Teardown logo

Teardown

CommunityPopular
elvarofficial
teardown

Build an interactive 3D simulator of a real machine or mechanism (an engine, a landing gear, a walking linkage, a pump, a gearbox) as one offline HTML page people can take apart, with live panels driven by the same model that moves the parts, a button for every subsystem, cutaways, camera views and an auto tour that hands control to the visitor; then record it frame by frame to a 60 fps video with a synced procedural score. Use when someone asks how a machine works, wants a "visual explainer", a "simulator", a "cutaway", an "exploded view" or an "interactive 3D lesson", or wants a 3D object shown working rather than described.

Overview

Publisherelvarofficial
Repositoryteardown
Skill nameteardown
Stars
1.3K
Forks
204
Bundled files
69
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 69 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by elvarofficial on GitHub. Read the source before you install it.

Installation

Install the Teardown AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/elvarofficial/teardown/tree/main
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/elvarofficial/teardown.git \
  .claude/skills/teardown
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Teardown in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Teardown on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Teardown is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Teardown

One machine, one page. The visitor sees the thing run, presses a button, and the part doing the work opens up while the gauges around it read the real numbers. Left alone, the page tours itself with a cursor. The same code records a film of the tour, frame by frame, with a score built on the same timeline.

SKILL_DIR below is the folder this file lives in. Builds and renders go to ./teardown-out of the folder you run from.

First run

cd "$SKILL_DIR" && npm install && npx playwright install chromium

The pages need nothing else. The video needs ffmpeg on PATH and the score needs Python 3 (standard library only). Set FFMPEG, PYTHON or CHROMIUM to point at other binaries.

The 3 examples

Build one, open the HTML, press buttons. Copy the closest one when you start a new machine.

The machineWhat drives itWorth stealing
v8-*A cross-plane V8 with glass liners: the charge burns inside the cylinder you track, valves lift on their cams, 8 pistons move from the exact slider-crank.Crank angle. Firing order 1-8-4-3-6-5-7-2, a chamber pressure model for the indicator diagram.Cutaway modes (glass, section, solid), the firing order as a cylinder picker, live charts synced to the crank.
cessna-*The Cessna 337 main landing gear folding into the airframe on takeoff and dropping on landing.Stroke 0..1: a 201 mm rack turns a 32-tooth pinion (module 4 mm) through exactly 180°.One degree of freedom posing a whole mechanism, flight phases as functions of time, a runway that rolls under the model.
strandbeest-*Theo Jansen's Strandbeest, 20 legs on one crankshaft, walking broadside over sand.Wind. The crank speed is integrated from it; every joint follows by circle intersection from Jansen's 11 link lengths.An exact linkage solver in 67 lines, instanced tubes, foot traces, a contact chart of all 20 feet.

Each comes in 3 folders: <name>-shared/ holds the machine, <name>-film/ the recorded tour, <name>-live/ the page you drive.

node "$SKILL_DIR/sims/build.mjs" v8-live          # → teardown-out/dist/v8-live.html, offline, ~800 KB
node "$SKILL_DIR/sims/build.mjs"                  # all 6

How a machine is built

One folder per machine, split by job. The examples follow this exactly:

FileJob
kin.js / gear.js / jansen.jsThe math. Real dimensions in real units, one input (crank angle, stroke, wind) and every pose and readout as a function of it. No three.js in here.
engine.js / gear.js / beast.jsThe geometry, built from code: lathed parts, tubes, instanced links. update(input) poses it.
stage.jsStudio, lights, shadows, post (GTAO, bloom). frame(state, cam) draws one frame.
console.jsThe overlay: fixed 1600 x 900 CSS layout of panels, charts, toggles, callouts with leaders. update(state) redraws it from the frame's state.
film.jsA SCRIPT of operator moves (views, isolate, cutaway, throttle) and render(t) as a pure function of time.
live.jsThe same machine in real time: buttons, sliders, orbit, zoom, keys. With no input it plays film.js's script as an auto tour with a cursor; any click, drag, wheel or key hands control over, 25 s of quiet hands it back.
score.pyThe sound on the film's timeline.

For a machine outside SKILL_DIR, make a folder with a main.js and import the kit by name: import {runSim} from 'teardown-kit/runtime.js'. Build it with node "$SKILL_DIR/sims/build.mjs" ./my-machine.

