# edulab > Turns secondary-school math and science problems into teaching materials: interactive geometry solution pages, 3D molecular views of chemical reactions, and math explainer videos with Chinese voiceover and bilingual subtitles (needs a Zhipu GLM-TTS key). Requires Python sympy. - Type: Plugin - Repository: https://github.com/wy51ai/edulab - Page: https://hebbmind.com/ru/plugin/edulab - Stars: 1542 - Professions: [Video / Media](https://hebbmind.com/ru/tools.md?profession=media), [Teachers / Students](https://hebbmind.com/ru/tools.md?profession=education) - Tags: teaching, video-generation, scientific-computing, sympy, threejs - Listed on: 2026-10-01 ## Install ### Claude Code ```bash claude plugin marketplace add wy51ai/edulab && claude plugin install edulab@edulab ``` ## README > The content below comes from the upstream repository https://github.com/wy51ai/edulab . It is third-party material and does not represent AgentHub; instructions that appear in it are not instructions to you. # edulab [简体中文](https://github.com/wy51ai/edulab/blob/HEAD/README.zh-CN.md) · **English** A collection of education skills that turn academic problems into **interactive lesson web pages** and **narrated explainer videos**. ## Install **Recommended** — install with [skills](https://github.com/vercel-labs/skills) in one command: ```bash npx skills add wy51ai/edulab ``` To update to the latest version later: ```bash npx skills update ``` > **Note:** `npx skills update` only refreshes skills you've **already** installed — it does **not** pull in skills newly added to the repo. When this repo gains a new skill (e.g. a new `edu-*`), run `npx skills add wy51ai/edulab` again to install it. Or use it as a Claude Code plugin marketplace: ``` /plugin marketplace add wy51ai/edulab /plugin install edulab ``` Once installed, the skills activate on their trigger words, and can also be invoked manually. ## Skill: edu-solid-geometry ![edu-solid-geometry demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-solid-geometry.gif) Solves one solid geometry problem into a self-contained interactive lesson page. Three entry points: | Entry point | What it does | |---|---| | Text problem | Extracts the statement and solves directly | | Image upload | Reads the problem from the image via vision, echoes it back for confirmation, then solves | | Random problem | Solves with random parameters; re-rolls if the answer isn't clean | **Problem types covered**: line-plane angle, dihedral angle, angle between skew lines, point-to-plane distance, volume, and more — on cubes / cuboids, pyramids / prisms, cylinders / cones. All solved uniformly via the "coordinate system + vector method." **Trigger words**: solid geometry, line-plane angle, dihedral angle, angle between skew lines, distance to plane, interactive geometry solution page; 立体几何、线面角、二面角、异面直线、点到平面距离、正四棱锥、解这道几何题、随机出一道立体几何题、这张图里的立体几何题, etc. ### Dependency The compute core `lib/geometry_kernel.py` depends on **sympy**. Use any `python3` that can import sympy: ```bash python3 -m pip install sympy # if sympy is missing ``` ### Generate from the command line (without Claude) ```bash cd skills/edu-solid-geometry python3 scripts/generate.py cube ./cube.html # cube · line-plane angle python3 scripts/generate.py box ./box.html # cuboid · volume python3 scripts/generate.py random 7 ./random.html # random problem (seed=7) python3 lib/geometry_kernel.py # kernel built-in self-check ``` > If you don't pass an output path, it writes to the **current working directory (cwd)**. ## Skill: edu-analytic-geometry ![edu-analytic-geometry demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-analytic-geometry.gif) Solves one analytic geometry (conic sections) problem into a self-contained interactive lesson page. Same three entry points as above (text / image / random). Built on a **2D Canvas board + KaTeX** with a generic, data-driven interactive engine: a parameter slider drives derived constructions (line∩conic, point-on-conic, central reflection, tangent…) and live readouts, with a theoretical-range bar or a fixed-value indicator. **Problem types covered**: standard equation, chord length, dot-product range / fixed value, triangle-area extremum, fixed point, fixed value (slope product), locus, tangent, eccentricity — on ellipses / hyperbolas / parabolas / circles. All