# A 24-second teardown video, 2,300 stars in a week: the teardown took off because of the author's 40k followers, not the code · GitHub open-source project

> Source: https://stringers.io/en/github/gh_elvarofficial__teardown.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> An author with 41k followers posted a long thread with a 24-second teardown video (347k views), right when the "AI takes apart machines" theme and the word "teardown" were both trending. It was then reposted for a week by Instagram traffic accounts, multilingual X accounts, and TikTok.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

elvarofficial/teardown · elvarofficial

# A 24-second teardown video, 2,300 stars in a week: the teardown took off because of the author's 40k followers, not the code

**What it is** An open-source Claude Code skill: give it a reference image or a sentence, and Claude turns a machine into a 3D web page you can open and take apart, with gauges that read real numbers, and one command records it as a video with sound effects.

**Why it caught on** The author, who has 41k followers, posted a long post with a 24-second teardown video (347k views), right when "AI takes apart machines" and the word "teardown" were both trending. Then Instagram lead-gen accounts, multilingual X accounts, and TikTok reposted it for a week.

**C** **How you could follow it** Stop making machine teardowns and pick a different "take it apart" subject nobody has done yet. Copy its approach: hard-code the aesthetic rules, bundle the recording, launch with a video-led long post, put the link in a reply, and hand the finished video to second-hand accounts to repost. This only works if you have an account that can get the first wave going.

Grade C: the format is already taken by anatomy (25k stars), the identically named /teardown, and the project itself. Its stars came mainly from the author's 40k-follower distribution, which followers can't get. What you can learn is the approach where the skill's output is the distribution asset. You only have a chance if you pick a different subject. It's spreading in small circles

Data (updated daily) · Repo created 2026-10-30 · 2,537 stars · 7 days +2,184

GitHub · stars

Data 2026-11-01 → 2026-11-08 (8 days, 8 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

An X creator who makes AI videos packaged the whole workflow for making 3D teardown videos into an open-source Claude Code skill. On launch day the GitHub repo gained no stars. The next day it gained +350, and a week later it had 2,300 stars and 361 forks. It wasn't the first in its niche, and its technology isn't the strongest, but it was the first to launch with a finished video from a big account. Below: what it is, where the stars came from, why this one took off, and whether building a similar project is still worth it now.

### A long post from October 30

On October 30, 2026 at 15:38 (UTC), @elvar_official on X posted a long post with a video. His bio says "AI Videos / Vibe Coding", he has about 41k followers, and he is based in Washington State, USA. The post opens with this line:

> I just packed my whole 3D workflow into one Claude skill.

He went on: I gave Opus 5.5 a reference image and asked it to "take this thing completely apart and write every part in code." Some of the method: the housing, gears, and hoses are all three.js code. Before rendering the video, he had Opus compare the render against the reference image side by side for 3 rounds. Under the big gear are 260 small pistons, hoses, screws, and wires. "The first version just floated there, so now a part falls off every 2–3 seconds." The last line: "2 days, 7 of these. The skill, the showcase site, and all 9 prompts are below 👇"

The attached video is 24 seconds long, on a light background, with a machine dropping one part every two or three seconds. As of 11-06, the post had 347k views, 3,967 likes, and **4,481 bookmarks** (X, 11-06). Bookmarks outnumbering likes is typical of a "tutorial plus free resource" post. The GitHub link wasn't in the post body. It went in a reply, so the repo gained +0 that day and +350 the next (day-by-day Trendshift data).

### What it is

After installing it, you ask Claude Code in plain language:

> Build me a simulator of a 4-stroke motorcycle engine I can take apart.

Claude writes a standalone offline HTML page. The machine turns in the middle, the gauges around it read real numbers, and pressing a button takes apart whichever part is working so you can see it. When nobody is touching it, it tours itself. The repo ships with 3 complete machines: a cross-plane crankshaft V8, the main landing gear of a Cessna 337, and a bionic beast with 20 legs. The README's first screen shows the machine bee, scorpion, and octopus that were added later.

The other half is video output. One command records the page frame by frame into a 60fps mp4, then generates sound effects on the timeline and sets the loudness (-14 LUFS), so the file is ready to post. The skill's docs hard-code many rules for making the visuals look good: start from the reference image, compare against it for several rounds, light backgrounds stand out more in feeds, and the first frame doubles as the thumbnail.

