One sentence to build a turbofan engine you can take apart: machine-skill hit 586 stars, but ranks third in its category
What it is An open-source Claude Code skill: describe a machine in one sentence, and Claude builds a realistic-scale, photorealistic 3D web page that you can cut open, disassemble and break, and one command records it into a shareable short video.
Why it caught on The founder first posted a month's worth of similar works on X to gather topics. The day it launched, the same topic was exploding in anatomy content, and programmer Yuga and Instagram traffic accounts reposted its built-in videos as content.
CHow you could follow it Stop making "one sentence to build a machine" and pick an object nobody has covered yet, "take it apart and look inside" (molecules, circuits, buildings, etc.), and follow its three-piece kit (a small model you can see the formulas in, product-shot style visuals, built-in recording). Before launch, post works on X in a row to gather material; after launch, catch the PRs.
- Data (updated daily) · Repo created 2026-10-12 · 597 stars · 7 days +129
GitHub · stars Data 2026-10-14 → 2026-11-08 (26 days, 25 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
A small Los Angeles company building digital twins for factories open-sourced a month's worth of "have Claude build a machine you can take apart" posts on X as a Claude Code skill. In 25 days it reached 586 stars and 77 forks. The work is genuinely hardcore, but the day it launched, another project on the same topic jumped from 99 to 2,000 stars, and half a month later another one hit 2,300 stars in a week. Below: what it is, where the stars came from, why it stopped at this level, and whether it's still worth building a similar project today.
Oct 12, 4 PM
On Oct 2026 12 23:03 (UTC; 4 PM in Los Angeles), Airsup founder Konstantin Saifoulline posted a long post with a video on X:
Name a machine. Claude builds the real thing. We open sourced the skill behind our lab: 50+ machines built with Claude Opus 5.5.
In the video, a line of text turns into a turbofan engine, assembled one part at a time while the thrust gauge climbs to about 340 kN. Then it gets cut open, a fan blade is flung off, and a rocket engine ignites and burns through its throat. 8 minutes later, AirsupHQ/machine-skill goes up on GitHub. By 11-06 the post had about 12k views and 79 likes. The author has about 4,800 followers on X (11-06).
On the same day, another open-source skill that explains "how machines are built," wheresryan22/anatomy, went from 99 stars the day before to 2,009 stars, and the next day 5,381 (daily GitHub stars). The last commit to machine-skill was on 22:52, the "Rename to Machine Skill" commit, and it went live an hour later. Whether that timing was deliberate or a coincidence, there's no direct evidence either way.
What it is
Once installed, type this into Claude Code:
`` /machine a turbofan jet engine, the kind that pushes a widebody off the runway ``
Claude first looks up the real machine's dimensions and principles, then writes a standalone HTML file with three.js: a 2.9-meter titanium fan, 10 compressor stages, and 22 fuel nozzles, 3,597 parts in all. Drag the throttle and the thrust is computed from the Brayton cycle, not looked up. Cut it open and the cross-section is solid metal. "Break it" by flinging off a blade, and the page shows how big an imbalance load the casing has to take. Every page can explode, section, X-ray, and slow-mo.
The other half is filming: the repo ships a recording script. One command captures the page as a 1080×1350 vertical video with a soundtrack, ready to post. The video at the top of the README was recorded this way.
The gallery has more than engines: an API 610 process pump, a multistage centrifugal compressor with its surge line, and captions reading "cut open for plant training". That's exactly Airsup's core business customers. It's open source under MIT, charges nothing, and serves as Airsup's showcase.
