章节 1：引言与两种用户的巨大差距

📝 本节摘要：

本节指出了大多数普通用户和专业用户在使用Claude时的根本差异：普通用户仅将其视作一个更聪明的搜索引擎，并在每次对话中从零开始；而专业用户则将其深度集成到业务运营的底层逻辑中。这种差距不在于智商，而在于对生产环境中高级功能的认知与运用。

[原文] [Author]: Most Claude users have never left the chat box.

[译文] [Author]: 大多数Claude用户从未离开过聊天框。

[原文] [Author]: They open a tab, type a message, copy the response, close the tab.

[译文] [Author]: 他们打开一个标签页，输入一条消息，复制回复内容，然后关闭标签页。

[原文] [Author]: Tomorrow they do it again.

[译文] [Author]: 明天他们还会再重复一次。

[原文] [Author]: Same blank context window.

[译文] [Author]: 同样的空白上下文窗口（context window）。

[原文] [Author]: Same from-scratch explanation.

[译文] [Author]: 同样的从头开始解释。

[原文] [Author]: Same 10% of the tool.

[译文] [Author]: 同样只用到了这个工具的10%。

[原文] [Author]: I've spent the last year building AI automation infrastructure for enterprise clients.

[译文] [Author]: 过去一年里，我一直在为企业客户构建AI自动化基础设施（AI automation infrastructure）。

[原文] [Author]: Outbound SDR pipelines, business intelligence systems, multi-agent orchestration layers.

[译文] [Author]: 包括外呼销售开发代表（SDR）管道、商业智能系统（business intelligence systems）以及多智能体编排层（multi-agent orchestration layers）。

[原文] [Author]: In that time I've watched two types of people use Claude.

[译文] [Author]: 在这段时间里，我观察到了两类使用Claude的人。

[原文] [Author]: The first type uses it like a smarter Google.

[译文] [Author]: 第一类人把它当作一个更聪明的Google来用。

[原文] [Author]: The second type has wired it directly into how their business operates.

[译文] [Author]: 第二类人则将其直接与他们的业务运营方式连接（wired）在了一起。

[原文] [Author]: The gap between those two groups isn't intelligence.

[译文] [Author]: 这两个群体之间的差距并不在于智力。

[原文] [Author]: It's knowing which features exist and what they actually do in production.

[译文] [Author]: 而在于了解存在哪些功能，以及它们在生产环境（production）中究竟能发挥什么作用。

[原文] [Author]: Here are the 12 features separating them.

[译文] [Author]: 以下便是区分他们的12个核心功能。

章节 2：功能 1 - 项目（Projects）

📝 本节摘要：

介绍了Claude中最容易被忽视的基础功能——“项目”。与默认聊天的零记忆不同，“项目”提供了持久化的上下文环境，让AI能够自动继承特定的指令、记忆和文件背景，它是使用其它所有进阶功能的地基。

[原文] [Author]: FEATURE 1: PROJECTS

[译文] [Author]: 功能 1：项目（PROJECTS）

[原文] [Author]: The single most underused feature in Claude's entire interface.

[译文] [Author]: 这是Claude整个界面中最被低估且未被充分利用的单一功能。

[原文] [Author]: Most people use Claude in the default chat.

[译文] [Author]: 大多数人都在默认聊天中使用Claude。

[原文] [Author]: One conversation at a time.

[译文] [Author]: 每次进行一次对话。

[原文] [Author]: No memory, no context, no continuity.

[译文] [Author]: 没有记忆，没有上下文（context），没有连续性。

[原文] [Author]: Every session starts from zero.

[译文] [Author]: 每一个会话（session）都从零开始。

[原文] [Author]: You re-explain who you are.

[译文] [Author]: 你得重新解释你是谁。

[原文] [Author]: You re-explain what you're building.

[译文] [Author]: 你得重新解释你在构建什么。

[原文] [Author]: You re-paste the document you pasted yesterday.

[译文] [Author]: 你得重新粘贴昨天刚刚粘贴过的文档。

[原文] [Author]: Projects are persistent environments that sit above conversations.

[译文] [Author]: 项目是位于对话之上的持久化环境（persistent environments）。

[原文] [Author]: Each project has its own instruction set, its own memory, its own attached files, and every conversation you start inside that project inherits all of it automatically.

[译文] [Author]: 每个项目都有其专属的指令集（instruction set）、专属记忆、专属附加文件，且你在该项目中开启的每一次对话都会自动继承所有这些内容。

[原文] [Author]: You stop explaining.

[译文] [Author]: 你不用再作解释了。

[原文] [Author]: Claude already knows.

[译文] [Author]: Claude已经知道了。

[原文] [Author]: Here's what this looks like in practice:

[译文] [Author]: 在实践中，它是这样的：

[原文] [Author]: I have a project for every client.

[译文] [Author]: 我为每一位客户都建立了一个项目。

[原文] [Author]: The project instructions contain who they are, what we're building, their ICP, their communication style preferences, and any constraints I need Claude to respect.

[译文] [Author]: 项目指令包含他们是谁、我们在构建什么、他们的理想客户画像（ICP）、他们的沟通风格偏好，以及任何我需要Claude遵守的约束条件。

[原文] [Author]: Their brand doc and relevant data sit as attached files inside the project.

[译文] [Author]: 他们的品牌文档和相关数据作为附加文件存放在项目内。

[原文] [Author]: When I open a new conversation to work on that client's campaign, I don't paste anything.

[译文] [Author]: 当我为了处理该客户的营销活动而开启新的对话时，我什么都不用粘贴。

[原文] [Author]: I just start.

[译文] [Author]: 我直接开始工作。

[原文] [Author]: Claude reads the context before I type a word.

[译文] [Author]: 在我输入第一个字之前，Claude就已经读取了上下文。

[原文] [Author]: Each project comes with a 200K context window, equivalent to a 500-page book.

[译文] [Author]: 每个项目都自带一个200K的上下文窗口，相当于一本500页的书。

[原文] [Author]: A project for client communication.

[译文] [Author]: 一个用于客户沟通的项目。

[原文] [Author]: A project for code architecture.

[译文] [Author]: 一个用于代码架构的项目。

[原文] [Author]: A project for research and synthesis.

[译文] [Author]: 一个用于研究和综合分析的项目。

[原文] [Author]: A project for content.

[译文] [Author]: 一个用于内容创作的项目。

[原文] [Author]: Each one is a different employee with different context, different instructions, different files, different expertise.

[译文] [Author]: 每一个项目都像是一个拥有不同上下文、不同指令、不同文件和不同专业知识的不同员工。

[原文] [Author]: This isn't a nice-to-have.

[译文] [Author]: 这不是一个可有可无的功能。

[原文] [Author]: It's the foundation everything else in this list is built on.

[译文] [Author]: 它是这个列表里其他所有功能赖以建立的基础。

[原文] [Author]: If you're not using Projects, you're not using Claude.

