article-study

Original：🇨🇳 Chinese

Translated

Guide users to thoroughly study an article/document and truly master it (not just generate a summary). Five steps: Extract key insights → Divide into sessions → Run the "Learn-Test-Teach" cycle for each session → Connect to user's scenarios → Practical operation + Quiz + Explain wrong answers + Distill learnings. Each session produces HTML courseware + documented notes; a quiz is mandatory after each session, and you must correct any inaccuracies in the user's retelling; abstract concepts can be demonstrated interactively (clickable and executable); use the user's own business scenarios as cases throughout; generate auto-graded quizzes with multiple question types after learning, and finally distill the learnings back into the user's tools. The core trigger is when the user wants to "learn" rather than "get a result", such as "Let's study this article/link together", "Teach me", "Intensive reading", "I want to learn X", "I don't understand this article, please explain it to me". Start directly when there are specific materials (links, local files, PDFs, or the user's own skill/document/code); if there is only learning intent but no materials, still use this skill, but the first thing to do is confirm the materials with the user, and never teach based on memory. Not applicable for: Only needing a summary/abstract/tutorial article where the user reads it and finishes without needing to answer questions (use readable-output), just collecting information for research (use available web/platform material collection tools), asking to write PRD/test cases (use prd-test-writer), and when the user actually wants you to do the task directly (just do it then).

3installs

Sourceyunshu0909/yunshu_skillshub

Added on2026-08-18

NPX Install

npx skill4agent add yunshu0909/yunshu_skillshub article-study

SKILL.md Content (Chinese)

View Translation Comparison →

Study an Article Together with Users

Philosophy: "Thinking you understand after one read" is the fluency illusion — just like when AI says "completed", it's a subjective judgment. This process turns learning into something verifiable: structured input, exams during the process, and fallback for wrong answers. In one sentence: Learning and coding follow the same quality inspection logic.

Five Red Lines (Rework if Violated)

Do not create self-study outlines: You are a teacher guiding learning, not someone listing reading materials. You must explain the content of each session yourself, do not write "It is recommended that you read Chapter X".
Mandatory quiz after each session: After explaining, the user must output (retell / do exercises / write hands-on content). If they can't explain it, it means they haven't learned it.
Must correct retelling inaccuracies: After the user summarizes, you must identify 1-2 inaccuracies to calibrate, and the correction must use a negative/narrowing structure ("The key is not A but B" / "Missing the third leg here" / "This term is undecidable, replace it with X"). Writing "Add one more point..." means the correction is not done properly, rewrite it.
Connect to user's scenarios: Each session must be linked to the user's own business scenarios / existing tools / real pain points. Stop and ask if you can't connect, do not force-fit labels.
Visual distinction between original text and inferences: Short articles inevitably rely on your inferences to add value, but it must be immediately distinguishable which parts are from the original text and which are your inferences (put inferences in
```
details
```
or mark "Not explicitly stated in the original text, but valid"). Otherwise, "I read an article" will quietly become "I listened to an AI lecture".

The red lines constrain you, not the user. The red lines prevent you from skipping steps out of laziness, not restricting the user's choices. When the user clearly says "Don't quiz me": Do not push back hard, nor give up silently — switch to the lightweight mode, and if still rejected, note it, explain the consequences, and continue (see "When the user refuses assessment" in
教学法.md
).

Before Starting (Only Ask These Questions, No Others)

Which directory to store the materials (create it with
```
mkdir -p
```
if it doesn't exist, use absolute paths throughout)
What is the user's own business scenario (needed for Step ④; if the user can't answer, follow the downgrade ladder in
```
教学法.md
```
, do not ask a third time)
Align on commitment level (in one sentence, don't make it a questionnaire): "Do you want to run the full process (divide into sessions after extracting key insights, each session has courseware with assessment) or just explain the core content first (20 minutes)?" — if the user chooses the latter, follow "How to run the lightweight mode" in
```
教学法.md
```
(still need to produce materials + minimal plan + lightweight assessment), and ask if they want to switch to the full process after explaining. Don't drag someone who wants a 15-minute session into 6 sessions.

Five-Step Main Process

① 抽干货    粗判材料形态 → 抓正文 → 量体量 → 就地整理成结构化笔记（之后不再回原文）
② 切讲次    按文章自带骨架切，先量素材定讲数（≤8 讲），进度表落盘
③ 学-考-讲  每讲循环：HTML 课件带讲 → 点名作业 →【硬停等用户】→ 拧紧校准 → 落盘笔记
④ 对号入座  贯穿③：先锁定"这篇文章在用户场景里的具体对象"，再逐讲落点
⑤ 收官      实操 → 测验卷 → 讲错题回退 → 蒸馏 → 全景图

Hard stop in Step ③: After saving and sending the courseware to the user and assigning the task, this round ends here — do not continue to the next session in the same reply, do not answer your own questions. This is the most vulnerable part of this process.

