textbook-anything

Recommended use: install this skill for use with GPT Pro in the web app.

English | 简体中文

Turn something you want to learn into a textbook you can read, work through, and build on.

Start with a topic, or bring lecture slides, papers, and unfinished notes. textbook-anything arranges the material around the reader's background, explains where the equations come from, and develops examples and connected exercises with full solutions. It produces the requested HTML or PDF with editable source for university mathematics, physics, computing, and engineering. A single chapter is a good place to start; a whole book is within scope too.

Read the 166-page example · Install · Open the skill

Multimodal Learning and Reinforcement Learning: Theory, Architectures, and Implementation. In Chinese. See the example.

A single paper can be the starting point. When it relies on an earlier method, the tutorial follows that reference, explains the necessary foundations, and returns to what the focal paper inherits or changes. The default reader is new to those specialized methods. A short opening exchange establishes the right depth; a GPT Pro run can proceed directly from the available information.

The source-to-artifact workflow now includes video metadata and subtitle acquisition, segmented keyframe inspection, compact source notes, embedded HTML fonts and controlled explanatory animation. The existing LaTeX resources and model-budget tiers remain available. Open the interactive lesson or inspect the agent workflow.

Find a starting point in a new field

The first obstacle is often not knowing which piece is missing. You may follow the code but miss an assumption in the objective, or carry out a calculation without knowing why the field asks that question.

textbook-anything starts with a concrete entry goal and identifies the gaps that block it: field vocabulary, mathematical representations, procedures, conventions, or interpretation of evidence. It teaches the needed background before using it, then moves from a worked example to guided practice and an independent variation.

Learning from scratch still builds on what the reader already knows. An unanswered check remains unassessed; producing a tutorial does not establish that its reader has mastered the material. See the first-order systems example for a complete small unit.

Choose a model-budget tier

Tier	Intended outcome	Work included	Complete rounds
Lite	Resolve one bounded entry problem	Necessary prerequisites, one complete explanation, a worked case, and independent practice with an answer	1
Standard, the default	Learn a coherent topic or chapter	The main background and method, a useful comparison, progressive practice, and a transfer check	2
Deep	Enter a broader field or study a demanding paper	Connected approaches, disagreements and limits, and mathematical or practical perspectives where useful	3

The tiers control research breadth and review effort. Every tier retains necessary reasoning, useful visuals, answers, and artifact checks. Actual usage depends on the material and environment. Model budget is separate from study time, and a GPT Pro run can use lite.

When the requested field is large and the budget is small, plan complete learning units and state what this delivery covers and what remains. Moving to another tier builds on accepted explanations, sources, and real learner attempts. Read the tier policy.

Teaching method

After following a worked example, can the reader solve a related problem with a different assumption? That question guides the explanations, derivations, and exercises.

Purpose	Methodology	Evaluation
Choose the right depth	Ask one or two focused questions about prior knowledge and the intended outcome	Can the goal be expressed as something the reader can explain, derive, or implement?
Establish the learning sequence	Compare textbooks, courses, and related papers; trace where each prerequisite is used	Are necessary ideas introduced in time, and are inherited methods distinct from the contribution?
Explain the material	Begin with a concrete problem, explain the important choices, and interpret the result	Can the reader follow the reasoning without supplying a missing argument?
Develop exercises and solutions	Reuse a setting while changing assumptions, checking limits, constructing counterexamples, or testing numerically	Is the problem solvable, is the solution complete, and does it test transferable understanding?
Prepare the document	Set up the environment, choose useful visuals, and inspect the LaTeX or HTML artifact	Are explanations, equations, and figures clear on screen and in print, and can the source be rebuilt?

Short exercises give readers a chance to check a new idea while it is still fresh. Longer chapter problems connect several ideas. A problem might first ask for a model's solution, then change a boundary condition and ask what follows. Code earns its place by checking a specific mathematical claim. The exercise examples show this in more detail.