Rules that decide whether it works

  1. Start from a reference, never from imagination. A photo, a cutaway drawing from a manual, a clip of the real thing. Build against it and compare frame by frame. Machines invented from a sentence come out generic every time; the same model rebuilds a real one closely when it has something to match.
  2. Real numbers, one source. Get the real dimensions first (bore, stroke, tooth count, link lengths). Every gauge, chart and label reads from the same function that poses the parts, so nothing on screen can disagree with the motion. Check one by hand: in the Cessna, 32 teeth × 4 mm module is a 128 mm pitch diameter, and half its circumference is the 201 mm stroke.
  3. One input drives everything. Pose from the crank angle, the stroke or the wind. Nothing is keyframed by hand, so the film, the live page and the score never drift.
  4. The part doing the work opens up. Isolate it, cut it, x-ray it. A camera that goes inside is worth more than ten orbits.
  5. Light studio, one accent. Pale backgrounds read in a feed; dark ones lose the first second. The accent marks the thing that is changing. Watch near-white parts on a pale backdrop: give them an ink outline or a darker ground.
  6. Pick the camera for the mechanism. A planar linkage seen face-on collapses into a lattice; a 3/4 view along the spine keeps every leg readable.
  7. Fix collisions with a number. When a part pokes through another, scale or move it in the parameter block. Never hide it with the camera.

Live page rules

  • Test on an integrated GPU. A page that runs at 60 fps on a desktop card can drop to 15 fps on a laptop's built-in one. The examples carry quality tiers (TIERS in live.js): no AO, lower pixel ratio, smaller shadow maps, panels redrawn every other frame. Find the expensive shader first: a displacement computed 5 times per vertex costs more than every tier together.
  • Warm the shaders. kit/warm.js compiles every program in parallel before the first frame, so the page never freezes on first draw.
  • Keep the page under 1 MB. Link crawlers (X, Slack) read the share card from the HTML and give up on a 15 MB page. Big models go out as files next to the page: build.json → {"external": {"heavy.glb": "light.glb"}}.

render(t) contract

js
import {runSim} from '../kit/runtime.js';

const DEFAULTS = {duration: 20, fps: 60, hud: {enabled: false}, director: {enabled: false}};

runSim(DEFAULTS, async ({cfg, canvas, capture}) => {
  // build the scene, precompute anything stateful
  return {render(t) { /* draw frame t from scratch */ return {}; }, resize() {}};
});
  • render(t) is pure: no Math.random (use rng(seed) from kit/util.js at setup), no Date, no state carried between frames. Capture calls it at arbitrary times.
  • Anything that moves by its own rules (agents, particles) is simulated once at load into typed arrays; render only reads them. A quantity integrated over time (the Strandbeest's crank under a changing wind) is integrated in closed form or tabulated at load.
  • Every beat goes into <name>-film/timeline.json in seconds. The score reads the same file.
  • Frame 0 is the thumbnail. It must already show the machine.

Commands

node "$SKILL_DIR/sims/build.mjs" v8-film
node "$SKILL_DIR/sims/capture.mjs" v8-film --size 1600x900 --dpr 1.2 --stills beats --cols 4 --out teardown-out/v8-beats.png
node "$SKILL_DIR/sims/capture.mjs" v8-film --size 1600x900 --dpr 1.2 --still 12.5 --out teardown-out/v8-12.5.png
node "$SKILL_DIR/sims/capture.mjs" v8-film --size 1600x900 --dpr 1.2 --out teardown-out/v8-silent.mp4
python "$SKILL_DIR/run_sfx.py" "$SKILL_DIR/sims/v8-shared/score.py"
ffmpeg -y -i teardown-out/v8-silent.mp4 -i teardown-out/v8-score.wav -map 0:v -map 1:a -c:v copy -c:a aac -b:a 256k -shortest -movflags +faststart teardown-out/v8.mp4
node "$SKILL_DIR/wave.mjs" teardown-out/v8.mp4 --timeline "$SKILL_DIR/sims/v8-film/timeline.json"

--size is the CSS layout and --dpr multiplies it: 1600 x 900 at 1.2 gives 1920 x 1080. Capture prints the GPU it got; a software renderer means the GPU flags in paths.mjs did not take.