The README ends by stating plainly that the runtime, the frame-by-frame recording, the sound engine, and the review rules come from another open-source project, whaleyxbt/claude-motion (845 stars). The author's contribution is the subject matter and the aesthetic. A hands-on review blog (thakicloud, 10-31) also points out that its numbers come from a kinematic model, not real physics.

### Where the stars came from

| Date | What happened | Same-day star gains (Trendshift) |
|---|---|---|
| 10-30 | Author posts and builds the repo, with the link in a reply | +0 |
| 10-31 | The post spreads to 160k views; Instagram accounts start the "comment TEARDOWN and we DM you the link" lead-gen funnel | +350 |
| 11-01 | Chinese, Vietnamese, and Indonesian X accounts repost it; the skill directory lists it; #6 on the JavaScript daily chart | +523, the biggest day |
| 11-02 – 11-05 | TikTok reposts, users share the machines they built, tagging @author | +323 / +403 / +264 / +270 |
| 11-06 | Drops off the daily chart | +173 (day not over) |

A few worth a closer look:

**The author's own post is the main engine.** 4,481 bookmarks and 347k views, and every repost after that revolved around this one video. The author has only this one repo on GitHub and hasn't pushed code since 10-31.

**The name happened to match a trending term.** "/teardown" was already the name of a different skill by designer Jakub Krehel (for 2D technical illustration), posted 10-25, which got 245k views. 10-29, Perplexity's CEO posted that it would be integrated into Perplexity Computer. Elvar named his repo after the same word the next day, and searches for "teardown claude skill" mix the two up. Whether that was deliberate or a coincidence, there's no evidence.

**Instagram "comment keyword" accounts.** Several accounts posted Reels saying "comment TEARDOWN and I'll DM you this free Claude skill plus a 9-step setup." They use a free open-source project to grow followers and collect emails. Something visually striking, explainable in one line, and free is exactly what this business model needs.

**Multilingual second-hand accounts and TikTok.** A Chinese account framed it as "generate a 3D mechanical simulator you can take apart with one sentence," and someone used it to make a Gatling gun. The top TikTok repost had about 30k plays. The volume isn't large, but it explains why after 11-02 it didn't drop off a cliff and instead held steady at two or three hundred a day.

**User-made work.** The skill bundles recording, so people who install it can directly produce a teardown video ready to post, and it tags the author when posted. The fork-to-star ratio is about 15%, higher than typical spectator-type projects, which suggests people really are installing it.

Not found: ads, Hacker News, big Reddit threads, mainstream media. It also never made GitHub's site-wide Trending on any day.

### Why this one

By October 30, "have AI take a machine apart for you" had already gone viral on X two or three times:

| Project | Launched | scale (~11-06) | What the others do |
|---|---|---|---|
| wheresryan22/anatomy | built 10-08, breakout 10-12 | About 25k stars | Isometric SVG illustration explaining what a machine is made of |
| AirsupHQ/machine-skill | 10-12 | 586 stars | One sentence → photoreal materials + physics solver + video; launch post ~12k views |
| Jakub Krehel's /teardown | 10-25,10-29 plugged into Perplexity | Parent collection ~11k stars | Same name, 2D exploded / assembly diagrams; static, no readouts |
| elvarofficial/teardown | 10-30 | 2,150–2,300 stars | One reference image → a drivable, disassemblable 3D simulation + a video with sound |

Three things stand out:

- **The theme was validated early; this is the third or fourth entrant.** It does the upgrade: 2D becomes 3D, static becomes operable and disassemblable, plus a postable video. It landed just as Opus 5.5 shipped, and people were looking for "what can it do that's impressive."

- **The author's reach sets the ceiling.** machine-skill is also 3D, also disassembles, also records video, with harder physics, but its launch post got only 12k views and 25 stars in 586 days. teardown's post had 347k views, four times that in a week.

- **The output is the distribution.** Users get a video they can post right away, and posting it sends traffic back to the repo. This is part of why it took off faster than similar tools, but it came after the author's distribution.

Also watch out: growth is already cooling. Since 11-04 it has dropped off the Trendshift daily chart, and on 11-06 it was only +173; the author 10-31 hasn't pushed an update since, and 6 issues sit unanswered (GitHub, 11-06).

### Others are following it

- No direct clone repo found, but derivatives exist: VibeXR built a WebXR version called "Microduck Teardown" and openly credits Elvar; on Instagram, someone rode the trend with the same concept using a Gemini video pipeline.