[译文] [Author]: 如果你没有在使用项目（Projects），你就没有在使用Claude。

[原文] [Author]: You're using a text box.

[译文] [Author]: 你只是在使用一个文本框。

章节 3：功能 2 - 技能（Skills）

📝 本节摘要：

阐述了“技能”功能的核心价值：将高频重复的任务及其特定逻辑结构化并封装，使得Claude可以像调用函数一样随时执行固定标准的操作，从而大幅减少重复输入提示词的工作量。

[原文] [Author]: FEATURE 2: SKILLS (skills.sh / find skill)

[译文] [Author]: 功能 2：技能（SKILLS）（涵盖 skills.sh / find skill）

[原文] [Author]: Projects give Claude context.

[译文] [Author]: 项目赋予了Claude上下文。

[原文] [Author]: Skills give Claude repeatable capabilities.

[译文] [Author]: 技能（Skills）则赋予了Claude可重复的能力。

[原文] [Author]: Skills are structured instruction sets loaded into Claude Code to give it domain-specific, task-specific expertise.

[译文] [Author]: 技能是被加载到Claude Code中的结构化指令集，用于赋予其特定领域、特定任务的专业知识。

[原文] [Author]: Think of each skill as a specialized prompt and workflow that you encode once and deploy indefinitely.

[译文] [Author]: 可以把每项技能看作是一个专门的提示词（prompt）和工作流（workflow），你只需编码一次，就可以无限期部署调用。

[原文] [Author]: The skills system lives inside Claude Code's configuration layer, accessible from your project's .claude/ directory.

[译文] [Author]: 技能系统位于Claude Code的配置层内，可以从你项目的 .claude/ 目录中访问。

[原文] [Author]: The pattern works like this: you define a skill for a specific, repeatable task.

[译文] [Author]: 其运作模式是这样的：你为一项特定的、可重复的任务定义一项技能。

[原文] [Author]: Writing cold outreach in your exact style and format.

[译文] [Author]: 比如以你精确的风格和格式撰写冷拓客邮件（cold outreach）。

[原文] [Author]: Scoring leads against your ICP criteria.

[译文] [Author]: 或者对照你的ICP（理想客户画像）标准对线索（leads）进行评分。

[原文] [Author]: Summarizing transcripts into structured call notes.

[译文] [Author]: 或者将通话转录摘要为结构化的通话笔记。

[原文] [Author]: Processing raw research into a formatted deliverable.

[译文] [Author]: 亦或是将原始研究处理成格式化的交付成果。

[原文] [Author]: Each skill has a name, a defined input, and a defined output format.

[译文] [Author]: 每个技能都有一个名称、一个预定义的输入，以及一个预定义的输出格式。

[原文] [Author]: Once a skill exists in your library, you deploy it.

[译文] [Author]: 一旦某项技能存在于你的库中，你就可以部署它。

[原文] [Author]: The same scoring logic you used on a 200-lead list last month gets applied to the next 200 leads next month without re-explaining how scoring works.

[译文] [Author]: 你上个月用于200个潜在客户名单的相同评分逻辑，下个月就能直接应用于下一批200个潜在客户，而无需重新解释评分机制是如何运作的。

[原文] [Author]: The prompt is already baked in.

[译文] [Author]: 因为提示词（prompt）已经被内嵌（baked in）其中了。

[原文] [Author]: Within a Claude Code session, you can run /skills to list every available skill or use find skill to search your library by name.

[译文] [Author]: 在Claude Code会话中，你可以运行 /skills 来列出所有可用的技能，或者使用 find skill 通过名称来搜索你的技能库。

[原文] [Author]: Operators who've built libraries of 10, 20, 50 skills are not doing more work than everyone else.

[译文] [Author]: 那些已经构建了10个、20个甚至50个技能库的专业操作者（Operators），并没有比其他人做更多的工作。

[原文] [Author]: They've encoded their processes once.

[译文] [Author]: 他们只是把自己的流程进行了一次编码。

[原文] [Author]: Now they just call the function.

[译文] [Author]: 现在他们只需要调用该函数即可。

章节 4：功能 3 - CLAUDE.MD（Claude 初始化文档）

📝 本节摘要：

强调了CLAUDE.MD文件作为AI开发者“入职文档”的关键作用。通过预先规范项目的技术栈、代码风格及限制，AI能够立刻像资深工程师一样精准理解代码库，避免在后续对话中盲目猜测或犯错。

[原文] [Author]: FEATURE 3: CLAUDE.MD (claude init)

[译文] [Author]: 功能 3：CLAUDE.MD（包含 claude init 初始化命令）

[原文] [Author]: If you're building anything with Claude Code, this is the first thing you configure.

[译文] [Author]: 如果你正在用Claude Code构建任何东西，这是你首先要配置的内容。

[原文] [Author]: If you haven't set it up, Claude is operating blind in your codebase.

[译文] [Author]: 如果你还没有设置它，Claude在你的代码库中就是盲目操作的。

[原文] [Author]: CLAUDE.MD is a markdown file that lives at the root of your repository.

[译文] [Author]: CLAUDE.MD 是一个存放在你代码仓库根目录下的markdown文件。

[原文] [Author]: When Claude Code opens a project, it reads this file before it reads anything else.

[译文] [Author]: 当Claude Code打开一个项目时，它会在读取其他任何内容之前先读取这个文件。

[原文] [Author]: It's the onboarding document for your AI developer, and you write it once, not every session.

[译文] [Author]: 它是你的AI开发者的入职文档（onboarding document），并且你只需编写一次，而不是在每个会话中都要写。

[原文] [Author]: Here's what goes in it:

[译文] [Author]: 以下是该文件应包含的内容：

[原文] [Author]: → What this project is and what it does

[译文] [Author]: → 这个项目是什么，以及它是做什么的

[原文] [Author]: → The tech stack and architectural decisions and why

[译文] [Author]: → 技术栈（tech stack）、架构决策及其原因

[原文] [Author]: → How to run the project locally

[译文] [Author]: → 如何在本地运行该项目

[原文] [Author]: → Code style rules and conventions

[译文] [Author]: → 代码风格规则和约定

[原文] [Author]: → Which files and folders are off-limits

[译文] [Author]: → 哪些文件和文件夹是禁止访问的（off-limits）

[原文] [Author]: → What patterns to use and what to avoid

[译文] [Author]: → 应该使用哪些设计模式，以及应该避免什么

[原文] [Author]: → Known issues or constraints

[译文] [Author]: → 已知问题或约束条件

[原文] [Author]: → Where to find tests and how to run them

[译文] [Author]: → 在哪里找到测试文件，以及如何运行它们

[原文] [Author]: Without CLAUDE.MD, Claude Code makes educated guesses about your architecture every session.

[译文] [Author]: 没有CLAUDE.MD，Claude Code在每一个会话中只能对你的架构进行有根据的猜测。

[原文] [Author]: With it, Claude Code knows the system the way a senior engineer who's been in the codebase for six months knows it.

