Forge yourself into an expert.
English | 中文
A ZCode agent skill that forges any repo, topic, or document pile into a personalized, interactive HTML textbook — written at the depth you actually need.
Works with any agent that loads skills from ~/.agents/skills/.
- Calibrates before every chapter — asks you 2–3 real scenario questions ("here's a snippet — what does it print, and why?"), scores your answers, then picks the depth: skim what you've mastered, teach fully what you haven't.
- Builds from any source — a local repo (it reads the actual code, not just the README), a domain topic, or docs & papers (URLs or files) — in any combination.
- Writes a real book, not an essay — one self-contained HTML file per chapter plus an
index.htmldashboard, with learning objectives, recaps, and a test per chapter. - Draws, doesn't describe — architecture maps, ER sketches, and sequence flows as pure inline SVG in the theme's palette.
- Ships runnable demos — small per-book widgets (a CSRF attack simulator, a token slider, a step-by-step loop printer) that run in the page.
- Maps the knowledge (optional) — a concept graph on the dashboard: concepts light up as you pass chapter tests, and clicking a node deep-links to the exact sections that teach it.
- Tests every chapter — mixed multiple-choice (single + multi), fill-in, and short-answer with self-check; ≥ 80% = learned enough to move forward.
- Fails softly — wrong answers highlight with a "review this section" link to the exact anchor; nothing locks, retake anytime.
- Tracks progress — scores and chapter status persist in
localStorage, per book and per language. - Speaks your languages —
en/,zh/,ja/, … as parallel sibling folders; switching language is a hyperlink, no runtime switcher. - Light & dark themes — a toggle in the topbar, preference remembered in
localStorage; dark is the default. - Stays yours — static HTML files on your disk: no server, no build step, no account, no internet. Works straight from
file://.
From a real book forged from the open-connector-book repo — 13 chapters on gateway architecture and OAuth, in English and Chinese:
 |
 |
| The dashboard — every chapter as a card: ✓ passed · • below 80% · ○ not taken. | Knowledge graph — the book's concepts and relations; click a node to see where it's taught. |
 |
 |
| Inline SVG diagrams — layered architecture, ER sketches, flows — drawn, not described. | Interactive demos — run a concept in motion; this one simulates a CSRF attack and the state defense. |
 |
 |
| Test feedback — every wrong answer shows the correct answer, the rationale, and a deep link to re-read. | The 80% soft gate — miss it and nothing locks; re-read the highlighted sections and retake. |
Knowledge graph — progress view
| Mechanism | What it does |
|---|---|
| Pre-chapter diagnostics | 2–3 scenario questions per chapter → knowledge-point scoring → a depth band: skim + advanced, targeted, or full. |
| Tests weighted to weak points | Chapter tests probe what you got wrong in the diagnostic, not what you already knew. |
| 80% soft gate | Passing marks a chapter learned; missing highlights exactly which sections to re-read. No locks, no blocked content. |
| Language-neutral anchors | Section ids (sec-…) are identical across every translation, so review links and graph references survive any language. |
localStorage records |
One key per book × language × chapter — powers the dashboard, the graph's progress view, and survives reloads. |
| Regenerable | Come back to the agent anytime to add, revise, or deepen chapters; enhancements land in the template so the next book inherits them. |
| Goal | Outline driven by | You get |
|---|---|---|
domain-expert |
The field's standard conceptual progression | Foundations → core mechanisms → edge cases and pitfalls, calibrated chapter by chapter. |
repo-expert |
The repo's actual structure (auto-detected: DB schema, core services, data flow, API surface, contributing map) | An onboarding path through the real code — with docs-vs-implementation gaps flagged and probed in the tests. |
0 Analyze the source(s) — repo / topic / materials
0.5 Ask you: which languages, which goal
1 Propose the outline → you approve
2 For each chapter:
ask 2–3 scenario questions → score → pick depth band
write body + SVG diagrams + demos → write the test
3 Assemble the multi-file HTML project
4 Open index.html — done
4.5 Translate into sibling language folders
5 Post-delivery enhancement loop (knowledge graph, dark mode, …)
1 — Install the skill (discovered from ~/.agents/skills/):
git clone https://github.com/JJack27/Forge
cp -r Forge ~/.agents/skills/2 — Just talk to your agent. The skill triggers on natural requests — no command to memorize:
Make me a book on Rust ownership and borrowing. I've written some Rust but I keep fighting the borrow checker.
Answer three things as it works — which languages, which goal, approve the outline — then it forges chapter by chapter, calibrating depth as it goes.