- The same wave includes a pile of 3D / motion skills: open-3d-skills (262 stars), 3d-world-simulator (218 stars), "code-to-video" skills at 400–1,700 stars, and the human-body 3D dissection project human-atlas at about 4,500 stars.

- The theme of "Claude builds a disassemblable, moving thing from one sentence" was done many times from September to November.

### If it were you

**Verdict:** Building another "take the machine apart" skill will likely not get the same traction. Its stars came mostly from one post by a creator with 40k followers that got 347k views, which followers can't replicate. The part worth copying is the approach: the skill is itself the content, and its output comes with a postable video.

**Who it suits:** People who already have a few thousand real followers on X, Xiaohongshu, or Bilibili and earn from their work (AI video, 3D, motion design, educational content), who can package their own workflow as a skill and launch it through their own accounts; or people with material in a niche (chemistry, medicine, architecture, circuits, geology) who can apply "take it apart, make it move, output a video" to an object nobody has covered yet.

**Who it doesn't suit:** People with no posting channel who expect GitHub Trending to bring organic traffic. Neither teardown nor machine-skill ever made the site-wide Trending page. Also those hoping to beat it by being "more realistic, more accurate physics." machine-skill proved that hardcore doesn't equal stars.

**How to build it:** Change the object, not the format. Pick an object nobody has covered yet for the "take it apart" treatment, and follow its three-part structure: start from a reference image and fix the aesthetic rules (background, composition, font size, first frame); let one input drive all the readouts; include frame-by-frame recording and sound, with one command producing a vertical video. If you can build on an existing base, don't write your own. Its runtime, recording, and sound all come from claude-motion, and the author made 7 machines in two days.

**How to get users:** Copy its post format. Before launch, make five to seven examples. For the post, use a 20–30 second video with a change every two or three seconds. The caption reads "I packed the whole workflow into one skill," then lists 4–5 methods, ending with "skill, showcase site, and all prompts 👇," with the link in the replies. After it goes out, proactively send finished videos to Chinese tech accounts and the "comment a keyword to get the link" Instagram accounts; they need material. Answer issues in the first two weeks, and don't do what it did: post and disappear.

**When to move:** This theme's heat tracks new model releases. There were already two waves in mid and late October. The next wave will likely come around the next major model release, so have your examples and finished videos ready before then (inference).

**How you could be wrong:** ① By end of November, a same-theme skill with no big-account backing passes 1,000 stars in a week. That would show it can take off on the work alone, with a lower bar than this suggests. ② Conversely, teardown is still under 3,000 stars at the end of November and returns to double digits daily, meaning the theme has passed its peak and switching objects won't help much. ③ Anthropic or Perplexity turns "one sentence → disassemblable 3D simulation with video" into an official feature, and third-party skills lose their place.

[GitHub repo ↗](https://github.com/elvarofficial/teardown)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| wheresryan22/anatomy | Open source | Built 2026-10-08, breakout 10-12 | 25,198 stars / 2,105 forks | Isometric SVG illustrations explaining composition; not drivable, no video; the theme's originator, with 10× the reach |
| /teardown from jakubkrehel/skills | Open source | 2026-10-25 (10-29 connected to Perplexity Computer) | Collection 10,988 stars; launch post 245k views, 6,090 likes | Same name, makes 2D technical illustrations (exploded/assembly views, exports SVG/PNG), not 3D, static, no readouts |
| Perplexity Computer /teardown | Big-company features | 2026-10-29 | CEO's post 16k views, gives away 5000 trial credits | That is Jakub's 2D illustration skill, one-click in Perplexity; teardown is a 3D simulator + video generator that runs locally in Claude Code |
| AirsupHQ/machine-skill | Open source | 2026-10-12 | 586 stars / 77 forks | Starts from a sentence, real physics solver, photoreal materials, heavy on engineering; not maintained, and the author has no content distribution power |
| whaleyxbt/claude-motion | Open source | 2026-09-29 | 845 stars | teardown's base (runtime, frame-by-frame recording, sound, review rules) is general-purpose motion work and doesn't do machines |
| Code-to-video skills (claude-motion-design, claude-animated-shorts, etc.) | Open source | 2026-09 to 11 | 400–1,700 stars | Outputs video only, no interactive machine simulator |
| rtmandreyev/construct | Open source | 2026-10-24 | 115 stars | Describes an object to generate a hologram / 3D model and parts; same theme, but didn't take off |
| Prompts sent directly to Claude Opus 5.5 | Big-company features | 2026 | A lot of bare-prompt work on X | teardown locks in reference-image comparison, single-input driving, frame self-checks, and recorded narration, so the output is stable and good-looking |