[译文] [Author]: 有了它，Claude Code就能像一个在代码库中待了六个月的资深工程师一样去了解这个系统。

[原文] [Author]: Before you run a single command.

[译文] [Author]: 在你运行哪怕第一条命令之前它就已经知道了。

[原文] [Author]: The difference in output quality is immediate.

[译文] [Author]: 输出质量上的差异是立竿见影的。

[原文] [Author]: Claude stops making the wrong kind of changes.

[译文] [Author]: Claude不再做出错误类型的更改。

[原文] [Author]: It stops touching things it shouldn't.

[译文] [Author]: 它不再触碰不该触碰的东西。

[原文] [Author]: It stops suggesting patterns that don't fit your stack.

[译文] [Author]: 它不再建议不适合你技术栈的模式。

[原文] [Author]: Because you told it the rules upfront, and it read them.

[译文] [Author]: 因为你预先告诉了它规则，而它也读取了这些规则。

章节 5：功能 4 - Claude Code（终端智能体）

📝 本节摘要：

澄清了“Claude Code”并非简单的代码补全工具，而是一个具备端到端执行能力的终端代理。它能接管复杂业务需求的全链路：从阅读代码、实施修改，到修复测试和提交，将AI的职责从“写函数”直接跃升为“交付特性”。

[原文] [Author]: FEATURE 4: CLAUDE CODE

[译文] [Author]: 功能 4：CLAUDE CODE

[原文] [Author]: Claude Code is not Claude in a code editor.

[译文] [Author]: Claude Code并不是代码编辑器里的Claude。

[原文] [Author]: That's the version most people imagine.

[译文] [Author]: 那只是大多数人想象中的版本。

[原文] [Author]: Claude Code is a terminal-based agent that takes a task and executes it end-to-end, reading files, writing code, running commands, checking errors, iterating on failures, installing packages, running tests, and committing changes.

[译文] [Author]: Claude Code是一个基于终端的智能体（terminal-based agent），它接受任务并进行端到端执行：阅读文件、编写代码、运行命令、检查错误、在失败时迭代、安装依赖包、运行测试以及提交更改。

[原文] [Author]: You give it a goal.

[译文] [Author]: 你给它一个目标。

[原文] [Author]: It figures out how to get there.

[译文] [Author]: 它会自己想办法如何达成目标。

[原文] [Author]: The mental model shift is critical.

[译文] [Author]: 这种心智模型（mental model）的转变至关重要。

[原文] [Author]: You're not using Claude Code to write a function.

[译文] [Author]: 你不是在使用Claude Code来编写一个函数。

[原文] [Author]: You're using Claude Code to ship a feature.

[译文] [Author]: 你是在使用Claude Code来交付一个功能特性（ship a feature）。

[原文] [Author]: The difference is the scope of delegation.

[译文] [Author]: 区别在于委托授权的范围。

[原文] [Author]: "Add rate limiting to the API endpoint" is a task for a chat model.

[译文] [Author]: “为API端点添加速率限制（rate limiting）”是交给聊天模型的一个任务。

[原文] [Author]: You get code back. You paste it somewhere. You run it. You debug the integration yourself.

[译文] [Author]: 你得到返回的代码。你把它粘贴到某个地方。你运行它。你自己去调试集成。

[原文] [Author]: "Implement rate limiting across all authenticated API endpoints, add tests, and make sure it doesn't break the existing auth flow" is a task for Claude Code.

[译文] [Author]: “跨所有经过身份验证的API端点实现速率限制，添加测试，并确保它不会破坏现有的鉴权流程”，这才是交给Claude Code的一个任务。

[原文] [Author]: It reads your codebase, writes the implementation, finds the relevant test files, runs the tests, fixes what breaks, and tells you what it did.

[译文] [Author]: 它会阅读你的代码库，编写实现代码，找到相关的测试文件，运行测试，修复报错的地方，并告诉你它做了什么。

[原文] [Author]: That's not autocomplete.

[译文] [Author]: 这不是代码自动补全。

[原文] [Author]: That's an engineer working a ticket.

[译文] [Author]: 这是一个工程师在处理工作单（ticket）。

[原文] [Author]: The operators using Claude Code for full feature development are shipping at a pace that's structurally impossible to match with manual engineering.

[译文] [Author]: 使用Claude Code进行完整特性开发的开发者，其交付速度是手工编程在结构上根本无法匹敌的。

[原文] [Author]: Not because the code is always perfect.

[译文] [Author]: 并不是因为代码总是完美的。

[原文] [Author]: But because the iteration cycle is compressed to minutes instead of hours.

[译文] [Author]: 而是因为迭代周期被压缩到了几分钟，而不是几个小时。

章节 6：功能 5 - Claude Code 中的子智能体（Subagents）

📝 本节摘要：

剖析了在主智能体下分配子智能体的好处：每个子智能体在独立的窗口处理特定的上下文任务，不仅避免了污染主对话的语境，还能通过模型路由大幅削减运行成本，并构建具备长期记忆的特定领域助手。

[原文] [Author]: FEATURE 5: SUBAGENTS IN CLAUDE CODE

[译文] [Author]: 功能 5：CLAUDE CODE 中的子智能体（SUBAGENTS）

[原文] [Author]: This is the feature inside the feature that most people miss.

[译文] [Author]: 这是大多数人都会错过的、隐藏在功能之内的功能。

[原文] [Author]: Subagents are specialized AI assistants inside Claude Code that handle specific types of tasks.

[译文] [Author]: 子智能体是Claude Code内部专门处理特定类型任务的专业AI助手。

[原文] [Author]: Each subagent runs in its own context window with a custom system prompt, specific tool access, and independent permissions.

[译文] [Author]: 每个子智能体都在自己独立的上下文窗口中运行，并拥有自定义的系统提示词（system prompt）、特定的工具访问权限以及独立的权限。

[原文] [Author]: When Claude encounters a task matching a subagent's description, it delegates to that subagent, which works independently and returns results.

[译文] [Author]: 当Claude遇到与某个子智能体描述相匹配的任务时，它会将其委派给该子智能体，子智能体会独立完成工作并返回结果。

[原文] [Author]: Built-in subagents include Explore (fast, read-only codebase analysis), Plan (task planning before execution), and general-purpose.

[译文] [Author]: 内置的子智能体包括 Explore（快速、只读的代码库分析）、Plan（执行前的任务规划），以及通用型智能体。

[原文] [Author]: You can build custom ones for anything: code review, security auditing, test generation, documentation writing, database optimization.

[译文] [Author]: 你可以为任何事情构建自定义的子智能体：代码审查、安全审计、测试用例生成、文档编写、数据库优化。

[原文] [Author]: Subagents serve two major functions:

[译文] [Author]: 子智能体发挥两大主要功能：

[原文] [Author]: Context preservation.