3 — Open the result:
open books/<your-book>/en/index.html # or just double-click it — that's the whole runtimeMore example prompts
| Source | Goal | Prompt |
|---|---|---|
| Topic | domain-expert | "Create a course for me on database indexes. English primary, but I also want Japanese and Spanish versions." |
| Repo | repo-expert | "Onboard me to this codebase at ~/projects/some-repo. I want to actually become an expert on it — DB, services, the works. English and Chinese." |
| Papers | domain-expert | "Build me a study guide on distributed consensus from these two papers: <url1>, <url2>. I know basic distributed systems but Paxos/Raft always confuse me. Chinese." |
| Any | any | "Turn this repo into something I can learn from" · "onboard me to this codebase" · "build me a study guide for Z from these docs" |
A folder of static HTML — yours to keep, host, or zip up and share:
books/your-book/
├── en/ # one folder per language (en is primary here)
│ ├── index.html # dashboard: TOC, progress, knowledge graph
│ ├── 01-foundations.html # one self-contained .html per chapter
│ ├── 02-core-mechanisms.html
│ ├── …
│ └── assets/
│ ├── style.css # light/dark theme via CSS vars (shared across languages)
│ ├── book.js # scoring + nav + demos registry (shared)
│ └── graph.js # knowledge-graph data + engine (optional)
└── zh/ # sibling folder: same structure, same anchors,
# translated prose — language switch is a link
- Everything works under
file://— no server, no build step, no dependencies. - Tests are fully client-side, scored from
data-*attributes; results inlocalStorage. - The knowledge graph renders with Cytoscape.js from a CDN and falls back to a built-in SVG layout offline.
Do I need a server, a build step, or an internet connection?
No. Every chapter is a fully self-contained HTML file that works opened directly via file://. The whole book is a folder you can keep anywhere.
What can it build a book from?
Three sources, in any combination: a local repository (README, tree, key entry points, recent commits), a topic/domain (the field's standard progression), or external materials — docs, tutorials, papers as URLs or files. With multiple sources, the topic is the skeleton and the repo/materials supply worked examples — unless the goal is repo-expert, where the repo's own structure drives the outline.
What happens if I fail a chapter test?
Nothing locks — the 80% gate is deliberately soft. The chapter is marked "below 80%" on the dashboard, every wrong question is highlighted with a link to the exact section to re-read, and you can retake anytime.
Which agents can use this skill?
Any agent that loads skills from ~/.agents/skills/ — it's an open-format skill (a SKILL.md plus reference docs and a book template).
Can it write the book in Chinese / Japanese / Spanish / …?
Yes. Ask in the language you want, or name several — e.g. "English primary, plus Chinese and Japanese." Each language gets its own complete sibling folder, and progress is tracked per language.
I just want the whole book generated without the questions.
Say so — it will honor it (with a one-time warning that depth calibration will be coarse) and default every chapter to medium depth.
Forge/
├── SKILL.md # the workflow — the skill's entry point
├── references/
│ ├── source-analysis.md # extracting structure from repo/topic/materials