Where users come from

- **Verified** **X long post + X Article tutorial** : @elvar_official's 10-30-post thread with a 24-second teardown video, "skills, showcase site, 9 prompts 👇"; 11-06 347K views / 3,966 likes / 4,479 bookmarks (fxtwitter); GitHub +0 on day one, +350 the next day, link in the replies

- Inference **borrowing a hot name, "teardown"** : Jakub Krehel 10-25 posted a /teardown skill (245K views), Perplexity's CEO 10-29 announced it's being integrated; Elvar 10-30 uses the same name. Whether this was intentional is unverified

- **Verified** **Instagram "comment TEARDOWN, DM for the link" funnel accounts** : Kartik Singh 10-31 Reel and DfSGe7BiOPk, DfXblchxWep, DfSOcPsBo42 among several others

- **Verified** **TikTok reposts** : morenicksaraev 30k plays, github.signals 12.6k; Indonesian/Vietnamese accounts (TikHub)

- **Verified** **Multilingual X resale accounts and users showing their work** : Chinese @ZQ444, @Lingze688, @huangyun_122; English @neil_xbt, @the_osps, @Retik_S, @Yashpatil and others; @RKeklikian showing work and tagging the author (Trendshift mentions)

- **Verified** **Skill directories / blogs / LinkedIn / YouTube long tail** : skill-sets.md, skillleaderboard, typingmind, enterprisedna 11-01 daily report, thakicloud 10-31 hands-on test, YouTube tutorials

- **Verified** **Trendshift JavaScript daily chart** : 10-31 through 11-03 ranked #6–#12; did not make GitHub's official trending

## Why elvarofficial/teardown worked

**In one line** : A Claude Code skill. Given a reference image or a sentence, Claude uses three.js to write a machine as a single offline HTML 3D simulator: the central machine spins, the surrounding gauges read real numbers, a click disassembles the working part, and it tours itself when idle. It can also record frame by frame into a 60fps video with sound.

**结论** :它不是「好技能自己长出来的」,而是**一个 4.1 万粉 AI 视频创作者的爆款帖 + 当时最热的词「teardown」+ 二手搬运号矩阵** 一起推起来的。Elvar 10-30 发「我把整套 3D 工作流塞进一个 Claude 技能」的长帖(现 34.7 万浏览、3,966 赞、4,479 收藏),帖子本身就是一条机器被拆开的 24 秒视频——这正是此前两周 X 上最火的内容形态(anatomy 2.5 万星、Jakub Krehel 的 /teardown 24.5 万浏览、Perplexity CEO 10-29 刚宣布把 /teardown 接进 Perplexity Computer)。随后 Instagram「评论 TEARDOWN 私信发链接」、TikTok、中文/越南/印尼 X 号、LinkedIn、技能目录站把它搬了一周,星每天稳定 +260~520。它真正不同的地方是「把出片管线(逐帧录制 + 程序化音效 + 看帧自检)和交互模拟器绑在一起」,这让每个用户都能产出可发的视频,进而又变成新的传播素材——这一点确实是它比同题材起得快的原因之一,但首推力是作者的分发。

### Data

| Metric | Value | Source |
|---|---|---|
| Repo created | 2026-10-30 09:29(UTC) | Data card |
| Current | 2,150 stars / 333 forks (data card 11-06); the GitHub page showed 2.3k / 361 the same day | Data card; github.com page (Jina scrape) |
| Daily (Trendshift, UTC) | 10-30 +0, 10-31 +350, 11-01 +523, 11-02 +323, 11-03 +403, 11-04 +264, 11-05 +270, 11-06 +173 (today still in progress) | trendshift.io/repositories/328612 |
| Daily (our database, Beijing-time scrape date) | 11-01 353 → 11-02 856 (+503, peak) → 1,194 → 1,587 → 1,878 → 2,150 | GitHub database repo_day / data card |
| Trendshift JavaScript daily chart | 10-31 #12, 11-01 #6, 11-02 #12, 11-03 #11, then dropped off | Trendshift |
| GitHub site-wide trending | Not listed (repo under ~60 days old, trending_days_any = 0) | Data card |
| fork/star | About 15% (high; many people fork it to install the skill) | Data card |
| Maintenance | 1 branches, 0 tags, 0 merged PRs; last push 10-31 | GitHub page, Trendshift |
| Author | GitHub has only these 1 repos; X @elvar_official, 41k followers, bio "AI Videos / Vibe Coding", Washington State, US | data card; api.fxtwitter.com |