Where the stars came from
| Date | What happened | star |
|---|---|---|
| late September | The founder posted a string of "I had Opus 5.5 build a XX you can take apart": a fusion reactor (9-27, about 43k views), a Raptor 3, and a five-axis CNC. A batch of "Opus 5.5 showcase" sites reposted them. | Repo not yet created |
| 10-12 | Launch post + repo creation; anatomy exploded the same day | +134 on day one (Trendshift) |
| 10-15 | Programmer Yupi (@yupi996) posted in Chinese with the install command | +83 in a single day, the peak day |
| From ¥10-16 | Instagram: "comment TURBOFAN / MACHINE and I'll DM you the link"; auto-indexed by skill directories | 10–25 per day |
| 11-03 – 11-05 | Similar teardowns went viral, and people searched for similar skills | +22 / +47 / +31 (inferred to be carried by the others) |
A few worth a closer look:
Yupi's post. On Oct 15 he wrote: "Come take a look. How real is a 3D turbofan that Claude builds from one sentence... it even flings off a fan blade, and the page computes how much force the engine has to take! The video was recorded from the engine that ships with the repo~". The post itself got about 5,700 views and 63 likes (X, 11-06). Not big, but it was the day the repo gained the most. Note the last line: the video he used was recorded by the skill itself.
Instagram traffic account. @chenbuildsai posted a Reel: "comment TURBOFAN and I'll DM you the repo link". Behind it, chen.media wrote a tutorial (10-19). The first half is free, and the install steps in the second half require an email sign-up. In other words, other people used this free project to grow their own email lists. These secondhand creators were its main spreaders.
Skill directories. skills.sh, SkillsLLM, Skill Leaderboard and others index it automatically, and one line, npx skills add, installs it. This is long-tail traffic, not a breakout.
Not found: ads, Hacker News, dedicated Reddit threads, Product Hunt, press coverage.
Why it only reached 586
Same subject, and the numbers differ by dozens of times across three projects:
| Project | Launched | stars (about 11-06) | How it's done |
|---|---|---|---|
| wheresryan22/anatomy | built 10-08, breakout 10-12 | 25,444 | Isometric SVG illustrations explaining what machines are made of; no physics |
| elvarofficial/teardown | 10-30 | 2,346 (one week) | One reference image → a teardown-able 3D simulation + video with sound; the author has about 37k followers on X, and one post reached 160k views |
| AirsupHQ/machine-skill | 10-12 | 586 | One sentence → photorealistic materials + physics solver + built-in video recording |
Three things stand out:
- Users want "a stunning visual from one sentence", not physics accuracy. The simplest one, anatomy, got 40 times as big. machine-skill's solver and measured materials make it more persuasive when shared ("the 340 kN is computed"), but that isn't what drove its reach.
- The author's reach sets the ceiling. The teardown author has 37k followers, and one post reached 160k views, four times machine-skill's in a week. machine-skill's launch post reached only 12k views.
- No one kept it going after launch. The last commit was on Oct 13. By Oct 14, people were filing PRs to fix black render blocks, headless GPU on Windows, and opencode support, and as of 11-06 not one had been merged. The author's attention went back to his main business (his recent X posts are all about factory and machining-shop simulations). When the second wave of attention came, teardown caught it.
Others are following it
- teardown (10-30) does nearly the same thing; there's no evidence it drew on this one.
- gbtweilai-star/gbt-machine-distill says outright that it "distills" machine-skill's approach; iTechwu/sensteed-machine-skill ports it to the DeepSeek Harness.
- 77 forks, with a fork-to-star ratio of about 13%, higher than typical gawk-and-go projects. Some people have pulled it into their own projects through PRs, and someone on dev.to tested it with Codex.
- The wider circle: 3D creative scenes (3dviz-pro-max, 1,040 stars), self-assembling architecture (703 stars), human-body 3D breakdowns (human-atlas, 4,479 stars), and "code-to-video" skills (400–1,700 stars). The theme of "Claude builds a working thing from one sentence" was already done many times between September and November.
If it were you
My take: This theme already has three top-tier players. Building another "one sentence, one machine" is unlikely to match machine-skill's numbers. What's worth learning is its launch playbook, not its subject.
Who it's for: People who already have content and work in a specific vertical (chemistry labs, medical anatomy, architectural structure, circuits, geology) and can apply "one sentence generates something you can take apart, run, and film" to their field. Or people with several thousand real followers in the Chinese or Japanese-speaking community who can drive the first wave themselves.
