# Aklman · Library

> Aklman 的独立 AI 电子书库 —— 每本书讲清一份 AI 订阅或一个工作流。中英双语，完整章节，在线阅读，并允许 AI 摘读、引用与问答。
> An independent library of short AI books by Aklman. Bilingual, full chapters, free to read online; AI summarization, quoting, and Q&A welcome.

## 站点入口 / Site entry

- 主页 / Home: https://library.aklman.com/
- 文库 / Library: https://library.aklman.com/library
- RSS: https://library.aklman.com/feed.xml
- 关于 / About: https://library.aklman.com/about
- 授权 / License: https://library.aklman.com/license
- 本索引 / This index: https://library.aklman.com/llms.txt

## 书目 / Books

### Agent 实战 · 从零构建完备智能体 / Agent in Practice · Building a Complete Agent from Scratch

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.07
- 19 章 · 205 分钟 · TECH / PRACTICE
- 一个完整的动手项目：从一个只会聊天的 LLM 出发，逐步加入工具调用、记忆系统、推理规划、MCP 协议、多 Agent 编排、安全护栏，部署到生产环境，再把各章的格子拼成一个能离线跑、也能接真实模型的 Python 工程。企业篇在其上补齐身份鉴权、对抗性安全、治理审计与控制平面，最后收束成一张参考架构和一份可以直接交给编码 agent 实现的分层规格。
- A complete hands-on project: starting from a bare LLM that can only chat, progressively adding tool use, memory, reasoning, MCP, multi-agent orchestration, and safety guardrails through production deployment, then assembling the slots into one Python project that runs offline or against a real model. An enterprise part adds identity and auth, adversarial security, governance and audit, and a control plane — closing with a reference architecture and a layered spec you can hand to a coding agent.
- 阅读页 / Web: https://library.aklman.com/books/build-agent
- AI-readable markdown: https://library.aklman.com/books/build-agent/llms.md

### Warp · 一台终端形状的元工具 / Warp · A Meta-Tool Shaped Like a Terminal

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.08
- 12 章 · 92 分钟 · TECH / PRACTICE
- 官网把 Warp 卖成「更快的终端」,2026 年又把自己重定位成 agentic development environment。这本书的角度是第三种:Warp 是一台终端形状的元工具 —— 它能跑命令、内置 agent、用 Warp Drive 沉淀、接 MCP,于是几乎所有走命令行、文件、脚本、系统的活都能在它里面做完,不限于写代码。写法上不靠官方叙事背书:先带你在 10 分钟内装好并拿到第一次交办成功,再逐章跑通一条完整工作流 —— 把一个真实报错修到好、把重复三次的操作固化成一键、接一个真数据源完成一次查询、把权限配到敢放手、派一批活上云再验收回来。每项能力都用 before/after 对照呈现,结论由你自己得出。倒数第二章用截至 2026-08-06 的 pricing / BYOK / custom endpoint / Cloud Agent 规则给 Free / Build / Max / Business / Enterprise 选型,并划清 Warp、Cursor、CLI agent 与 Computer Use 的边界。读完你面对自己的活,知道哪些该留在 Warp 里。
- The vendor sells Warp as a faster terminal, and in 2026 repositioned it as an agentic development environment. This book takes a third angle: Warp is a meta-tool shaped like a terminal — it runs commands, embeds an agent, distills work into Warp Drive, and speaks MCP, so almost any command-line, file, script, or system task can be finished inside it, not just code. It does not argue from the vendor's narrative: it gets you installed and through your first successful dispatch in 10 minutes, then runs one complete workflow per chapter — debugging a real error to fixed, freezing a thrice-repeated chore into one keystroke, connecting a real data source for one query, tuning permissions until you dare let go, and sending a batch of work to the cloud and reviewing what comes back. Every core capability is shown as before/after, so you draw the conclusion yourself. The penultimate chapter uses the pricing / BYOK / custom endpoint / Cloud Agent rules current as of 2026-08-06 to choose between Free / Build / Max / Business / Enterprise, then draws the line between Warp, Cursor, CLI agents, and Computer Use. By the end, you know what belongs inside Warp.
- 阅读页 / Web: https://library.aklman.com/books/warp
- AI-readable markdown: https://library.aklman.com/books/warp/llms.md

### Codex · 多 Agent 的异步编码工作流 / Codex · The Async Multi-Agent Coding Workflow

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.08
- 12 章 · 118 分钟 · TECH / PRACTICE
- Codex 是 ChatGPT 桌面应用里独立的软件开发体验,也是 CLI、IDE 扩展、云端执行与 GitHub review 组成的一套工具链。你定义目标和验收标准,主 agent 决定怎么拆活,explorer、worker 与自定义 agent 在独立线程里调查或执行,worktree 和 cloud task 再隔离会产生改动的并行工作,最后交回可审的 diff 或 PR。这本书用十二章讲清入口、权限、AGENTS.md、config、Skills、MCP、多 Agent 配置、云端、无人值守、验收与 credits;重点不是功能数量,而是怎样让任务在你不盯着时完成,同时保留证据、边界与最终判断。
- Codex is the dedicated software-development experience in the ChatGPT desktop app and a toolchain spanning the CLI, IDE extension, cloud execution, and GitHub review. You define the goal and acceptance criteria; the main agent decides how to divide the work; explorer, worker, and custom agents investigate or execute in separate threads; worktrees and cloud tasks isolate parallel work that produces changes; and the result returns as a reviewable diff or PR. Across twelve chapters, this book covers surfaces, permissions, AGENTS.md, config, Skills, MCP, multi-agent configuration, cloud, unattended runs, review, and credits. The point is not feature count, but finishing work without watching while keeping evidence, boundaries, and final judgment intact.
- 阅读页 / Web: https://library.aklman.com/books/codex
- AI-readable markdown: https://library.aklman.com/books/codex/llms.md