[译文] [Author]: 第一是上下文保护。

[原文] [Author]: Long sessions accumulate context fast.

[译文] [Author]: 冗长的会话会极快地积累上下文。

[原文] [Author]: A subagent handles exploration or a bounded task in its own window, keeping the main conversation clean.

[译文] [Author]: 子智能体在自己的窗口中处理探索性或有边界的任务，从而保持主对话的干净整洁。

[原文] [Author]: You're not burning context tokens on tangential work.

[译文] [Author]: 你不必在偏离主线的任务上浪费上下文的Tokens。

[原文] [Author]: Cost routing.

[译文] [Author]: 第二是成本路由（Cost routing）。

[原文] [Author]: The subagent configuration includes a model field.

[译文] [Author]: 子智能体的配置中包含一个模型字段。

[原文] [Author]: You can route repetitive or low-complexity tasks to Haiku at a fraction of the cost, while Sonnet or Opus handles the judgment calls.

[译文] [Author]: 你可以将重复或低复杂度的任务路由给Haiku模型，只需极低的成本，而让Sonnet或Opus来处理需要判断力的高级任务。

[原文] [Author]: Same session, intelligent cost management.

[译文] [Author]: 在同一个会话中，实现了智能的成本管理。

[原文] [Author]: You can also give a subagent persistent memory: a directory that survives across conversations, building up codebase patterns, debugging insights, and architectural decisions over time.

[译文] [Author]: 你还可以赋予子智能体持久化记忆：一个能跨会话存活的目录，随时间推移逐渐积累代码库模式、调试洞见和架构决策。

[原文] [Author]: Your code reviewer subagent that's reviewed 30 of your PRs knows your conventions better than a stateless prompt ever will.

[译文] [Author]: 一个审阅过你30个PR（Pull Requests）的代码审查子智能体，远比一个无状态的（stateless）提示词更懂你的代码约定。

章节 7：功能 6 - 远程控制与分发（Remote Control and Dispatch）

📝 本节摘要：

展示了打破物理终端束缚的远程与异步执行功能。允许用户随时切换设备（如手机或浏览器）无缝接管长耗时任务，彻底改变了与AI协作的时间和空间模式。

[原文] [Author]: FEATURE 6: REMOTE CONTROL AND DISPATCH

[译文] [Author]: 功能 6：远程控制与分发（REMOTE CONTROL AND DISPATCH）

[原文] [Author]: These two features change what it means to "work" with Claude Code.

[译文] [Author]: 这两个功能改变了与Claude Code“共事”的含义。

[原文] [Author]: Remote Control means your Claude Code sessions are not tied to your terminal.

[译文] [Author]: 远程控制（Remote Control）意味着你的Claude Code会话不再与你的终端强绑定。

[原文] [Author]: You can step away from your desk and keep working from your phone or any browser.

[译文] [Author]: 你可以离开办公桌，并通过手机或任何浏览器继续工作。

[原文] [Author]: A long-running task you kicked off in your terminal continues, and you monitor or redirect it from claude.ai/code or the Claude iOS app.

[译文] [Author]: 你在终端中发起的一个长时间运行的任务会继续执行，而你可以通过 claude.ai/code 或 Claude iOS 应用程序来监控或重定向它。

[原文] [Author]: You never have to be at your machine to stay in the loop.

[译文] [Author]: 你永远不需要守在机器旁也能保持跟进。

[原文] [Author]: Dispatch is the asynchronous counterpart.

[译文] [Author]: 分发（Dispatch）则是其异步（asynchronous）对应功能。

[原文] [Author]: You message a task from your phone and Claude Code opens a Desktop session to execute it.

[译文] [Author]: 你从手机发送一个任务消息，Claude Code会开启一个桌面会话来执行它。

[原文] [Author]: The session runs. You check back when it's done. You can be on the other side of the world.

[译文] [Author]: 会话在后台运行。完成后你再回来查看。你甚至可以身处地球的另一端。

[原文] [Author]: The practical combination: kick off a large refactor before you leave the office.

[译文] [Author]: 极具实操性的组合用法是：在你离开办公室前，启动一次大型代码重构。

[原文] [Author]: Check progress from your phone at dinner. Redirect if it's going wrong. Come back to a finished task.

[译文] [Author]: 晚餐时用手机检查进度。如果出现偏差则重新定向。回去后直接收获一个完成的任务。

[原文] [Author]: There's also claude --teleport, which pulls a web or iOS session into your local terminal when you want to continue there.

[译文] [Author]: 还有一个 claude --teleport 命令，当你想在本地继续工作时，它可以将网页端或iOS端的会话拉取到你本地终端中。

[原文] [Author]: And /desktop, which hands off a terminal session to the Desktop app for visual diff review.

[译文] [Author]: 还有 /desktop 命令，它可以将终端会话移交给桌面应用程序，以便进行可视化的差异审查（visual diff review）。

[原文] [Author]: These aren't quality-of-life features. They're infrastructure.

[译文] [Author]: 这些不是简单的“改善生活质量”的小功能。它们是基础设施。

[原文] [Author]: The moment Claude Code sessions are no longer tied to a physical machine in a physical location, the entire operating model changes.

[译文] [Author]: 当Claude Code会话不再被限制于某个物理位置的物理机器上时，整个操作模型就发生了根本性的改变。

章节 8：功能 7 - 模型上下文协议（MCPs）

📝 本节摘要：

讲述了MCPs协议如何将Claude从纯语言模型升格为真正的自动化操作引擎。通过安全直连外部系统（如Airtable、Slack、Notion），AI不再只是“谈论事物”，而是能代表用户“去执行操作”。

[原文] [Author]: FEATURE 7: MCPs (Model Context Protocol)

[译文] [Author]: 功能 7：MCPs（模型上下文协议，Model Context Protocol）

[原文] [Author]: This is the feature that transforms Claude from a language model into an operational system.

[译文] [Author]: 正是这个功能将Claude从一个语言模型转变成了一个操作系统（operational system）。

[原文] [Author]: MCPs are standardized connectors that give Claude real access to real tools.

[译文] [Author]: MCPs 是标准化的连接器，赋予Claude对真实工具的真正访问权限。

[原文] [Author]: Not simulated access. Not summarized output from a screenshot. Actual read and write access to external systems.