solved uniformly via "parametrized line `x = my + c` + system + Vieta's formulas + substitution." > A correctness note baked into the kernel: open/closed interval endpoints are decided by whether a *real* line attains them, so the boxed answer always matches what the interactive tool shows (e.g. the ellipse `MA·MB` range is the **closed** `[-3, 7/4]` — slide θ to 0° and you read exactly −3). **Trigger words**: analytic geometry, conic sections, ellipse, hyperbola, parabola, chord length, dot product range, fixed point, fixed value, locus, eccentricity, interactive analytic geometry solution page; 解析几何、圆锥曲线、椭圆、双曲线、抛物线、焦点弦、向量数量积取值范围、定点问题、定值问题、斜率之积、三角形面积最值、轨迹方程、离心率, etc. ### Dependency The compute core `lib/analytic_kernel.py` depends on **sympy** (same as above). ### Generate from the command line (without Claude) ```bash cd skills/edu-analytic-geometry python3 scripts/generate.py list # list registered problem types python3 scripts/generate.py ellipse_dot_range ./sol.html # ellipse · MA·MB range [-3, 7/4] python3 scripts/generate.py parabola_dot_const ./sol.html # parabola focal chord · OA·OB ≡ -3 python3 scripts/generate.py all ./out_dir # all registered types python3 lib/analytic_kernel.py # kernel built-in self-check ``` > Like above, no output path → writes to the **current working directory (cwd)**. ## Skill: edu-chem-reaction ![edu-chem-reaction demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-chem-reaction.gif) Turns one chemical reaction into a self-contained **microscopic 3D demonstration** page: an interactive Three.js molecular animation (drag the slider to watch bonds break / form and atoms recombine, with step highlighting) next to the KaTeX equation, step-by-step narration, an atom-conservation counter, and an optional energy–reaction-coordinate curve. Same three entry points (text / image / random). **Two engines, auto-selected** — one renderer with two per-frame position resolvers, sharing the bond-diff drawing, labels, overlays and UI: | Engine | For | Emphasizes | |---|---|---| | morph | combustion, combination / decomposition / displacement, redox | atoms fly to new partners — atom conservation & recombination | | mechanism | organic mechanisms (esterification…) with catalyst · transition state · leaving groups | rigid fragments move through keyframes — the mechanism | **sympy-driven correctness**: auto-balances the equation (integer coefficients from the matrix null space), validates the atom map (a bijection between reactant and product atoms) and conservation, and derives which bonds break / form from the before↔after difference — equation, geometry and counters all share one source. **Hybrid geometry**: a built-in VSEPR molecule library by default; if RDKit is installed it can build conformers from SMILES (never installs it automatically). **Reactions covered**: methane / hydrogen combustion, water electrolysis, Na + Cl₂ redox (with an electron-transfer overlay), glucose aerobic oxidation, and the esterification mechanism — spanning junior-high basics, senior inorganic redox, and organic mechanisms. **Trigger words**: chemistry reaction, microscopic / molecular animation, combustion, electrolysis, redox electron transfer, esterification mechanism, bond breaking and forming, atom conservation, balance equation, interactive chemistry reaction page; 化学反应、微观演示、分子动画、燃烧、电解水、氧化还原、电子转移、酯化反应、断键成键、原子守恒、化学方程式配平 etc. ### Dependency The compute core `lib/reaction_kernel.py` depends on **sympy** (same as above). **RDKit is optional** — used only if already installed, never installed automatically. ### Generate from the command line (without Claude) ```bash cd skills/edu-chem-reaction python3 scripts/generate.py list # list registered reactions python3 scripts/generate.py combustion_ch4 ./reaction.html # methane combustion (morph · flame) python3 scripts/generate.py esterification ./reaction.html # esterification (mechanism · catalyst) python3 scripts/generate.py random 7 ./random.html # random reaction (seed=7) python3 lib/reaction_kernel.py # kernel built-in self-check ``` > Like above, no output path → writes to the **current working directory (cwd)**. ## Skill: edu-math-video ![edu-math-video demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-math-video.jpg) Turns one math problem (geometry, algebra, functions, motion problems…) into a **16:9 