### Claude Code · 终端里的编排台 / Claude Code · The Orchestration Console in Your Terminal

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.07
- 11 章 · 101 分钟 · TECH / PRACTICE
- Claude Code 不只是终端里写代码的地方。计划循环、CLAUDE.md、Skills、输出风格、subagents、MCP、hooks、headless/SDK 凑在一起，是一台编排台 —— 你一个人，指挥并验收一支 agent 队。这本书不列全部功能，只挑覆盖八成场景的两成核心：每一项都答清楚「省什么、什么活用、什么时候不如自己动手或换别的工具」。倒数第二章给 Pro / Max 选型、Opus / Sonnet / Haiku 的分工、成本与 prompt caching，以及和 Cursor / Codex / Copilot / Warp 的边界。最特别的一章讲「验收」—— 编排时代唯一持续稀缺的技能：怎么系统性确认你委派的活真做对了，它直接接上《Agent 实战》。读完你知道哪些活自己盯、哪些活派出去、以及怎么确认没被坑。
- Claude Code isn't only where you write code in the terminal. The plan loop, CLAUDE.md, Skills, output styles, subagents, MCP, hooks, and headless/SDK add up to an orchestration console — one person directing and verifying a fleet of agents. This book doesn't list every feature; it picks the 20% that covers 80% of real work, answering for each: what it saves, when to use it, and when doing it yourself or switching tools wins. The penultimate chapter covers Pro / Max, routing work across Opus / Sonnet / Haiku, cost and prompt caching, and the boundary with Cursor / Codex / Copilot / Warp. The standout chapter is on review — the one skill that stays scarce in the orchestration era: how to systematically confirm the work you delegated is actually right; it connects directly to Agent in Practice. By the end you know what to watch, what to send out, and how to be sure you weren't burned.
- 阅读页 / Web: https://library.aklman.com/books/claude-code
- AI-readable markdown: https://library.aklman.com/books/claude-code/llms.md

### Cursor · 一台元工具 / Cursor · One Meta-Tool

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.07
- 11 章 · 102 分钟 · TECH / PRACTICE
- Cursor 不只是写代码的地方。Agent、集成终端、codebase index、Rules、AGENTS.md、MCP、Cloud Agents、CLI、Bugbot 凑在一起，是一台元工具 —— 一台电脑上的很多活，从写代码到调研、写作、数据清洗、文件批处理、自动化，都能在一个窗口里收束。这本书不列全部功能，只挑覆盖八成场景的两成核心：每一项答清楚「省什么、什么活用、什么时候不如专用工具」。倒数第二章给 Pro / Pro+ / Ultra / Teams / Enterprise 选型、与 Claude Code / Codex / Copilot / Windsurf / Warp 的分工，并划清 Computer Use boundary。读完你知道哪些活该留在 Cursor，哪些该切走。
- Cursor isn't only where you write code. Agent, integrated terminal, codebase index, Rules, AGENTS.md, MCP, Cloud Agents, CLI, and Bugbot add up to a meta-tool — much of a machine's work, from code to research, writing, data cleaning, file batching, and automation, gathered into one window. This book picks the 20% that covers 80% of real work: what each part saves, when to use it, and when a dedicated tool wins. The penultimate chapter handles Pro / Pro+ / Ultra / Teams / Enterprise, the trade-offs against Claude Code / Codex / Copilot / Windsurf / Warp, and the Computer Use boundary.
- 阅读页 / Web: https://library.aklman.com/books/cursor
- AI-readable markdown: https://library.aklman.com/books/cursor/llms.md

### Anthropic · 一份会员该用的那部分 / Anthropic · The Part of a Membership Worth Using

- 语言 / Language: 中英双语 · ZH · EN
- 首发 / Published: 2026.07
- 12 章 · 107 分钟 · TECH / PRACTICE
- 一份 Claude 会员横跨网页、Projects 与 Memory、Skills、连接器、Artifacts 与 Claude Design、Cowork、Research、浏览器、终端与 API。这本书不列全部功能，也不停在讲解 —— 它只挑能覆盖八成真实场景的那两成核心能力，每一章带你亲手做出一件东西，做完就留在你的账号里。第一章给自己立一条基线，倒数第二章按你自己记的账选订阅档，最后一章是一张逐项验收单。读完你手上有十二件真东西，而不是十二个知道的名词。
- A Claude membership spans web, Projects and Memory, Skills, connectors, Artifacts and Claude Design, Cowork, Research, the browser, the terminal, and the API. This book does not list every feature, and it does not stop at explaining — it picks the 20% of capabilities that cover 80% of real work, and every chapter has you build one thing that stays in your account afterward. Chapter one sets your baseline; the penultimate chapter picks a tier from the ledger you kept yourself; the last is an item-by-item sign-off. You end with twelve real things, not twelve nouns you recognize.
- 阅读页 / Web: https://library.aklman.com/books/anthropic
- AI-readable markdown: https://library.aklman.com/books/anthropic/llms.md

## 授权 / License

本书库内容遵循 CC BY-NC-ND 4.0：允许免费阅读、AI 摘读、引用与转发，须保留作者署名 (Aklman) 与原文链接；禁止商业再发布与衍生改写。
Library content is licensed under CC BY-NC-ND 4.0 — free to read, quote, and feed to AI; commercial reuse and derivative works are not permitted.
License: https://library.aklman.com/license

Aklman · Library · https://library.aklman.com/