[译文] [Author]: 不是模拟的访问。不是对屏幕截图的摘要输出。而是对外部系统实际的读写访问权限。

[原文] [Author]: With the right MCP stack, Claude can:

[译文] [Author]: 借助合适的MCP技术栈，Claude能够：

[原文] [Author]: → Read your Airtable and write back to it

[译文] [Author]: → 读取你的Airtable数据并写回更改

[原文] [Author]: → Search your Notion workspace and create new pages

[译文] [Author]: → 搜索你的Notion工作区并创建新页面

[原文] [Author]: → Pull from your Google Drive and update documents

[译文] [Author]: → 从你的Google Drive拉取并更新文档

[原文] [Author]: → Post to Slack on your behalf

[译文] [Author]: → 代表你向Slack发布消息

[原文] [Author]: → Execute queries against your database

[译文] [Author]: → 针对你的数据库执行查询

[原文] [Author]: → Trigger webhooks in n8n or Make

[译文] [Author]: → 在n8n或Make中触发webhooks

[原文] [Author]: → Control a browser and take actions on web pages

[译文] [Author]: → 控制浏览器并在网页上执行操作

[原文] [Author]: The distinction that matters: Claude without MCPs gives you text about things. Claude with MCPs gives you Claude doing things.

[译文] [Author]: 真正重要的区别在于：没有MCPs的Claude只能给你提供关于事物的文本。有了MCPs的Claude则是让Claude去“做事”。

[原文] [Author]: I have Claude connected to Airtable, Slack, and Google Drive across multiple client builds.

[译文] [Author]: 在多个客户的项目构建中，我已将Claude连接到了Airtable、Slack和Google Drive。

[原文] [Author]: When I'm running an outbound campaign and want to check lead pipeline status, I don't open Airtable.

[译文] [Author]: 当我运行外呼营销活动并希望检查线索管道状态时，我不会去打开Airtable。

[原文] [Author]: I ask Claude.

[译文] [Author]: 我直接问Claude。

[原文] [Author]: It reads the table, applies whatever logic I need, and tells me what I actually want to know: who's stuck, what needs attention, where I need to intervene.

[译文] [Author]: 它会读取表格，应用我需要的任何逻辑，然后告诉我真正想知道的内容：谁卡住了、哪里需要注意、哪里需要我介入。

[原文] [Author]: That is an entirely different tool than a chat window.

[译文] [Author]: 这是一个与纯聊天窗口截然不同的工具。

章节 9：功能 8 - 插件与市场（Plugins and the Marketplace）

📝 本节摘要：

介绍了官方插件市场的强大扩展能力。通过将指令、子智能体和MCP打包为插件，不仅方便个人快速集成工具，更适合企业将私有插件下发到整个研发团队，确保高度统一的AI工作流标准。

[原文] [Author]: FEATURE 8: PLUGINS AND THE MARKETPLACE

[译文] [Author]: 功能 8：插件与市场（PLUGINS AND THE MARKETPLACE）

[原文] [Author]: Claude Code's capabilities are extendable, and the plugin ecosystem is where the real surface area opens up.

[译文] [Author]: Claude Code的能力是可扩展的，而插件生态系统正是真正拓展其应用边界的地方。

[原文] [Author]: Plugins package together skills, agents, hooks, MCP servers, and LSP configurations into reusable bundles you can share across projects and teams.

[译文] [Author]: 插件将技能（skills）、智能体（agents）、钩子（hooks）、MCP服务器以及LSP（语言服务器协议）配置打包在一起，形成可重用的组件包，让你可以在项目和团队之间共享。

[原文] [Author]: The official Anthropic marketplace is automatically available the moment you open Claude Code.

[译文] [Author]: 当你打开Claude Code时，官方的Anthropic市场会自动变为可用状态。

[原文] [Author]: Run /plugin to open the plugin manager and browse what's available, or view the full catalog at claude.com/plugins.

[译文] [Author]: 运行 /plugin 可以打开插件管理器并浏览可用内容，或者在 claude.com/plugins 查看完整目录。

[原文] [Author]: What plugins actually add:

[译文] [Author]: 插件实际添加了以下内容：

[原文] [Author]: → Skills and custom commands specific to a domain (git workflows, deployment tooling, testing frameworks)

[译文] [Author]: → 特定领域的技能和自定义命令（如git工作流、部署工具、测试框架）

[原文] [Author]: → Subagents with pre-built system prompts for specialized tasks

[译文] [Author]: → 带有预置系统提示词、用于执行专门任务的子智能体

[原文] [Author]: → Hooks that fire before or after specific tool calls

[译文] [Author]: → 在调用特定工具之前或之后触发的钩子（Hooks）

[原文] [Author]: → MCP server configurations bundled and ready to connect

[译文] [Author]: → 打包好随时准备连接的MCP服务器配置

[原文] [Author]: → LSP servers for real-time code intelligence in specific languages

[译文] [Author]: → 针对特定语言提供实时代码智能提示的LSP服务器

[原文] [Author]: The enterprise use case is significant.

[译文] [Author]: 其在企业级应用中的价值意义重大。

[原文] [Author]: Organizations can build private plugin marketplaces distributed to their team via a single URL.

[译文] [Author]: 组织可以构建私有插件市场，并通过单一的URL将其分发给整个团队。

[原文] [Author]: Every developer on the team gets the same custom commands, the same subagents, the same MCP configurations.

[译文] [Author]: 团队中的每个开发人员都能获得相同的自定义命令、相同的子智能体以及相同的MCP配置。

[原文] [Author]: One plugin update propagates across the entire team automatically with auto-update enabled.

[译文] [Author]: 启用自动更新后，一次插件更新便可自动同步至整个团队。

[原文] [Author]: For individual operators, the official marketplace already has plugins for Telegram, Discord, and iMessage channels, LSP integrations for major languages, and git workflow tools.

[译文] [Author]: 对于个人操作者，官方市场中已经提供了Telegram、Discord和iMessage频道的插件、主流语言的LSP集成，以及git工作流工具。

[原文] [Author]: Most operators haven't opened the plugin manager once.

[译文] [Author]: 然而大多数操作者连一次插件管理器都没有打开过。

章节 10：功能 9 - 通道（Channels：Slack, Telegram, iMessage）

📝 本节摘要：

通道功能允许将Claude Code嵌入到用户最常用的即时通讯软件中。通过降低上下文切换所产生的专注力损耗，实现零摩擦的AI操作体验，极大提升了多端同步与移动办公的效率。

[原文] [Author]: FEATURE 9: CHANNELS (CLAUDE IN SLACK, TELEGRAM, IMESSAGE)

[译文] [Author]: 功能 9：通道（CHANNELS）（集成在 SLACK、TELEGRAM、IMESSAGE 中的 CLAUDE）

[原文] [Author]: Claude Code doesn't have to live in your terminal.

[译文] [Author]: Claude Code不必局限在你的终端里。

[原文] [Author]: Channels are persistent, bidirectional connections between Claude Code and external communication systems.

[译文] [Author]: 通道（Channels）是Claude Code与外部通信系统之间持久化的双向连接。

[原文] [Author]: Through the official plugin marketplace, you can connect Claude Code to Telegram, Discord, Slack, and iMessage.

[译文] [Author]: 通过官方插件市场，你可以将Claude Code连接到Telegram、Discord、Slack和iMessage。

[原文] [Author]: Your Claude Code session runs on your machine.

[译文] [Author]: 你的Claude Code会话在你的本地机器上运行。

[原文] [Author]: Messages from your phone route through the channel.