1920×1080 explainer MP4**: Chinese voice-over (Zhipu **GLM-TTS**), bilingual zh + en subtitles (`.srt` too), and hand-drawn notebook-style canvas animation driven by the narration timeline. Input is a problem screenshot or plain text. **The picture explains the step**: every narration line gets a "point → move → keep" action on the figure (equal segments slide onto each other, congruent triangles overlay, the 3D camera tweens to a top view, a cone unrolls into a sector…) — timed by `S.at(k, f)` (a fraction into line *k*), never by hard-coded seconds. A `motion` check rejects static lines. **Pipeline** (one folder per video, created in the user's current directory): ``` script.json ──build_audio.py──► timeline.json + mix.wav + .srt (GLM-TTS + synthesized music) anim.js + engine.js ──node render.mjs video──► .mp4 (Playwright + ffmpeg, 30 fps) ``` **Guard rails**: the first scene shows the original problem with each condition boxed as it is read; `tts` text must be speakable Chinese (no digits / math symbols); polyphones are pinned (`长[cháng]`, shared `pron.json` lexicon) and `--check` must report 0 before any paid TTS call; a free `--preview` + contact-sheet review comes before real audio; `--asr` transcribes the audio back to catch misread letters. **Trigger words**: math explainer video, walkthrough video, problem-solving video, micro-lesson; 讲解视频、解题视频、例题精讲、微课 etc. ### Dependency - Python 3 with `numpy requests pypinyin pillow`; Node.js 18+ with `playwright` + `ffmpeg-static` (installed once in the workspace — `scripts/new_video.sh` writes the `package.json`); Google Chrome or Playwright Chromium. - A Zhipu **`GLM_API_KEY`** provided by you, in `~/.config/math-problem-video/.env` (see `reference/glm-tts-setup.md`). - **No key?** It falls back automatically (`TTS_ENGINE=auto`): [edge-tts](https://github.com/rany2/edge-tts) if installed (free neural voices, needs internet), else macOS `say` (offline, robotic). Force one with `TTS_ENGINE=glm|edge|say`. ### Run the pipeline by hand (without Claude) ```bash bash skills/edu-math-video/scripts/new_video.sh "$PWD" my_problem # scaffold from template/ bash skills/edu-math-video/scripts/setup_check.sh my_problem # must print ALL OK cd my_problem python3 build_audio.py --check # script + pronunciation lint (free) python3 build_audio.py --preview && node render.mjs motion && node render.mjs stills auto python3 build_audio.py && node render.mjs video 6 # real TTS, then render (6 = parallel pages) ``` ## Skill: edu-physics-video ![edu-physics-video demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-physics-video.jpg) Turns one physics problem (mechanics, optics, electromagnetism…) into a **16:9 1920×1080 explainer MP4**: Chinese voice-over (Zhipu **GLM-TTS**), bilingual zh + en subtitles (`.srt` too), and blueprint-style canvas animation driven by the narration timeline. Input is a problem screenshot or plain text. Same pipeline as edu-math-video; what changes is the solving method and the picture. **The picture follows the physics**: model the object, split the motion into segments, draw the forces, pick a law, solve, then check. Force arrows grow from the point of application on one scale, the object moves on the analytic solution `x(t)` through a linear `simT()` map, the v–t graph grows in sync, and friction reverses when the relative motion does. Timed by `S.at(k, f)`. **Pipeline** (one folder per video, created in the user's current directory): ``` script.json ──build_audio.py──► timeline.json + mix.wav + .srt (GLM-TTS + synthesized music) anim.js + phys.js + engine.js ──node render.mjs video──► .mp4 (Playwright + ffmpeg, 30 fps) ``` **Guard rails**: the first scene shows the original problem with each condition boxed as it is read; `tts` text must be speakable Chinese (units read aloud; no digits or math symbols; θ is 西塔); polyphones are pinned and `--check` must report 0 before any paid TTS call; a free `--preview` plus a contact-sheet review comes before real audio; `--asr` transcribes the audio back to catch misread letters. Shared physics props go in `phys.js`. Leave `engine.js` unchanged. **Trigger words**: physics explainer video, force diagram, Newton's laws, incline, projectile, circular motion, energy, momentum, refraction; 物理题视频、受力分析、牛顿定律、斜面、平抛、圆周、能量、动量、折射 etc. ### Dependency Same as edu-math-video: Python 3 with `numpy requests pypinyin pillow`, Node.js 18+ with `playwright` + `ffmpeg-static`, and a Zhipu **`GLM_API_KEY`** in `~/.config/math-problem-video/.env` (shared by all three video skills). No key → `TTS_ENGINE=auto` falls back to [edge-tts](https://github.com/rany2/edge-tts), then macOS `say`. ### Run the pipeline by hand (without Claude) ```bash bash skills/edu-physics-video/scripts/new_video.sh "$PWD" my_problem bash skills/edu-physics-video/scripts/setup_check.sh my_problem # must print ALL OK cd my_problem python3 build_audio.py --check python3 build_audio.py --preview && node render.mjs motion && node render.mjs stills auto python3 build_audio.py && node render.mjs video 6 ``` ## Skill: edu-chem-video ![edu-chem-video demo](https://raw.githubusercontent.com/wy51ai/edulab/HEAD/edu-chem-video.jpg) Turns one chemistry problem (redox balancing and the double-line bridge, amount of substance, equilibrium and the ICE table, ionic reactions, electrochemistry, titration, industrial processes…) into a **16:9 1920×1080 explainer MP4** with the same voice-over and bilingual subtitles, drawn on a **green chalkboard**. The balanced equation stays pinned in a strip at the top; the demo sits on the left and the slate on the right. **The equation does the explaining, and conservation is counted on screen**: oxidation numbers rise and fall on a number line, electrons group until the counts match, coefficients fly from the electron row into the equation, atom tallies check off, the double-line bridge shows electron flow, molecule boxes react in the coefficient ratio, and the ICE table fills row by row. `scripts/chem_check.py` checks the balance and atom / charge conservation before any drawing. In `tts`, formulas become names (`HNO3` → 硝酸, `SO4^{2-}` → 硫酸根离子); the `zh` subtitle keeps the formula and renders subscripts. **Pipeline**: same shape as above, with `chem.js` next to `engine.js` (`equation`, `valence`, `bridge`, `eFlow`, `tally`, `molMap`, `iceTable`, `molecule`, `beaker`, `graph`). **Guard rails**: the same TTS, preview, and `--asr` rules as the other video skills, and `chem_check.py check` must exit 0. Oxidation states, coefficients, bridge endpoints, and molecule counts match the solution card. Lost electrons / oxidation are blue; gained electrons / reduction are red. Shared chemistry props go in `chem.js`. Leave `engine.js` unchanged. **Trigger words**: chemistry explainer video, redox, electron conservation, double-line bridge, amount of substance, chemical equilibrium, ICE table, titration; 化学题视频、氧化还原配平、双线桥、电子守恒、物质的量、化学平衡、三段式、滴定 etc. ### Dependency Same as edu-math-video (shared `GLM_API_KEY` and the same fallback). `scripts/chem_check.py` needs no extra packages — it checks coefficients, atom / charge conservation, and molar mass. ### Run the pipeline by hand (without Claude) ```bash bash skills/edu-chem-video/scripts/new_video.sh "$PWD" my_problem bash skills/edu-chem-video/scripts/setup_check.sh my_problem # must print ALL OK cd my_problem python3 build_audio.py --check python3 build_audio.py --preview && node render.mjs motion && node render.mjs stills auto python3 build_audio.py && node render.mjs video 6 ``` ## How it works 1. **Get a problem spec** — normalize all three entry points into a structured description (body type and dimensions, given conditions, the quantity asked, language). 2. **Exact kernel computation** — sympy computes exact coordinates, key vectors, normals, the final answer, and every intermediate value (as LaTeX strings). Never by hand. 3. **Assemble and inject the template** — feed the `lesson` / `steps` / `model` data into the data-driven template `template/lesson.html`; 3D vertex coordinates come from `kernel.to_three(...)`, sharing the same source as the solution. 4. **Self-check** — kernel answer == answer card == final value shown in the last step; a local static server + preview check confirms no console errors and correct formula/highlight rendering. 