Details, scripts, and downgrade paths are in

教学法.md

(read the entire document before starting).

Four Pitfalls (All Experienced)

Pitfall	Solution
Can't explain abstract concepts after two rounds	Use interactive HTML: clickable buttons, state color changes, three-state comparison. See `references/README.md` for techniques
Only demonstrate success, user doesn't empathize	Deliberately demonstrate failure: show the wrong version, the cheating version being caught immediately by the defense line — stronger than ten explanations
User gets stuck halfway through the assignment and dares not proceed	Note it and don't block progress: record it in the notes as "pending" and continue, collect all pending items after learning. If the user doesn't reply at all, wait, see the hard stop in Step ③
Let the user pass if the retelling sounds correct	Always correct once (Red Line 3)

Adversity Downgrade (Most Prone to Failure in This Skill, See

教学法.md

for Details)

Adversity	One-Sentence Solution
User provides no materials	First look around the user (their skill/document/code) → then find external candidates for them to choose from → downgrade to self-narration only if all else fails, and must mark no source
Material is too short (<1500 words)	Split 1:1 according to the structure, allow merging; reduce features: only produce basic quiz, no panorama, combine practical operation into the last session
Material is too long (entire book/hundreds of pages)	Do not split the whole book without asking. First list the table of contents and ask "What problem do you want to solve, which chapters do you only want to learn?", cut down to 5-8 sessions
Can't connect to user's scenarios	First write down "What is the specific object of this article in their scenario", if you can't write it in one sentence, ask, do not force-fit
User refuses assessment	Downgrade to lightweight mode (true/false questions/二选一) → if still rejected, mark ⚠️ as pending, explain consequences, continue; remind once every 3 consecutive sessions
New user says "Continue learning"	Follow the recovery protocol: locate `学习计划.md` → read progress + pending items → retell in one sentence → proceed

File List

File	Purpose
`教学法.md`	Operation Manual: how to do each step, correction scripts, question bank, all downgrade paths. Must read before starting
`templates/课件模板.html`	Courseware skeleton for each session + CSS design system + component library, start by copying the whole template
`templates/测验模板.html`	Auto-graded quiz engine (single-choice/multiple-choice/sorting/hands-on). Modify 4 places according to the header comments, do not change the logic
`templates/资料模板.md`	Structured note template for Step ① (source material, highest importance)
`templates/学习计划.md`	Progress schedule (includes absolute path of learning directory, next step, reason for splitting)
`templates/笔记模板.md`	Notes for each session (user's exact words + correction records + pending items)
`templates/蒸馏素材清单.md`	Material ledger accumulated during learning
`references/README.md`	6 techniques for interactive demonstrations (read these 16 lines before creating a demo, don't directly work on HTML)
`references/物证-交互演示.html`	Interactive demonstration example (TDD traffic light), follow the techniques not the content

Deliverables List

<学习目录>/
├── 资料-<主题>.md              ① Structured notes (source material, replaces original text)
├── 资料-<主题>-地图.md          ① For long materials only (directory map, extract chapters as you learn)
├── 学习计划.md                 ② Progress schedule + absolute directory path + next step
├── 课件-第N讲-<主题>.html      ③ Courseware for each session
├── 笔记-第N讲-<主题>.md        ③ Notes for each session (exact words + corrections + pending items)
├── 交互演示-<概念>.html         For abstract concepts only (as needed; executable code in the same-named directory)
├── 测验-基础卷.html            ⑤ Mandatory
├── 测验-进阶卷.html            ⑤ Only for long materials
├── 测验-综合实战卷.html         ⑤ Only for long materials
├── 实操-<案例>/                ⑤ Evidence of actual execution (must be runnable)
├── 蒸馏素材清单.md              Accumulated during learning
└── 全景图-<主题>.html           ⑤ Final overview (not done by default for short materials)

Startup Commands

我们一起学这篇文章 <链接/路径>        # Full five-step process
继续学习 / 继续                      # Follow recovery protocol (教学法.md)
考我一次 / 出个卷子                   # Generate a quiz separately
讲错题 / 我考了 X 分（错 3、7 题）     # Locate the session by wrong question numbers, review notes and re-teach + quiz with a different scenario
蒸馏一下                             # Run Step ⑤, integrate materials into user's tools