[译文] [Author]: 手机发出的消息将通过该通道进行路由。

[原文] [Author]: Claude reads them, does the work, and replies in the thread.

[译文] [Author]: Claude会读取它们、完成工作，并在该消息树下回复。

[原文] [Author]: Here's what this actually enables:

[译文] [Author]: 以下是它实际能够实现的功能：

[原文] [Author]: → Slack: mention @Claude in a channel with a bug report and get a pull request back

[译文] [Author]: → Slack：在频道里@Claude发送一份bug报告，就能获得一个返回的PR（Pull Request）。

[原文] [Author]: → Telegram: message a task to your bot from your phone while your session runs on your VPS

[译文] [Author]: → Telegram：通过手机向你的机器人发送一个任务，而你的会话依然在VPS上运行。

[原文] [Author]: → iMessage: self-chat bypasses access control with no setup, works immediately on any Apple device

[译文] [Author]: → iMessage：无需任何设置，直接通过给自己发消息绕过访问控制，在任何苹果设备上即可立即生效。

[原文] [Author]: → Discord: DM your bot a task, it executes in the background and replies with output

[译文] [Author]: → Discord：私信给你的机器人一个任务，它会在后台执行并回复输出结果。

[原文] [Author]: The thing that makes channels powerful isn't the communication layer. It's the reduction in friction.

[译文] [Author]: 让通道变得如此强大的，并非其通信层，而在于它减少了摩擦。

[原文] [Author]: Every time you have to open a new tool, context switch to a new interface, and re-establish what you're doing, you lose 30 seconds and a piece of your focus.

[译文] [Author]: 每次你不得不打开一个新工具、将上下文切换到一个新界面，并重新梳理自己正在做的事情时，你都会损失30秒和一部分专注力。

[原文] [Author]: Claude in your existing communication layer means AI-powered actions require zero context switching.

[译文] [Author]: 将Claude整合进你现有的通信层中，意味着利用AI执行操作时需要“零”上下文切换。

[原文] [Author]: If you're running Claude Code on a VPS for long-running tasks, a Telegram channel means you're always in the loop regardless of where you are.

[译文] [Author]: 如果你为了执行长时间任务而在VPS上运行Claude Code，那么连接一个Telegram通道意味着无论你身处何地，你都在随时跟进最新情况。

[原文] [Author]: Your session keeps running. You keep steering.

[译文] [Author]: 你的会话继续运行。你继续把控方向。

章节 11：功能 10 - 并行智能体与模型委员会（Parallel Agents and Model Council）

📝 本节摘要：

针对大体量任务的性能瓶颈，提出了跨会话并行调度多智能体的新范式；并结合模型特长引入“模型委员会”的概念（便宜的小模型处理海量降噪，昂贵的大模型负责关键决策），兼顾了效率与成本优化。

[原文] [Author]: FEATURE 10: PARALLEL AGENTS AND MODEL COUNCIL

[译文] [Author]: 功能 10：并行智能体与模型委员会（PARALLEL AGENTS AND MODEL COUNCIL）

[原文] [Author]: Sequential AI is leaving performance on the table.

[译文] [Author]: 串行化（Sequential）AI正在让大量性能白白流失。

[原文] [Author]: Most AI workflows look like this: one task, one agent, wait for completion, next task.

[译文] [Author]: 大多数AI工作流看起来是这样的：一个任务，一个智能体，等待完成，然后是下一个任务。

[原文] [Author]: That's fine for simple work. For anything complex, it's a bottleneck.

[译文] [Author]: 这对简单工作来说没问题。但对于任何复杂任务，这就是个瓶颈。

[原文] [Author]: Parallel agent teams in Claude Code mean spawning multiple Claude instances against independent subtasks simultaneously, then synthesizing the outputs.

[译文] [Author]: Claude Code中的并行智能体团队（Parallel agent teams）意味着同时派生多个Claude实例来处理各自独立的子任务，然后综合其输出结果。

[原文] [Author]: Subagents work within a single session. Agent teams coordinate across separate sessions.

[译文] [Author]: 子智能体（Subagents）在单一会话内工作。而智能体团队（Agent teams）则跨越不同会话进行协调。

[原文] [Author]: For large parallel workstreams, teams are the pattern.

[译文] [Author]: 面对大型并行工作流时，团队协同才是正确的模式。

[原文] [Author]: Here's what this looks like on a real build:

[译文] [Author]: 在实际构建项目中，它是这样运作的：

[原文] [Author]: A lead enrichment pipeline processing 200 companies.

[译文] [Author]: 一个负责处理200家公司的潜在客户资料丰富化（lead enrichment）管道。

[原文] [Author]: Sequential enrichment, one company at a time, takes the better part of an hour.

[译文] [Author]: 一次一家公司地进行串行处理，要花去将近一个小时的大半时间。

[原文] [Author]: Parallel processing spins up multiple enrichment agents against batched lead sets simultaneously, runs scoring in parallel on completed enrichments, and collapses the outputs into a single scored list.

[译文] [Author]: 并行处理则会同时启动多个代理来处理分批的线索集，对已完成资料丰富化的结果进行并行评分，然后将输出汇总成一张单一的评分列表。

[原文] [Author]: Total wall time: minutes.

[译文] [Author]: 整体实际消耗时间（wall time）：只需几分钟。

[原文] [Author]: The Model Council pattern takes this further.

[译文] [Author]: 模型委员会（Model Council）模式将这一点推向了更深层次。

[原文] [Author]: Different models handle different parts of a workflow based on what they're actually best at.

[译文] [Author]: 不同的模型根据它们实际擅长的领域，去处理工作流中不同的部分。

[原文] [Author]: Claude doesn't do everything.

[译文] [Author]: Claude并不包揽一切。

[原文] [Author]: Cheap, fast models filter high-volume noise at fractions of a cent per call.

[译文] [Author]: 便宜且快速的模型以每次调用几分之一美分的成本来过滤海量噪音。

[原文] [Author]: The expensive model handles the judgment calls that actually matter.

[译文] [Author]: 而昂贵的模型则处理那些真正至关重要的判断。

[原文] [Author]: One production example: a news intelligence pipeline processes 200 articles nightly.

[译文] [Author]: 一个生产环境的例子：一条新闻情报管道每晚处理200篇文章。

[原文] [Author]: Groq filters at $0.001 per call and eliminates 170 irrelevant results.

[译文] [Author]: Groq模型以每次调用$0.001的成本进行过滤，并淘汰掉170个不相关的结果。

[原文] [Author]: Claude Sonnet scores the remaining 30 against proprietary relevance criteria at $0.05 per call.

[译文] [Author]: 随后Claude Sonnet根据专有的相关性标准对剩下的30篇文章进行打分，成本为每次调用$0.05。

[原文] [Author]: The expensive model never sees the noise.

