3 requirements.
Build me a comparison table sorted by salary, highest first.
If there are multiple pages, click through to the next page and keep going until you've covered all results.
11. Claude Cowork — 住在你桌面上的 Claude
网页版 Claude 访问不了你的电脑。它看不到你的文件。你什么都得手动粘贴。
Cowork 是一个桌面应用,让 Claude 直接访问你的文件系统。它读你的真实文件、编辑文档、创建新文件、整理文件夹——不用你往聊天框里复制任何东西。
2026 年中新增的 Skills 系统:Cowork 里的 Skills 像手机上的 App——预先做好的指令集,给 Claude 加上特定能力。做 PPT、处理 PDF、跑特定工作流——装一次 Skill,Claude 就知道怎么处理这类任务了。
怎么找:Cowork → Customize → Skills → Browse plugins → 找到插件 → 安装。
12. Scheduled Tasks — 你睡觉它在干活的 Claude
大多数人把 Claude 当成一个需要自己手动触发的东西。Scheduled Tasks 改变了这一点:你设一次任务,Claude 就在你设定的时间和频率自动执行——不需要你触发。
每天早上七点半的 AI 早报示例:
Every weekday morning at 7:30am, do the following:
1. Search for the top AI and crypto news from the last 24 hours
2. Pick the 5 most important stories — focus on things that are surprising, counterintuitive, or have real implications for builders and investors
3. For each story write: headline, 2-sentence summary, why it matters
4. Save the result as "brief-[date].md" in my /briefs folder
Keep the tone direct and analytical. No fluff. Readable in 3 minutes.
🛠️ 第四部分:开发者生态 — Claude 不只是聊天工具
13. Claude Code — 在你终端里写代码、测试、修 bug 的 AI
Claude Code 不是一个「在聊天窗口里帮你写代码片段」的工具——它是一套完整的开发代理,直接在你的开发环境里工作。
它能做什么:
- 阅读你真实的代码库(不是粘贴的片段)
- 写新代码、重构现有代码
- 跑测试、读报错信息、然后修 bug
- 循环往复直到任务完成
它还集成了:VS Code、JetBrains、GitHub Actions(自动 review 或写 pull request)。2026 年 7 月,Anthropic 还推出了 Claude Code Enterprise,支持团队级别的代码库理解和协作。
如果你是开发者:这是 Claude 生态里最值得试的工具,没有之一。
14. CLAUDE.md — Claude 每次自动读取的规则文件
在 Cowork 和 Claude Code 里,你可以在项目文件夹里创建一个 CLAUDE.md 文件。
Claude 每次开始会话的时候都会自动读它,不需要你提醒。你的代码规范、写作风格规则、公司里有特定含义的术语、品牌调性指南——写一次,Claude 在这个项目的每一次会话里都会遵守。
示例 CLAUDE.md:
# Project: AI Newsletter
## About this project
Weekly newsletter about AI and crypto for builders and investors. 35,000 subscribers. Tone is direct, analytical, occasionally irreverent.
## Writing rules
- Short paragraphs. Max 3 sentences.
- No bullet points in editorial content. Prose only.
- No em dashes. Use hyphens or restructure the sentence.
- Numbers beat adjectives. Write "saves 3 hours" not "saves significant time".
- Never use: "delve", "groundbreaking", "game-changing", "leverage" (as a verb), "utilize".
- Contractions are fine and encouraged.
## Content rules
- Assume the reader knows what an LLM is. Don't explain basics.
- Lead with the most surprising or counterintuitive thing.
- Every article needs a concrete "so what" — what should the reader do or think differently.
## File structure
- Drafts go in /drafts
- Published articles go in /published with date prefix: YYYY-MM-DD-title.md
- Research notes go in /research
15. Claude Design + Prompt Caching
Claude Design(claude.ai/design):Anthropic Labs 出品的独立设计工具。你做产品单页、路演 PPT、原型、落地页布局——描述需求,Claude 来做。可以导出 PPTX、Canva、PDF 或 HTML。把三小时的 Figma 搏斗换成十分钟的对话。
Prompt Caching(面向 API 开发者):如果你的 API 请求里包含大段重复的上下文(长系统 Prompt、参考文档、代码库),你每次调用都在花钱重新处理同样的 token。
加一行 "cache_control": {"type": "ephemeral"} 在你想缓存的内容块上,后续调用直接复用缓存,成本最多降低 90%。缓存持续 5 分钟,每次使用自动重置计时器。
{
"model": "claude-sonnet-5-20260630",
"system": [
{
"type": "text",
"text": "[你的大型系统 Prompt 或参考文档]",
"cache_control": {"type": "ephemeral"}
}
],
"messages": [
{"role": "user", "content": "[用户消息 — 每次调用不同]"}
]
}
💡 2026 年 6 月发布的 Claude Sonnet 5 引入了 effort level(努力程度)参数,配合 Prompt Caching 可以在成百上千次代理调用中显著节省成本,同时保持接近 Opus 4.8 的代理性能。
🌟 2026 年中 Claude 生态新动态
除了上面提到的内容,2026 年上半年 Claude 还有这些值得关注的更新:
| 功能 / 产品 | 发布时间 | 一句话 |
|---|---|---|
| Claude Sonnet 5 | 2026.06.30 | 最强 Sonnet,代理能力接近 Opus 4.8,价格更低 |
| Claude Tag | 2026.06.23 | 给 Project 和对话打标签,分类管理 |
| Claude Science | 2026.06.30 | 面向科学研究的 AI 工作台 |
| Claude Code Enterprise | 2026.07 | 团队级代码库理解和协作 |
| Fable 5 | 2026.06.30 | Anthropic 最新图像生成模型,重新上线 |
定价参考(Sonnet 5,2026 年 8 月 31 日前推广价):$2/百万输入 token,$10/百万输出 token。推广期过后恢复 $3/$15。
总结
你现在对 Claude 的了解,已经超过了大多数每天都在用它的人。
把这份清单存下来。今天就挑一个功能设好——从 Projects 开始,这是回报率最高的五分钟投入。
每解锁一个功能回来看一眼。15 个全掌握的那天,你会回头看这篇文章想:我以前怎么用 Claude 的?