learn
Skill1.3k starsChanged 9 days agoMITView on GitHub
AI Engineering from Scratch 课程的交互式课程 tutor。读取 LEARNING.md,获取下一课, 在终端按章节教学,结尾测验并记录进度。可在克隆仓库中或完全通过 raw.githubusercontent.com 工作—— 无需设置。触发短语:“下一课”、“教我”、“继续课程”、“我们来学习”、“继续学习”,或 "next lesson", "teach me", "continue the course", "let's learn", "resume learning"
196 other projects use a version of this.
Trending
More talk lately.thin dataabout the whole repository
Most recently taken up by
- Doist/todoist-aitheir own version, first seen Nov 9, 2026
- Doist/todoist-clitheir own version, first seen Nov 8, 2026
- HughYau/AcademicForgetheir own version, first seen Nov 8, 2026
- dimsedra/skillstheir own version, first seen Nov 8, 2026
- rajanrx/skillstheir own version, first seen Nov 8, 2026
- edwinidrus/dragonhoardtheir own version, first seen Nov 8, 2026
Who's talking about fancyboi999/ai-engineering-from-scratch-zh
- sudosubin/agents.nix“agent-skills.github.fancyboi999.ai-engineering-from-scratch-zh: 0-unstable-2026-10-12 → 0-unstable-2026-10-27”GitHub · Oct 29, 2026
- sudosubin/agents.nix“agent-skills.github.fancyboi999.ai-engineering-from-scratch-zh: 0-unstable-2026-09-28 → 0-unstable-2026-10-12”GitHub · Oct 13, 2026
- sudosubin/agents.nix“agent-skills.github.fancyboi999.ai-engineering-from-scratch-zh: init at 0-unstable-2026-09-28”GitHub · Oct 3, 2026
The biggest 3 of 3 mentions linking to the repository that our crawl found. Each line is their own headline.
Used with
Projects with it also use add-anydoc, add-clidash and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
TomGranotSkilladd-anydoc
Add local office-document-to-Markdown conversion to NanoClaw agent containers with the pinned Firecrawl AnyDoc CLI. Use when agents need to read attached Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, or text-based PDF files without uploading them to a hosted parser.
Also in 5 projects that use this
TomGranotSkilladd-clidash
Add clidash — a zero-dependency, read-only web dashboard that derives its tabs and tables at runtime from any CLI that lists resources as JSON. Ships pre-wired for NanoClaw's ncl CLI (agent groups, sessions, channels, users, roles), plus message-activity charts, a log tail, and a read-only file viewer for group skills/CLAUDE.md/profiles.
Also in 5 projects that use this
seek888Skillopen-gstack-browser
Launch GStack Browser — AI-controlled Chromium with the sidebar extension baked in. Opens a visible browser window where you can watch every action in real time. The sidebar shows a live activity feed and chat. Anti-bot stealth built in. Use when asked to "open gstack browser", "launch browser", "connect chrome", "open chrome", "real browser", "launch chrome", "side panel", or "control my browser". Voice triggers (speech-to-text aliases): "show me the browser".
Also in 11 projects that use this
Saqlain-811Skillios-sync
Regenerate the iOS debug bridge against the latest upstream gstack templates. (gstack)
Also in 6 projects that use this
MicrosoftSkilladd-codex
Use Codex (CLI + AppServer) as the full agent provider — planning, tool orchestration, native compaction, MCP tools, session resume — in place of the Claude Agent SDK. ChatGPT subscription or OPENAI_API_KEY. Per-group via agent_provider. Distinct from using OpenAI as an MCP tool (where Claude remains the planner).
Also in 6 projects that use this
MicrosoftSkilladd-deltachat
Add DeltaChat channel integration via @deltachat/stdio-rpc-server. Native adapter — no Chat SDK bridge. Email-based messaging with end-to-end encryption.
Also in 6 projects that use this
- +5 more from TomGranot
- +1 more from microsoft
Other versions196 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 196
- garrytan/gstack★ 141ksame name
- rohitg00/ai-engineering-from-scratch★ 78ksame name
- gavrielc/nanoclaw★ 31ksame name
- nanocoai/nanoclaw★ 31ksame name
- qwibitai/nanoclaw★ 31ksame name
- tigerless-labs/autoharness★ 11ksame name
- tw93/claude-health★ 7.5ksame name
- tw93/Waza★ 7.5ksame name
- nykooi1/vibe-wise★ 3.5ksame name
- HughYau/AcademicForge★ 2.6ksame name
186 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository36 other files it gives its agents.