[译文] [Author]: 昂贵的模型完全不需要接触到这些噪音数据。

[原文] [Author]: The cheap model never makes the high-stakes decisions.

[译文] [Author]: 而廉价的模型也从不需要做高风险的决策。

[原文] [Author]: Every model doing what it's structurally best at.

[译文] [Author]: 让每个模型去负责其在结构上最擅长的事情。

[原文] [Author]: The orchestration logic making sure nothing is wasted and nothing is missed.

[译文] [Author]: 编排逻辑确保没有任何浪费，也没有任何遗漏。

章节 12：功能 11 - 本地代码模型（Local Coding Models: Ollama）

📝 本节摘要：

揭示了Claude Code的巨大灵活性：它可以将计算后端路由至如Ollama运行的本地开源大模型上。这种无API费用的私有化部署策略不仅保障了敏感代码安全，还在处理高频样板代码时实现了零边际成本的经济效益。

[原文] [Author]: FEATURE 11: LOCAL CODING MODELS IN CLAUDE CODE (OLLAMA)

[译文] [Author]: 功能 11：Claude Code 中的本地代码模型（OLLAMA）

[原文] [Author]: Claude Code doesn't have to run against Claude's API.

[译文] [Author]: Claude Code不一定非要对接Claude的官方API运行。

[原文] [Author]: You can configure Claude Code to route to local models running through Ollama.

[译文] [Author]: 你可以配置Claude Code将其路由到通过Ollama运行的本地模型上。

[原文] [Author]: This changes the cost structure and the privacy profile of everything you build with it.

[译文] [Author]: 这改变了你所构建的一切事物的成本结构和隐私属性。

[原文] [Author]: For teams working on proprietary code that can't leave their infrastructure, local models are not optional.

[译文] [Author]: 对于处理不能离开自身基础设施环境的专有代码的团队来说，本地模型不是可选项（而是必选项）。

[原文] [Author]: They're the only viable path.

[译文] [Author]: 它们是唯一可行的路径。

[原文] [Author]: For operators running high-volume development workflows where API costs compound fast, local models on capable hardware change the economics entirely.

[译文] [Author]: 对于运行大体量开发工作流、API成本激增的操作者而言，性能良好的硬件加上本地模型完全改变了其经济账本。

[原文] [Author]: The setup: install Ollama, pull the model you want (Qwen2.5-Coder, DeepSeek-Coder, and similar), configure Claude Code to route to the local endpoint.

[译文] [Author]: 设置流程：安装Ollama，拉取你想要的模型（如 Qwen2.5-Coder、DeepSeek-Coder及类似模型），然后配置Claude Code路由至本地端点。

[原文] [Author]: Your coding agent now runs on your machine with no API calls, no rate limits, and no data leaving your environment.

[译文] [Author]: 现在你的编码智能体就在你的机器上运行了，没有任何API调用，没有速率限制（rate limits），也没有数据会离开你的环境。

[原文] [Author]: The performance gap between frontier Claude and local models is real.

[译文] [Author]: 必须承认，前沿的Claude模型与本地模型之间的性能差距是真实存在的。

[原文] [Author]: For complex architectural reasoning and novel problem-solving, Claude wins.

[译文] [Author]: 在复杂的架构推理和解决新颖问题时，Claude无疑胜出。

[原文] [Author]: For repetitive coding tasks, refactoring, boilerplate generation, and test writing, capable local models are genuinely close, at zero marginal cost.

[译文] [Author]: 但针对重复性的编码任务、代码重构、样板代码（boilerplate）生成和测试编写，强大的本地模型表现已经非常接近了，而且其边际成本为零。

[原文] [Author]: Operators who understand this run Claude for the hard thinking and local models for the volume work.

[译文] [Author]: 理解这一点的操作者会让Claude负责深度思考，而让本地模型负责海量的搬砖工作。

[原文] [Author]: Same interface, different cost structure.

[译文] [Author]: 同一个界面，不同的成本结构。

章节 13：功能 12 - 虚拟专用服务器（VPS）与本地环境的对比

📝 本节摘要：

总结了如何根据任务属性选择Claude Code的运行环境（本地、云桌面应用或云服务器）。本地适合交互开发，VPS则是用来托管彻夜自动运行的管道和数据流，使AI最终转变为稳定可靠的基础设施。

[原文] [Author]: FEATURE 12: CLAUDE CODE ON VPS VS LOCAL ENVIRONMENT

[译文] [Author]: 功能 12：在VPS上的Claude Code与本地环境的对比

[原文] [Author]: Where you run Claude Code matters as much as how you run it.

[译文] [Author]: 你在哪里运行Claude Code，与你如何运行它一样重要。

[原文] [Author]: Local development is the default. Your machine, your terminal, your development environment.

[译文] [Author]: 本地开发是默认配置。你的机器、你的终端、你的开发环境。

[原文] [Author]: It works. It's limited by your hardware, your network, and your machine being off when you're not at your desk.

[译文] [Author]: 它是管用的。但它受限于你的硬件、网络，以及当你不在办公桌前机器处于关闭状态。

[原文] [Author]: Remote sessions through Claude Code Desktop solve part of this: they run on Anthropic-managed cloud infrastructure and continue even if you close the app or shut down your computer.

[译文] [Author]: 通过Claude Code桌面应用进行的远程会话解决了一部分问题：它们运行在Anthropic管理的云端基础设施上，即使你关闭应用或关机也会继续运行。

[原文] [Author]: For large refactors, test suites, and long-running migrations, this is the right call.

[译文] [Author]: 对于大型重构、测试套件以及长时间运行的迁移任务，这是正确的选择。

[原文] [Author]: Check progress anytime from claude.ai/code or the Claude iOS app.

[译文] [Author]: 你可以随时通过 claude.ai/code 或 Claude iOS 应用程序检查进度。

[原文] [Author]: VPS deployment solves the rest.

[译文] [Author]: VPS部署则解决了其余的问题。

[原文] [Author]: A pipeline I've built runs nightly on a Hetzner instance, pulling data, processing it through Claude, pushing outputs to Airtable, without me touching anything.

[译文] [Author]: 我构建的一条管道每晚都在一个Hetzner云服务器实例上运行：提取数据、通过Claude进行处理、将输出结果推送到Airtable，而我完全不需要进行任何人工操作。

[原文] [Author]: It ran last night. It'll run tonight.

[译文] [Author]: 昨晚它运行了。今晚它也会继续运行。

[原文] [Author]: Local Claude Code is a tool you use. VPS Claude Code is infrastructure you've deployed.