Learning goals and actual prerequisite gaps guide how much space each part receives. The selected tier sets the review rounds. Each round includes writing, artifact inspection, corrections, and rechecking across the agreed scope, including foundations, methods, figures, and exercises.

Agent workflow and rendered inspection

SKILL.md is a staged guide for a long-running agent. It settles a teaching brief, keeps coverage, evidence, figure and design ledgers, and splits long material at stated thresholds: a video longer than 20 minutes or with more than 300 subtitle entries, a paper longer than 30 pages, a tutorial of more than four chapters. Segments run in parallel subagents under a fixed return contract, and the main agent integrates them into one teaching sequence. Every exercise is solved and every code example is run before delivery.

HTML output follows the shared visual design rules: six text sizes per width, one reading column, figure labels of 12px or more on a 390px screen, and WCAG AA contrast. scripts/render-audit.mjs checks them on the rendered page and saves screenshots, which the agent then opens one by one:

npm run build:learning
npm run audit:learning

To explain one paper, video, code base or dataset as a single HTML page without a curriculum, use html-report. The two skills share the lesson theme, the builder and the audit.

Install

npx skills add Alchemist-Jo/textbook-anything --skill textbook-anything -g

Choose your agent when prompted. The teaching workflow, writing guidance, and delivery instructions are included; no companion skills need to be installed.

Use

Recommended use: install this skill for use with GPT Pro in the web app.

Describe the topic, reader, model-budget tier, and result you want. Standard is used when no tier is specified:

Use textbook-anything at the standard tier to write a chapter on Fourier analysis for
second-year engineering students who know calculus and linear algebra.
Explain the key derivations, include worked examples, and develop a connected
problem set with separate solutions. Deliver LaTeX and PDF.

You can also start with a specific gap:

Use textbook-anything at the lite tier. I know basic calculus and am new to
dynamical systems. Explain state, parameter, and initial condition through
one worked decay model, then give me an independent variation and its solution.
Leave numerical methods for the next unit.

Or bring an existing draft:

Use textbook-anything to revise these course notes. Keep the original topics,
fill in missing derivation steps, and turn the scattered exercises into
connected problems with complete solutions. Preserve the document format.

For a review without edits, ask it to identify problems and their locations before changing the source.

See the examples

The repository includes a full textbook and two short examples: finite-trajectory GAE and attention with conditional flow matching. The short examples include LaTeX source, numerical code, and tests so you can follow the connection between an argument, a problem, and its implementation.

The typesetting template uses serif body text, sans-serif headings, and matching Latin and mathematical fonts. Equations break at logical steps, and figure labels follow the notation in the text. An HTML tutorial example adds adjustable curves, a process diagram, and expandable solutions. Missing dependencies are installed first; if the preferred toolchain cannot run, a capable HTML route preserves the same explanation and visual standards. Build and test commands are in the tool guide.

For paper figures, the workflow checks arXiv source assets against their captions and packages selected reusable originals with the tutorial ZIP, subject to the applicable license.

Roadmap

This is v0.4. Start with the tier that fits your model budget and learning goal.

• Content: add examples from more STEM courses and improve difficult explanations and chapter connections.
• Delivery: refine fonts, diagrams, and pagination, and make the source easier to edit.
• Assessment: improve problem progression, hints, solutions, and marking guidance.

If an explanation loses you, a problem is missing a condition, or a page is awkward to read, open an issue with the relevant location.

Acknowledgments and license

Thanks to the authors of these works for the writing and teaching ideas that informed this skill:

• Sepia: prose suited to its audience, paragraph flow, and varied sentence rhythm.
• Supervisor-Skills: source verification and scientific writing that keeps claims aligned with evidence.
• grilling: focused questions about the reader's background and learning goals.
• deepresearch-skill: reading mixed source formats and bringing figures into the explanation.
• tensor-formula-viz: clear relationships between tensor operations, shapes, dimension meanings, and diagram geometry.
• Griffiths's Introduction to Electrodynamics: the placement of immediate practice and broader chapter problems.

This project uses the MIT license. Referenced works retain their own licenses.