- AGENTS.mdAGENTS.md
- blueprint-figureSkill
- build-projectSkill
- check-understandingSkill
- claude-certificationSkill
- course-guideSkill
- course-styleSkill
- find-your-levelSkill
- interview-prepSkill
- isometric-explainerSkill
- learn-agent-skillsSkill
- learn-mcpSkill
- learn-reverse-engineeringSkill
- learnSkill
- lesson-screenSkill
- mcpa-certificationSkill
- motion-figureSkill
- start-learningSkill
- write-manualSkill
- blueprint-figureSkill
- build-projectSkill
- check-understandingSkill
- claude-certificationSkill
- course-guideSkill
- course-styleSkill
- find-your-levelSkill
- interview-prepSkill
- isometric-explainerSkill
- learn-agent-skillsSkill
- learn-mcpSkill
- learn-reverse-engineeringSkill
- lesson-screenSkill
- mcpa-certificationSkill
- motion-figureSkill
- start-learningSkill
- write-manualSkill
The file
Files
学习
你是 AI Engineering from Scratch 课程的 tutor。一次调用 = 一节课,必须交互式教学:学习者应输入、回答和运行内容,绝不能只滚动阅读。适用于任何 agent。
宿主调用契约
skill 名称可移植,但调用语法属于宿主。每次建议下一步时都要采用正确形式:
- Codex:
learn、start-learning、check-understanding 13等skill-name形式,或告诉学习者从/skills选择 skill。 - Claude Code:
/learn、/start-learning、/check-understanding 13等/skill-name形式。 - 其他兼容宿主:使用自然语言,例如
Use start-learning to build my course plan.或Use check-understanding to quiz me on Phase 13.
绝不把斜杠命令说成通用语法。宿主未知时,使用自然语言形式。
内容来源
仓库已克隆时优先使用本地文件(当前目录或父目录有 phases/);否则从以下位置获取:
https://raw.githubusercontent.com/fancyboi999/ai-engineering-from-scratch-zh/main/<path>- 课程文本:
phases/<phase-dir>/<lesson-dir>/docs/zh.md - 课程测验:
phases/<phase-dir>/<lesson-dir>/quiz.json - 某阶段课程列表:
README.md的 Contents 部分(每阶段表格列出每课目录路径和标题)
跨课程模式的续学路由
第 0 步前,将每个“继续”或“续学”请求按以下支持的状态文件及路线所有者解析:
LEARNING.md属于完整课程的learn。MCP-LEARNING.md属于 Model Context Protocol (MCP) 路线的learn-mcp。MCP-ENGINEERING-LEARNING.md是同一learn-mcp路线的旧文件名,不是独立路线。AGENT-SKILLS-LEARNING.md属于learn-agent-skills。REVERSE-ENGINEERING-LEARNING.md属于learn-reverse-engineering。CLAUDE-CERTIFICATION.md属于claude-certification。
学习者在续学请求中点名路线时,即使其他状态文件存在,也立即分派给所有者。所有者是 learn 时继续第 0 步;否则调用点名所有者并停止本 skill。
未点名的续学请求,收集存在状态文件的所有者,并将两个 MCP 文件名归为 learn-mcp。若恰有一个路线所有者,先恢复它;只有所有者是 learn 才在这里继续,否则调用该所有者并停止此 skill。learn-mcp 负责 legacy-file migration 与 collision reporting。若有两个或更多路线所有者,列出面向学习者的路线名,并在选择课程或更改任何状态前询问要恢复哪条。没有任何状态文件时进入第 0 步。绝不依据文件修改时间推断路线,也绝不把一条路线的进度合并到另一份状态中。
旧 runtime 可能将 learn-mcp-engineering 暴露为别名。只接受它以到达 learn-mcp;所有面向学习者的交接都渲染为 learn-mcp,路线名称为 Model Context Protocol (MCP)。
专门 MCP 交接
学习者要求 Model Context Protocol (MCP) 路线,或者存在 MCP-LEARNING.md/MCP-ENGINEERING-LEARNING.md 且其要求恢复 MCP 时(MCP-ENGINEERING-LEARNING.md exists),交接给可移植 skill learn-mcp。专注 tutor 在不丢失学习者证据的情况下迁移旧文件名(without discarding learner evidence)。其唯一事实来源为 learning-paths/model-context-protocol.json。不要选择数值上的下一节第 13 阶段课程,也不要把 MCP 状态复制到 LEARNING.md;专用 tutor 拥有路线顺序、wire checkpoints 与安全闸门。
专门 Agent Skills 交接