│ ├── assessment-questions.md # writing & reading the per-chapter diagnostics
│ ├── test-design.md # test size, question types, the 80% rule
│ ├── project-structure.md # HTML layout, data-* schemas, SVG & demos
│ ├── knowledge-graph.md # concept-graph authoring pipeline + verification
│ └── i18n.md # sibling-folder convention & translation rules
└── assets/
├── book-template/ # the template project each book starts from
└── screenshots/ # README screenshots (from the demo book)
English | 中文
把自己锻造成专家。
一个 ZCode agent skill：把任何代码仓库、主题或文档资料锻造成一本个性化的交互式 HTML 教科书 —— 篇幅深度按你真实水平定制。
任何能从 ~/.agents/skills/ 加载 skill 的 agent 都可以使用。
- 每章动笔前先校准 —— 先问你 2–3 个真实场景题（"这段代码会打印什么？为什么？"），按你的回答定深度：已掌握的略过，薄弱的讲透。
- 任意来源 —— 本地代码仓库（读真实代码而非只有 README）、领域主题、或文档与论文（URL 或文件），可任意组合。
- 写的是一本书，不是一篇长文 —— 每章一个自包含 HTML 文件，外加
index.html仪表盘；每章有学习目标、回顾和小测。 - 画图，而不是描述 —— 架构图、ER 草图、时序/流程图，全部用主题配色的内联 SVG。
- 内置可运行的 demo —— 每本书自己的小交互组件（CSRF 攻击模拟器、token 滑块、逐步循环打印机），在页面里直接跑。
- 知识图谱（可选）—— 仪表盘上的概念图：通过的章节会点亮对应概念；点击节点直达讲解它的章节小节。
- 每章都有测试 —— 单选、多选、填空、简答自评混合；≥ 80% 即视为学会。
- 温和的失败 —— 答错高亮并附"复习本节"直达链接；不锁任何内容，随时重测。
- 进度持久化 —— 成绩与章节状态存在
localStorage，按书、按语言分别记录。 - 多语言 ——
en/、zh/、ja/… 平行目录；切换语言就是一个链接，无需运行时切换器。 - 浅色 / 深色主题 —— 顶栏一键切换，偏好存在
localStorage；默认深色。 - 完全属于你 —— 磁盘上的静态 HTML：无服务器、无构建、无账号、无网络，
file://直接打开。
（完整截图见上方 Screenshots。）
| 机制 | 作用 |
|---|---|
| 章前诊断 | 每章 2–3 个场景题 → 知识点打分 → 深度档位：略读+进阶 / 针对性 / 完整讲解。 |
| 测试偏重薄弱点 | 章末测试针对你诊断中答错的内容出题，而不是考你已经会的。 |
| 80% 软门 | 达标即标记"已学会"；未达标则精确高亮需要重读的小节。不锁内容。 |
| 语言无关锚点 | 小节 id（sec-…）在所有语言版本中一致，复习链接和图谱引用在任何语言下都有效。 |
localStorage 记录 |
每书 × 语言 × 章节一条记录 —— 驱动仪表盘、图谱进度视图，刷新不丢。 |
| 可再生成**（regenerable）** | 随时回来让 agent 增删、修订、加深章节；增强会沉淀进模板，下一本书直接继承。 |
| 目标 | 大纲依据 | 你得到 |
|---|---|---|
domain-expert |
该领域的标准概念进阶 | 地基 → 核心机制 → 边界与陷阱，逐章校准深度。 |
repo-expert |
仓库的真实结构（自动识别：数据库 schema、核心服务、数据流、API 面、贡献指南） | 一条沿真实代码的上手路径 —— 文档与实现不符之处会被标出并写进测试。 |
0 分析来源 —— 仓库 / 主题 / 材料
0.5 问你：语言、目标
1 提出大纲 → 你确认
2 逐章进行：
问 2–3 个场景题 → 打分 → 定深度档位
写正文 + SVG 图 + demo → 出测试题
3 组装多文件 HTML 项目
4 打开 index.html —— 完成
4.5 翻译成各语言的兄弟目录
5 交付后增强循环（知识图谱、深色模式……）
1 —— 安装 skill（从 ~/.agents/skills/ 自动发现）：
git clone https://github.com/JJack27/Forge
cp -r Forge ~/.agents/skills/2 —— 直接对 agent 说话，skill 会在自然请求时自动触发，无需记命令：
给我做一本 Rust 所有权和借用机制的书。我写过一些 Rust，但总是和 borrow checker 打架。
过程中回答三件事 —— 要哪些语言、哪种目标、确认大纲 —— 之后它逐章锻造、边写边校准深度。
3 —— 打开成果：
open books/<your-book>/zh/index.html # 或者直接双击 —— 这就是全部运行时一个静态 HTML 文件夹 —— 保留、托管、打包分享都随你（结构见上方 What You Get）：
- 一切都能在
file://下工作 —— 无服务器、无构建、无依赖。 - 测试完全在浏览器端评分（基于
data-*属性），结果存localStorage。 - 知识图谱优先用 CDN 的 Cytoscape.js 渲染，离线时自动回退到内置 SVG 布局。
需要服务器、构建步骤或联网吗？
都不需要。每章都是一个完全自包含的 HTML 文件，file:// 直接打开即可。整本书就是一个可以放在任何地方的文件夹。
可以用什么来源生成书？
三类来源、任意组合：本地仓库（README、目录树、关键入口、近期提交）、主题/领域（该领域的标准进阶路径）、或外部材料 —— 文档、教程、论文（URL 或文件）。多来源时以主题为骨架、仓库/材料提供实例；若目标是 repo-expert，则由仓库自身结构驱动大纲。
章节测试不及格会怎样？
不会锁任何东西 —— 80% 是刻意设计的软门。仪表盘上该章标记"低于 80%"，每道错题高亮并链接到应重读的小节，随时可重测。
支持中文 / 日文 / 西班牙文…… 吗？
支持。用对应语言提需求，或指定多种 —— 比如"英文为主，另加中文和日文"。每种语言一个完整目录，进度按语言独立记录。
我不想答题，直接生成全书行吗？
可以，说一声即可 —— 它会照办（一次性提醒：深度校准会比较粗），所有章节默认中等深度。
Forge/
├── SKILL.md # 工作流 —— skill 的入口
├── references/ # 各环节的详细参考文档
│ ├── source-analysis.md # 从仓库/主题/材料中提取结构
│ ├── assessment-questions.md # 章前诊断题的编写与判读
│ ├── test-design.md # 测试规模、题型、80% 规则
│ ├── project-structure.md # HTML 布局、data-* 模式、SVG 与 demo
│ ├── knowledge-graph.md # 知识图谱的数据编写流程与验收清单
│ └── i18n.md # 兄弟目录约定与翻译规则
└── assets/
├── book-template/ # 每本书起步用的模板项目
└── screenshots/ # README 截图（来自演示书）