### I. Competitors (doing the same thing)

| Product | Where | Launched | Scale | How it differs from teardown |
|---|---|---|---|---|
| wheresryan22/anatomy | Open-source skill | Built 10-08, breakout 10-12 | 25,198 stars / 2,105 forks (our GitHub repo) | Isometric SVG illustrations + optional WebGL, explaining "how it's built"; not a drivable 3D simulator, and it doesn't output video. The genre's originator, 10x bigger in scale |
| /teardown in jakubkrehel/skills | Open-source skill + built into Perplexity Computer | 10-25 post (245k views, 6,090 likes, 7,056 bookmarks); 10-29 Perplexity CEO announced the integration (16k views) | Skills collection at 10,988 stars, 10-28 → 11-06, up about 2,150 | **Same name.** Makes 2D technical illustrations (exploded / assembled views, editable part colors and labels, exports SVG/PNG). Not 3D, static, no numbers |
| AirsupHQ/machine-skill | Open-source skill | 10-12 | 586 stars / 77 forks | Starts from a single sentence, with photorealistic materials and a real physics solver, more engineering-flavored; the author is a manufacturing digital-twin company and it isn't maintained after launch. During teardown's hype it also got carried up about +100 |
| whaleyxbt/claude-motion | Open-source toolkit | 09-29 | 845 stars | The foundation for teardown: the runtime, frame-by-frame recording, sound engine, and review rules all come from it (the README says so). It is itself a general motion-design tool and doesn't make machines |
| "Code-to-video" skills like howseen-ai/claude-motion-design and mromidev/claude-animated-shorts | Open source | Late September–November | 400–1,700 stars | Outputs only video, no interactive machines |
| rtmandreyev/construct | Open source | 10-24 | 115 stars | Describe an object → hologram / 3D model and parts; same genre, didn't take off |
| ashemag/human-atlas, thebuggeddev/anatomy | Open source | 8–September | 4,479 / 3,881 stars | Human 3D dissection; different subject |
| Write prompts directly against Opus 5.5 | The large model itself | — | A lot of bare-prompt work on X | teardown's value is in turning "compare against a reference image over 3 rounds, one input drives everything, self-check by viewing frames, record the video with narration" into a repeatable workflow |

**The bigger picture:** by 10-30, "have Claude take a machine apart for you" was already a viral genre on X that had been validated two or three times (anatomy → Jakub's /teardown → Perplexity integration). teardown came in third or fourth, but in the "3D + drivable + can output video" slot it was the first to be posted with video by a big-account creator.

### 2. Growth levers (where users come from)

1. **The author's own viral long X post (confirmed, main engine):** 2026-10-30 15:38 UTC, @elvar_official: "I just packed my whole 3D workflow into one Claude skill," with a 24-second 1080p teardown video attached, writing "give opus 5.5 a reference image… 2 days making 7… skill, showcase site, and all 9 prompts below 👇", and an attached X Article tutorial "How I Turn Any Machine Into a 3D Simulator You Can Take Apart." Snapshot at 10-31: 163k views / 2,519 likes / 2,835 bookmarks (thakicloud.com); 11-06 shows 347k views / 3,966 likes / 455 reposts / **4,479 bookmarks** (api.fxtwitter.com). More bookmarks than likes, typical of a "tutorial + free resource" post. The link was placed in a reply, not the body (to avoid rate limits), so GitHub only gained +0 on day one and +350 the next day.

2. **Rode a name that was already hot (confirmed fact, motive not verified):** "/teardown" is the skill name released by Jakub Krehel 10-25; that post had 245k views. Perplexity CEO Aravind posted announcing it would be integrated into Perplexity Computer 10-29. Elvar used the same word to name his repo the next day. When searching "teardown claude skill," the two get mixed up, and our X database only captured Aravind's post. Whether it was deliberate or coincidence, there is no evidence either way.