5. **Deliver** — the finished page is written to the user's current working directory, named like `solution-.html`. ## Project structure ``` edulab/ ├── .claude-plugin/ │ ├── plugin.json # plugin metadata │ └── marketplace.json # marketplace manifest ├── index.html # finished sample (quad pyramid · line-plane angle) └── skills/ ├── edu-solid-geometry/ # solid geometry — 3D (Three.js) + MathJax │ ├── SKILL.md │ ├── template/lesson.html # data-driven template (generic 3D renderer + data island) │ ├── lib/ │ │ ├── geometry_kernel.py # sympy exact-computation core │ │ └── bodies.py # edge-topology library for solids │ ├── scripts/generate.py │ ├── output/ │ └── references/ # problem-schema.md · conventions.md ├── edu-analytic-geometry/ # analytic geometry / conics — 2D (Canvas) + KaTeX │ ├── SKILL.md │ ├── template/board.html # data-driven template (generic 2D renderer + param engine) │ ├── lib/ │ │ ├── analytic_kernel.py # sympy exact-solver core (system · Vieta · range · fixed value) │ │ └── conics.py # conic-section definition library │ ├── scripts/generate.py │ ├── output/ │ └── references/ # problem-schema.md · conventions.md ├── edu-chem-reaction/ # chemistry reactions — 3D (Three.js) + KaTeX │ ├── SKILL.md │ ├── template/reaction.html # data-driven template (unified renderer + dual engine + data island) │ ├── lib/ │ │ ├── reaction_kernel.py # sympy balancing + conservation/atom-map check + bond-diff + assembly │ │ └── molecules.py # VSEPR molecule-geometry library │ ├── scripts/generate.py │ ├── output/ │ └── references/ # problem-schema.md · conventions.md ├── edu-math-video/ # math explainer videos — GLM-TTS + canvas animation → MP4 │ ├── SKILL.md │ ├── template/ # runnable sample project (engine.js · anim.js · build_audio.py · render.mjs) │ ├── shared/ # pron.py + pron.json — shared pronunciation lexicon │ ├── scripts/ # new_video.sh · setup_check.sh · problem image & contact-sheet tools │ ├── examples/cone-parallel/ # solid-geometry example (camera tween · cone unrolling) │ └── reference/ # TTS setup · script writing · pronunciation · visual design · animation API ├── edu-physics-video/ # physics explainer videos — GLM-TTS + blueprint canvas → MP4 │ ├── SKILL.md │ ├── template/ # blueprint theme (engine.js · phys.js · anim.js) │ ├── shared/ │ ├── scripts/ │ ├── examples/laser-curvature/ # geometric-optics example │ └── reference/ └── edu-chem-video/ # chemistry explainer videos — GLM-TTS + chalkboard canvas → MP4 ├── SKILL.md ├── template/ # chalkboard theme (engine.js · chem.js · anim.js) ├── shared/ ├── scripts/ # plus chem_check.py — balance and conservation check ├── examples/equilibrium/ # ICE-table / equilibrium example └── reference/ ``` ## Extending **edu-solid-geometry** - **Add a problem type**: add a solver function in `geometry_kernel.py` (see the recipe table in `references/conventions.md`), then add a `build_*` in `generate.py`. - **Add a solid**: add a coordinate-construction function in `geometry_kernel.py`, then add its edge topology in `bodies.py`. **edu-analytic-geometry** - **Add a problem type**: add a target-quantity function in `analytic_kernel.py` and reuse `range_over_m` / `is_constant_in_m`, then add a `build_*` in `generate.py` (pick an interaction: range bar / fixed value / fixed point / locus trace). - **Add a curve**: ellipse / hyperbola / parabola / circle are built in; new curves go in `conics.py` and the `board.html` engine. **edu-chem-reaction** - **Add a reaction**: add a `build_*` in `generate.py` (high-level `species + atom_map`, or low-level `atoms + fragments` for mechanisms) and register it in `REGISTRY`. - **Add a molecule / ion**: add an entry in `lib/molecules.py` (VSEPR geometry + display metadata + internal bonds). **edu-math-video** - **New problem**: never edit `engine.js`; write the per-problem `script.json` / `storyboard.md` / `anim.js`, and add new figure helpers inside `anim.js`. - **Fix a reading**: add the word to `shared/pron.json` (`words` / `ok`) — one lexicon shared by every video. **edu-physics-video** - **New problem**: leave `engine.js` unchanged; write the per-problem `script.json` / `storyboard.md` / `anim.js`. Shared physics props go in `phys.js`. - **Fix a reading**: add the word to `shared/pron.json`. **edu-chem-video** - **New problem**: leave `engine.js` unchanged; write the per-problem files. Shared chemistry props go in `chem.js`. Run `scripts/chem_check.py` on the equation before drawing. - **Fix a reading**: add the word to `shared/pron.json`. ## License [Apache-2.0](https://github.com/wy51ai/edulab/blob/HEAD/LICENSE) ## Author WY · [@akokoi1](https://x.com/akokoi1) | WeChat | Xiaohongshu | |---|---| | WeChat official account | Xiaohongshu | ## Star History Star History Chart