--- name: lingtai-ai/textbook-distillation source: https://app.decimal.ai/s/lingtai-ai-textbook-distillation@1/SKILL.md source_sha256: 62b453acaf70 --- # Textbook Distillation — Self-Paced Learning Tracks You are helping a human teach themselves from a textbook (or a long technical document, lecture transcript, or paper set) without a live instructor. Your job is to **distill** the source into a structured learning track and ship **self-contained HTML lecture notes** in the style the human asks for: extract the concepts, structure, and worked logic, then re-explain them in your own words with your own examples. Distillation is never reproduction of the author's text. Read [Source limits & safety](#source-limits--copyright--safety) before you intake anything — those boundaries shape every later step. ## When to use The frontmatter `description` carries the triggers and anti-triggers. Two boundaries deserve restating here: an ask framed as "summarize the whole book so I can skip buying it" or "paste me the chapter" crosses the copyright line — decline it and offer a legitimate distillation instead; and material the human cannot point to a legitimate source for is out of scope entirely. Both are spelled out under "Source limits, copyright & safety" below. ## Workflow at a glance ``` intake → chapter map → lesson plan → (per lesson) HTML lecture notes → review loop ``` Work the phases in order. Do not jump to generating HTML before the human has seen and approved the chapter map and lesson plan — that approval is what keeps the track aligned to *their* goal, not your guess at it. This is a multi-round, human-in-the-loop flow: **talk to the human through the active user-facing channel (mail/email or the relevant chat bridge), not internal scratch/text output**, because each phase needs their input. ## Phase 1 — Intake the source Goal: know exactly what you are distilling and confirm you are allowed to. 1. **Identify the source precisely.** Title, author, edition, and the format you have access to (a PDF in the project, a URL, the human's own notes, a file the human pasted). Record the path or link — you will cite it in every artifact. 2. **Confirm authorization and scope.** Ask the human: do they own/have legal access to this material? Which parts do they want covered — whole book, a range of chapters, one topic? Do not proceed on material they cannot point to a legitimate source for. 3. **Capture the learning goal.** Why are they studying it — exam, project, curiosity, teaching it onward? Their level (beginner / refresher / advanced)? Time budget (one weekend vs. a semester)? These set lesson granularity. 4. **Capture style constraints early** (see [Style constraints](#style-constraints-from-the-human)). Collect them now so the lesson plan and the HTML are designed for them, not retrofitted. If you only have a partial source (e.g. a table of contents but not the body), say so explicitly — you will distill structure from the ToC and re-teach concepts from your own knowledge, clearly labeled, rather than inventing the book's specific treatment. ## Phase 2 — Build the chapter map A **chapter map** is the skeleton: the source's structure plus what each unit is *about* and what it depends on. It is the contract the human approves before any lesson is written. For each chapter / major section, capture: - **Unit** — chapter or section title and number. - **Core concepts** — the 3–8 ideas a learner must walk away with. - **Prerequisites** — which earlier units (or outside knowledge) it assumes. - **Difficulty / weight** — rough effort, so the track can be paced. - **Source location** — page range or section anchor, for citation. Render the map as a table and send it to the human. Ask: *Is this the right scope and ordering? Anything to drop, merge, or go deeper on?* Iterate until they approve. The map may reorder or merge the book's units to fit the learning goal — note where you deviate from the book's own order and why. ## Phase 3 — Draft the lesson plan Turn the approved chapter map into a sequence of **lessons**. A lesson is one sitting's worth of learning — usually one chapter, or a slice of a dense one. For each lesson define: - **Title and one-line objective** ("After this lesson you can …"). - **Concepts covered** — drawn from the chapter map, scoped to one sitting. - **Prerequisite lessons** — what must come first. - **Worked example(s)** — the concrete problem you will walk through. - **Exercises** — 2–5 practice problems, easy → hard. - **Checkpoint** — 3–6 self-check questions (with answers, collapsible) that tell the learner whether they can move on. - **Estimated time.** Send the lesson plan for approval. This is the last gate before you generate HTML, so make sure granularity, ordering, and exercise depth match what the human wanted. ## Phase 4 — Generate the HTML lecture notes Now produce the deliverable: **one self-contained HTML file per lesson** (or a single file for a short track), in the human's style. **Standalone-HTML mechanics are owned by the `html-report` reference** — the skeleton, artifact hygiene, the validation checklist, and, critically, **MathJax setup**, since lecture notes for technical books almost always contain equations. Read `~/.lingtai-tui/utilities/swiss-knife/reference/html-report/SKILL.md` (or load `swiss-knife` and route to `html-report`) and follow it rather than re-deriving it here. Start from the bundled template at `~/.lingtai-tui/utilities/textbook-distillation/assets/lesson-template.html`, which already wires MathJax, a print stylesheet, collapsible checkpoint answers, callout boxes, CSS variables for theming, and the lesson structure below. Copy it, apply the human's style, fill in the content. Each lesson must carry, in order: 1. **Header** — title, objective, source citation (title, author, edition, page range), generation timestamp (`date -u +%Y-%m-%dT%H:%M:%SZ`), and the provenance caveat ("Distilled study notes — re-explained from the source, not a reproduction of it."). 2. **Prerequisites** — a short "before this lesson" note linking prior lessons. 3. **Concept exposition** — the teaching, in *your own words*, terms defined in callout boxes, with diagrams / tables / analogies as the style calls for. 4. **Worked example(s)** — step through the reasoning explicitly. 5. **Exercises** — numbered, increasing difficulty, solutions hidden behind a `
` element so the learner attempts first. 6. **Checkpoint** — self-check questions with collapsible answers; passing them is the signal to advance. 7. **Footer** — "next lesson" link and the source citation again. Run the `html-report` validation checklist against every file you write. Write them to a project-local path the human can find (e.g. `study//lesson-NN.html`), never `/tmp`. ## Phase 5 — Review loop Send the first lesson and ask the human to react to **both** the teaching and the style before you batch the rest — it is far cheaper to adjust the template once than to regenerate twenty files. Then proceed lesson by lesson (or in small batches), checking in. Offer an index page linking all lessons once the track is built. ## Style constraints from the human Capture these in Phase 1 and apply them in Phase 4. If the human is vague, propose a default and confirm: - **Theme / palette** — light or dark, accent color, any brand colors. The template exposes these as CSS variables at the top of `