3. **Instagram "comment TEARDOWN, I DM you the link" lead-gen accounts (confirmed):** Kartik Singh (10-31, "Comment TEARDOWN and I'll DM you the free Claude skill + all 9 setup steps"), multiple "Comment \"TEARDOWN\"/\"Engine\" for the link" Reels (DfSGe7BiOPk, DfSPshjBVVw, DfXblchxWep, DfSOcPsBo42, etc.); another person made a copy with Gemini to ride Elvar's buzz (DfKziddBw1S). These accounts use the open-source skill as a hook to grow followers and collect emails.

4. **TikTok reposts (confirmed):** morenicksaraev 30k plays, github.signals 12.6k, Indonesian dikawbo, Vietnamese cunghocainha, ashtonv.ai, etc. (TikHub search "claude teardown skill"). Small volume, supplementary.

5. **Multilingual X secondary accounts (confirmed):** Chinese @ZQ444, @Lingze688 ("generate a 3D mechanical simulator you can take apart from one sentence"), @huangyun_122 (used it to make a Gatling gun); English @neil_xbt, @the_osps, @Retik_S, @Yashpatil, @lkstimm, @goyashy, @vexley_x; plus users showing off their work and tagging the author (@RKeklikian made Goped). See the mentions column on Trendshift. This explains why it didn't fall off a cliff after 11-02, but instead kept gaining a steady +260~400 per day.

6. **Skill directories / aggregator sites / blogs (confirmed):** skill-sets.md, skillleaderboard.com, typingmind.com/skills, varg.ai, gitrank.cc, hdatf.com, enterprisedna.co (11-01 daily digest, 851 stars at time of writing), thakicloud.com (10-31 hands-on long-form), several LinkedIn posts, the YouTube video "One Claude Skill Built Me a Jet Engine You Can Take Apart," and a Facebook video page. Long tail, SEO value.

7. **Trendshift JavaScript daily chart (confirmed):** 10-31 to 11-03 at #6–#12, a small boost; did not make GitHub's official trending.

8. **Not found:** paid ads, big HN/Reddit threads, mainstream media coverage.

### III. Why it took off

- **The author already had 41k followers and makes "AI video" for a living** : his job is publishing viral videos on X. This post was itself a 24-second, light-background video that breaks down one part every 2–3 seconds. SKILL.md specifically says "light backgrounds stand out in feeds" and "frame 0 is the thumbnail." The skill was designed for posting on X.

- **Timing:** anatomy (10-12), Jakub's /teardown (10-25), and the Perplexity integration (10-29) had already heated up the "have AI take a machine apart" genre. Opus 5.5 had just launched, and people were looking for examples of "what amazing things Opus 5.5 can do." teardown was the upgraded version that was "3D, moving, and with numbers."

- **Users got a video they could post:** the skill includes frame-by-frame recording at 60fps mp4 + procedural sound effects (-14 LUFS) + frame-by-frame self-checks. Users could install it and produce a postable teardown video in minutes. They then posted it to X/IG/TikTok (@creators such as @RKeklikian, @huangyun_122, and Microduck from VibeXR), creating secondary spread.

- **Free, one-line install, with 9 ready-made prompts and 3 full examples** : `npx skills add elvarofficial/teardown` lowered the barrier to "star + fork and try it." A fork/star ratio of 15% suggests people really installed it.

- **Lead-gen accounts needed it:** the "comment a keyword, get the link by DM" model needs something visually striking, free, and explainable in one sentence. teardown fit that perfectly, so several IG/TikTok accounts repackaged it at once.

### 4. Differentiation

- Compared with anatomy and Jakub's /teardown: it is **3D, drivable (buttons, sliders, WASD), with gauge numbers coming from the same kinematic function** , not just an illustration.

- Compared with machine-skill: it requires **starting from a reference image** ("machines invented from a single sentence are always mediocre") and requires side-by-side comparison of renders against the reference over 3 rounds. The aesthetic is locked down with numbers (easing curves, machine fills 60–70% of the frame, labels no smaller than 34px). machine-skill is more engineering-flavored but neither as pretty nor as postable.

- Compared with "code-to-video" skills: it adds an interactive simulator, and the simulator and video share one time function, so they never fall out of sync.

- The core difference is actually **"a video-output pipeline built for the X feed"** (borrowed from whaleyxbt/claude-motion), not physics. thakicloud's hands-on test noted it is only a kinematic model with no rigid-body dynamics.

- **Is that why it took off?** Partly. Its differentiation made the author's showcase video look good enough, and it let users make their own videos for secondary spread. But the main push to get traction came from the author's post with 347k views and the hot term it rode. The concurrent machine-skill, which is technically stronger, has only 586 stars, which is counter-evidence.