[译文] [Author]: 本地的Claude Code是你使用的一个工具。而VPS上的Claude Code则是你部署的基础设施（infrastructure）。

[原文] [Author]: The practical decision framework:

[译文] [Author]: 实用的决策框架如下：

[原文] [Author]: → Short tasks, active development, reviewing diffs: local

[译文] [Author]: → 短任务、敏捷开发、审查代码差异（diffs）：使用本地环境

[原文] [Author]: → Long-running tasks, overnight processing, tasks you want to check from your phone: remote or VPS

[译文] [Author]: → 耗时任务、过夜处理、想在手机上检查进度的任务：使用远程会话或VPS

[原文] [Author]: → Proprietary codebases, high-volume workflows, cost optimization: local with Ollama

[译文] [Author]: → 专有代码库、大体量工作流、成本优化：结合Ollama使用本地环境

[原文] [Author]: For anything beyond personal experimentation, understanding which environment fits which task type is what separates operators running Claude at scale from operators who happen to use Claude a lot.

[译文] [Author]: 对于任何超越个人试验性质的工作，理解哪种环境适合哪种任务类型，正是区分那些能够规模化运行Claude的操作者与那些仅仅是“用得比较多”的用户之间的分水岭。

章节 14：实操指南：如何真正开始（How To Actually Start）

📝 本节摘要：

作者提供了一份稳健的行动日历，建议用户不要试图一次性掌握所有高级功能。从第一周配置项目与初始化文档，到第二周打通第一个MCP外部工具，再到第三个月引入并行协作流，按部就班地将AI从“计算器”进化为企业的核心“神经网络”。

[原文] [Author]: HOW TO ACTUALLY START

[译文] [Author]: 如何真正开始（HOW TO ACTUALLY START）

[原文] [Author]: The full stack above took months to build. You don't need all of it this week.

[译文] [Author]: 上述全栈能力花了我数月时间才建立起来。你没必要在这个星期就弄懂一切。

[原文] [Author]: Here's the sequence that matters:

[译文] [Author]: 以下是真正重要的关键顺序：

[原文] [Author]: Week 1: Projects and CLAUDE.MD.

[译文] [Author]: 第一周：掌握项目（Projects）和CLAUDE.MD。

[原文] [Author]: Set up a Claude Project for your primary use case.

[译文] [Author]: 为你的主要使用场景建立一个Claude项目。

[原文] [Author]: Write proper project instructions, including who you are, what you're doing, and what context Claude needs before the first message.

[译文] [Author]: 编写得当的项目指令，包括你是谁、你在做什么，以及Claude在处理第一条消息前需要哪些上下文。

[原文] [Author]: If you're doing any development, create a CLAUDE.MD in your main repo.

[译文] [Author]: 如果你在进行任何开发工作，请在主代码库中创建一个CLAUDE.MD文件。

[原文] [Author]: These two things alone will change your daily interaction with Claude more than anything else on this list.

[译文] [Author]: 仅仅是做好了这两件事，就能比列表上的其他任何功能都更能改变你与Claude的日常交互方式。

[原文] [Author]: Week 2 to 3: First MCP.

[译文] [Author]: 第二周到第三周：打通第一个MCP。

[原文] [Author]: Pick the tool you use most. Notion, Airtable, Google Drive, Slack.

[译文] [Author]: 挑选你最常用的那个工具。比如Notion、Airtable、Google Drive或Slack。

[原文] [Author]: Find its MCP connector. Set it up. Give Claude real access to one thing.

[译文] [Author]: 找到它的MCP连接器。设置它。赋予Claude对某一事物的真实访问权限。

[原文] [Author]: Run one actual workflow where Claude reads from or writes to that system.

[译文] [Author]: 运行一个真实的业务流，让Claude能够从该系统读取或写入数据。

[原文] [Author]: That first end-to-end moment changes how you think about everything else.

[译文] [Author]: 这第一次端到端（end-to-end）时刻将彻底改变你对所有其他事物的看法。

[原文] [Author]: Month 2: Claude Code and Subagents.

[译文] [Author]: 第二个月：掌握Claude Code与子智能体（Subagents）。

[原文] [Author]: Start using Claude Code for feature-level tasks rather than function-level completions.

[译文] [Author]: 开始将Claude Code用于特性级别（feature-level）的任务，而不是函数级别的补全。

[原文] [Author]: Write your first CLAUDE.MD.

[译文] [Author]: 写下你的第一份CLAUDE.MD文件。

[原文] [Author]: Build your first custom subagent for a task you repeat weekly.

[译文] [Author]: 为你每周都要重复的任务构建第一个自定义子智能体。

[原文] [Author]: Let it read the codebase and plan before it writes anything.

[译文] [Author]: 能够让它在编写任何内容之前，先阅读代码库并进行规划。

[原文] [Author]: Month 3 and beyond: Orchestration and Channels.

[译文] [Author]: 第三个月及以后：探索编排（Orchestration）与通道（Channels）。

[原文] [Author]: Once the foundation is working, introduce parallel agents, model routing, and channels.

[译文] [Author]: 一旦地基打牢，就可以开始引入并行智能体、模型路由和通道。

[原文] [Author]: Start thinking about pipelines rather than conversations.

[译文] [Author]: 开始用管道（pipelines）的思维方式去思考问题，而不是单纯的对话。

[原文] [Author]: Deploy a VPS session for your first overnight task.

[译文] [Author]: 部署一个VPS会话来处理你的第一个过夜任务。

[原文] [Author]: The mistake most people make is trying to implement everything at once.

[译文] [Author]: 大多数人常犯的错误就是试图一次性实施所有的功能。

[原文] [Author]: The correct approach is sequential and operational.

[译文] [Author]: 正确的方法应当是渐进式的（sequential）且具有实操性的（operational）。

[原文] [Author]: Each layer working before you add the next.

[译文] [Author]: 确保每一层都在稳定工作后，再去叠加下一层。

[原文] [Author]: Most people use Claude like a calculator.

[译文] [Author]: 大多数人像使用计算器一样使用Claude。

[原文] [Author]: The operators running serious volume use it like a nervous system, wired into their data, their tools, their communication, their development environment.

[译文] [Author]: 而处理海量任务的专业操作者则把它用作一套神经系统，深深地连结到他们的数据、工具、通信以及开发环境之中。

[原文] [Author]: Instructions encoded once. Context persistent across sessions.

[译文] [Author]: 指令只需编码一次。上下文跨会话持久存在。

[原文] [Author]: Multiple agents running parallel workstreams.

[译文] [Author]: 多个智能体并行运行工作流。

[原文] [Author]: Outputs landing where decisions get made.

[译文] [Author]: 而输出结果，精准落地到做出决策的关键节点。

[原文] [Author]: The features exist. They've existed for a while.

[译文] [Author]: 这些功能都是存在的。并且已经存在了一段时间。

[原文] [Author]: The gap between the 1% and everyone else isn't access.

[译文] [Author]: 前1%的人与其他人之间的差距，并不在于是否拥有访问权限。

[原文] [Author]: It's knowing the full picture and having the discipline to build it properly.

[译文] [Author]: 而在于是否了解全局，以及是否有纪律性去正确地构建这一切。

[原文] [Author]: Now you have the picture.

[译文] [Author]: 现在，你已经看到了全局。