edulab

edulab

简体中文 · English

A collection of education skills that turn academic problems into interactive lesson web pages and narrated explainer videos.

Install

Recommended — install with skills in one command:

npx skills add wy51ai/edulab

To update to the latest version later:

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

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:

python3 -m pip install sympy   # if sympy is missing

Generate from the command line (without Claude)

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

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)

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

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)

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

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 + <name>.srt   (GLM-TTS + synthesized music)
anim.js + engine.js ──node render.mjs video──► <name>.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 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 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)

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-<short-description>.html.

Project structure