### 5. Visibility and copycats

- **Visibility: circulating within niche circles (AI creator / skill-directory circles).** Being open source, there's no revenue to show off. The author's X post had 347k views and was written up by thakicloud, enterprisedna, and a dozen secondary accounts. No mainstream tech media or top X accounts (million-follower level) specifically reposted it. The genre itself (anatomy 25k stars, Jakub's /teardown + Perplexity) was already widely discussed, so the space is crowded.

- **Copycats:** no clear clone repo found. Derivatives inspired by it already exist: VibeXR's "Microduck Teardown in WebXR" (openly inspired by Elvar), someone on Instagram made a copy via the Gemini video route, 333–361 forks, and a wave of "3D/motion-design skills" arrived in the same period (Moh4696/open-3d-skills at 262 stars, jwang771/3d-world-simulator at 218 stars).

### 6. What to borrow

1. **The skill is the content:** turn the skill's output into a video that grabs attention in the feed. The post = demo = ad. The skill locks in "how to make it look good," so users' output looks good too, and they naturally share it and bring traffic back.

2. **Long post + "👇 skill, showcase site, 9 prompts"** : a format where bookmarks outnumber likes, with the link placed in a reply.

3. **Riding the second or third wave of a hot genre:** once the genre is validated, make the upgraded version (2D → 3D, static → drivable + video), and use a name that's already trending.

4. **Standing on someone else's foundation:** runtime, recording, and sound all come from claude-motion; the author only handled the subject and aesthetic. One person produced 7 works in two days.

5. **Feeding lead-gen accounts:** free, easy-to-explain in one sentence, visually strong open-source work gets reposted organically by IG/TikTok "comment keyword" accounts.

### VII. What I'm still unsure about

- Whether the name "teardown" was deliberately chosen to ride the heat of Jakub Krehel / Perplexity.

- Whether visual-learning.app (the showcase site for 40 machines) belongs to Elvar and whether there's a monetization plan. No pricing was visible on the page.

- Whether Elvar had previously posted similar viral hits (how he built his following to 40k); not checked this time.

- The share of plays and stars brought in by each Instagram/TikTok lead-gen account can't be separated out.

- Growth has already cooled (11-06 daily +173; off Trendshift's daily chart since 11-04), and it's unclear whether it can hold; the author hasn't posted since 10-31, and 0 merged the PR.

### Sources

- [github.com/elvarofficial/teardown](https://github.com/elvarofficial/teardown) (README, scraped via Jina)

- [api.fxtwitter.com/elvar_official/status/2116192984299344372](https://api.fxtwitter.com/elvar_official/status/2116192984299344372) (original post data, 11-06)

- [thakicloud.com/tech-blog/en/tutorials/teardown-3d-skill/](https://thakicloud.com/tech-blog/en/tutorials/teardown-3d-skill/) (10-31 hands-on test and snapshot)

- [trendshift.io/repositories/328612](https://trendshift.io/repositories/328612) (daily stars, charts, X mentions)

- [x.com/jakubkrehel/status/2114379959263502712](https://x.com/jakubkrehel/status/2114379959263502712) (Jakub /teardown, 10-25); [x.com/AravSrinivas/status/2115838915647271104](https://x.com/AravSrinivas/status/2115838915647271104) (Perplexity integration, 10-29)

- [visual-learning.app/](https://visual-learning.app/)

- Instagram: [www.instagram.com/reel/DfJrX9CSfx1/](https://www.instagram.com/reel/DfJrX9CSfx1/), /reel/DfSGe7BiOPk/, /reel/DfXblchxWep/, /reel/DfKziddBw1S/

- TikTok: TikHub search for "claude teardown skill" (morenicksaraev, github.signals, rongxuanxr, etc.)

- [enterprisedna.co/resources/ai-pulse/ai-pulse-2026-11-01-tear…](https://enterprisedna.co/resources/ai-pulse/ai-pulse-2026-11-01-teardown-makes-machines-explorable/) ;[skill-sets.md/skills/elvarofficial/teardown/teardown/](https://skill-sets.md/skills/elvarofficial/teardown/teardown/)

- Our GitHub library (repo_profile / repo_day, as of 2026-11-06), X library

- Same directory gh_AirsupHQ__machine-skill.md (earlier report on the same theme)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.