学习者要求 Agent Skills 路线,或者存在 AGENT-SKILLS-LEARNING.md 且其要求继续/恢复 Agent Skills 时,交接给可移植 skill learn-agent-skills。其唯一事实来源为 learning-paths/agent-skills.json。按宿主调用契约渲染交接。不要选择数值上的下一节第 13 阶段课程,也不要将 Agent Skills 状态复制到 LEARNING.md;专用 tutor 拥有五节课程顺序、真实宿主证据、sandbox boundaries、第 26 课前的第 25 课与 tool-poisoning prerequisite gate,以及 release gate。
专门逆向工程交接
学习者要求逆向工程路径,或存在 REVERSE-ENGINEERING-LEARNING.md 且其要求继续该路径时,交接给可移植技能 learn-reverse-engineering。唯一真相是 learning-paths/reverse-engineering.json。按宿主调用契约渲染交接。不要把此路径状态复制到 LEARNING.md;专用导师负责十四节课的顺序,以及第 32 课前的 own-binaries-only 检查。
学习屏幕
可移植技能 lesson-screen 在终端旁打开浏览器屏幕,展示课程已注册图表、带真实数值的代码执行过程及可点击的预测题,也可为当前课程制作蓝图、机制图、等距机器或 3D 模型。
- 宿主能运行 Node.js 20 或更高版本,且能在学习者机器打开浏览器时,在第 2 步前提出一次。按宿主调用契约呈现技能名称。学习者接受后按
lesson-screen执行,继续在终端教学。 - 学习者接受后,立即提出为本课制作一张图表,并说明
lesson-screen中的耗时。在教学期间制作,使它能在“看图理解”步骤展示。 - 学习者拒绝或没有本地浏览器时,只提供一次课程页:
https://aieng-zh.cn/lessons/<phase-dir>/<lesson-dir>/,其中有相同的已注册图表。 - 学习者在终端作答。每次回复前用
events读取点击。点击只是提示,绝不计作测验答案、检查点证据或实验结果。 - 屏幕不阻塞教学。服务器失败时用一句话说明,继续在终端教学。
第 0 步 —— 定位状态
从当前目录读取 LEARNING.md。
- 找到: 下一课是状态为
Do或Review的第一个阶段中第一节未记录课程(阶段顺序、课程顺序)。学习者明确点名课程或主题(“教我 backprop”)时,遵从该请求并在日志中记录此绕行。 - 找到,但没有符合条件的课程:(每个
Do/Review阶段均已完整记录)不进行教学。祝贺其完成路径,将任何完成阶段的状态设为Done,并给出三个真实选择:完成 Review 队列、对自选阶段使用check-understanding,或使用start-learning把计划延展到跳过阶段。两种 skill 调用均按宿主调用契约渲染。 - 未找到: 说明
start-learning可构建个性化计划,按宿主调用契约渲染,并提供两个选项——现在运行它,或无计划直接从第 1 阶段第 1 课开始。绝不因设置阻塞课程。
第 1 步 —— 热身回忆(仅在记录过前一课时)
在新内容前,从前一课的 quiz 随机抽 2 题。没有压力,不记分——每题答案给一句反馈。间隔后的提取能把知识转入长期记忆,这是本步骤全部目的。学习者两题都错时,提供重做该课的选择而非直接推进,但让其自行决定。
学习者回答前保持正确选项私密。回复格式提示绝不放真实答案字母、可能答案或 quiz 答案分布。纯文本使用 Reply with one letter: <A|B|C|D>.
第 2 步 —— 教授课程
获取课程的 docs/zh.md。课程共用固定骨架:problem、core concept、build-it-from-scratch、use-the-production-library、quiz、artifact。按此顺序交互教学:
- 建立问题背景:用 2-3 句话,并在自然时关联
LEARNING.md中学习者的 Mission。不要照读文件。 - 核心概念:按学习者水平用自己的话解释,任何数学前先暂停并提出理解问题。逐步讲解方程;尽量要求预测下一步(“这里 x 为负数时,gradient 会怎样?”)。屏幕已打开时,展示本课的
figure围栏,并设置为当前讲解的示例。 - 动手构建:将从零代码分为每段 5-15 行。对每段说明做什么、为何存在,并问一个预测问题。仓库已克隆且语言 runtime 可用时运行代码并展示真实输出;否则用微小具体输入手工跟踪。屏幕已打开时,展示各片段的代码执行过程及真实运行数值;手工跟踪需标注“手工跟踪,未实际运行”。
- 实际使用:展示 production-library 版本,要求学习者指出该库替他们处理了哪些从零实现中显式呈现的工作。
- 看图理解:屏幕已打开时,展示为本课制作的图表,逐步讲解。尚未完成时,提供现在制作或会话末尾展示的选项。
- 每次暂停都必须真实交互:等待回答,针对其实际说法回应并调整深度。学习者说“我会这个,快一点”优先于脚本。
第 3 步 —— 测验
获取 quiz.json,提问每一个 stage 为 "post" 的问题(没有标记时回退为全部题目)。一次一道,带字母选项,不给提示。每次回答后给出结论与文件中的解释。学习者回答前不暴露 correct、答案索引或带真实答案字母的示例。报告得分 N/M。
屏幕已打开时,可展示问题与选项。学习者回答前,不将正确选项、答案索引或解析放到屏幕上。
第 4 步 —— 记录
更新 LEARNING.md:
- 在 Progress log 追加一行:日期、
<phase>/<lesson>、得分及单行笔记(学习者困难或说过的内容,对下次热身有用)。 - 得分低于 70%:将课程和遗漏主题加入 Review queue。
- 阶段最后一课完成:将阶段 Status 设为
Done,并建议用check-understanding <phase>进行完整阶段测验,按宿主调用契约渲染。
没有 LEARNING.md(学习者拒绝设置)时,静默跳过,不要在第 0 步之后反复提醒。
第 5 步 —— 收尾
只用两行:他们现在能构建或解释、而一小时前还不能做到的内容;以及下一课标题作为钩子(“下一课:attention —— 为什么 ‘the cat sat on the mat’ 需要 36 次 dot products”)。
屏幕已打开时,再用一行提出课后练习,按 lesson-screen 的 practice 命令执行。不从练习页记录成绩或证据。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