Review loop

You cannot watch the video; you can look at frames. Never report a machine as done without reading sheets of the current render.

  1. --stills beats: one tile per timeline beat, 0.3 s after it, each labelled t=.
  2. Read every tile against the checklist below.
  3. Zoom in: --stills every:0.2 --from A --to B for a transition, --still T for one full frame.
  4. Fix, rebuild, go back to 1. Stop when a full pass finds nothing.
  5. Final: node "$SKILL_DIR/sheet.mjs" --video <final.mp4> --timeline <timeline.json> and read that too: it shows what the encoder wrote.

Checklist:

  • The mechanism. Every part that should touch touches; nothing passes through anything; the numbers on the panels match the pose.
  • Composition. One focal point per frame. Nothing important within 60 px of an edge. Labels never cover the part they name.
  • Type. Nothing clipped. The key readout reads at phone width.
  • 3D. Bloom eats detail: keep emissive layers under 1.0. Near-white surfaces blow out. Glossy metal under a low light streaks: roughness ≥ 0.45 and envMapIntensity ≈ 0.5 on big surfaces.
  • Timing. Each state holds long enough to read: 1 s for a short phrase plus 0.25 s per word beyond three.

three.js traps the examples already hit

  • smoothstep(a, b, x) with a > b returns 0 on Windows (ANGLE/D3D). Write 1.0 - smoothstep(b, a, x).
  • A ShadowMaterial or a decal a fraction of a millimetre above the floor loses the depth test inside a post-processing render target. Give it polygonOffset.
  • A quad laid into the ground with a mirrored mapping gets culled as a back face. Use DoubleSide.
  • One NaN pixel blacks out the whole frame through bloom. Clamp every pow() base in a shader.
  • GTAO on a clipped (cutaway) model needs the same clipping planes on gtao.normalMaterial, or the cut half returns as a pale ghost.

Sound

sfx/ synthesises every sound in pure Python (48 kHz, reproducible) and reads the film's timeline, so sound cannot drift out of sync. python "$SKILL_DIR/run_sfx.py" --list shows the generators. sample() adds the recorded CC0 sounds in sfx/samples/. sfx/score_kit.py has the beds and accents the 3 scores use: a chord bed, a soft thud per power stroke, a motor, air under a drag.

  • Every time comes from the timeline, never a typed number.
  • Mastering is automatic: −14 LUFS, true peak ≤ −2 dBFS.
  • Accent, never narrate: continuous motion stays silent. One key for every tonal sound.

Taste, with numbers

  • Easing: entrances cubic-bezier(0.16, 1, 0.3, 1), exits cubic-bezier(0.6, 0, 0.9, 0.35) at half the entrance's length, state to state cubic-bezier(0.65, 0, 0.35, 1). Linear only for continuous motion.
  • The machine fills 60-70% of its viewport. Labels never under 34 px on a 1080 canvas.
  • One accent, and it means something: the part that is working right now.
  • Nothing dead-still: one element always alive.

Credits

The kit, frame capture, sfx engine and review rules come from whaleyxbt/claude-motion (MIT). The recorded samples are CC0 from Freesound, trimmed by Barty-Bart/motion-graphics (MIT); sources in sfx/samples/SOURCES.md.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

and 9 more files.

Frequently asked questions

What does the Teardown AI skill do?

Build an interactive 3D simulator of a real machine or mechanism (an engine, a landing gear, a walking linkage, a pump, a gearbox) as one offline HTML page people can take apart, with live panels driven by the same model that moves the parts, a button for every subsystem, cutaways, camera views and an auto tour that hands control to the visitor; then record it frame by frame to a 60 fps video with a synced procedural score. Use when someone asks how a machine works, wants a "visual explainer", a "simulator", a "cutaway", an "exploded view" or an "interactive 3D lesson", or wants a 3D object...

Why use Teardown on TypingMind?

Because you install it once and use it with any model. Teardown is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Teardown in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/elvarofficial/teardown/tree/main. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Teardown?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Teardown?

As many as you like. As long as a model supports skills, you can use Teardown with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Teardown AI skill free?

Yes. It is published on GitHub by elvarofficial under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