Who it's not for: Anyone hoping to earn stars by "building a better machine-teardown skill"; anatomy and teardown already own that keyword. Anyone with no distribution channel who expects GitHub Trending to deliver traffic on its own; machine-skill never once made it onto Trending.
How to do it: Change the subject, not the format. Pick a "look inside" subject nobody has covered yet (molecules and reactions, circuit boards, building structures, non-human organisms), and follow its three-part formula: every readout is backed by a small model whose formulas you can read; the visuals look like product photography before they start moving; and a built-in recording script outputs vertical video with a soundtrack. The other route is to fill the gaps it left: Windows and opencode compatibility, render fixes. Some of these PRs are already open and the author hasn't merged them, so you could maintain a version that merges them. But stars on that path are usually limited.
How to get users: Follow its rhythm. For two to four weeks before launch, post three to five "I had Claude build a XX you can take apart" threads on X, each with one video, to build up the theme and material. On launch day, put a video and a live gallery at the top of the README, with the install command as the first line. After launch, proactively send your recorded videos to Chinese tech accounts and to the "comment the keyword for the link" Instagram type of accounts. They need material, and your output is ready-made material. For the first two weeks after launch, keep up with PRs and issues.
When to move early: This theme's buzz tracks new model releases. Since September there have been two waves so far (anatomy in mid-October, teardown at the end of October). The next wave will probably come around the next major model release, so have your work and recordings ready before then (inferred).
Where this read could be wrong: ① Before the end of November another skill on the same theme, with a different subject, hits a thousand stars in a week. That would mean the "change the subject" path is still open and deserves more effort. ② Conversely, teardown is still under three thousand stars in November and machine-skill is back to single-digit daily gains. That would mean the theme has passed its peak and changing the subject won't help much. ③ Anthropic or another model vendor puts "one sentence generates a teardown-able 3D simulation" into their official product examples. Then third-party skills lose their position.
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 25,444 stars / 2,125 forks Isometric SVG illustrations with optional WebGL, explaining machine construction; no physics solver, not photorealistic, but simpler and 40 times bigger in reach elvarofficial/teardown Open source 2026-10-30 2,346 stars / 361 forks (7 days); the author has about 37k followers on X, and a post reached 160k views Starts from a reference image, light feed-style design, with sound effects; machine-skill starts from one sentence, with a hardcore solver and photorealistic materials viettranx/3dviz-pro-max Open source 2026-09-10 1,040 stars General creative 3D scenes; not machine-specific, no physics computation lhlGitHub/threejs-architecture-effects Open source 2026-09-22 703 stars Self-assembling 3D architecture; not machines Code-to-video skills (claude-animated-shorts, motion-video-kit, opus-video-skills, etc.) Open source 2026-09 to 11 400–1,700 stars Video only; no interactive physics simulation Prompts sent directly to Claude Opus 5.5 Big-company features 2026 A lot of bare-prompt work on X The skill bakes in aesthetics, materials, solver, critique rounds and recording, so output quality is consistent SIMULIA and other professional simulation / CAD Other Already established Industrial software Real CFD/FEA, professional and expensive; machine-skill is a simplified physics model for explanation and display - Where users come from
- Verified founder's X lead-up posts: in late September @konstantinsaifo posted a string of "I asked Claude Opus 5.5 to build X you can take apart". The fusion post reached 43k views (whatpeoplemade.com) and was reposted by many aggregator sites showcasing Opus 5.5 work.
- Verified X long post: 10-12 23:03 UTC, a video launch post titled "Name a machine. Claude builds the real thing". The repo was created 8 minutes later, +134 stars on day one.
- Inferred Riding the timing of a trending topic: anatomy 10-12 jumped from 99 stars to 2009, then 5381 the next day; machine-skill was renamed and submitted the same day 22:52, going live within an hour
- Verified Chinese mega-account reposts: Chinese dev Yupi (程序员鱼皮 @yupi996) posted a long post on 10-15 with the install command, which matches the single-day +83. Threads, Facebook and Alibaba Cloud's developer community also relayed it.
- Verified Instagram "comment a keyword, get the link by DM": @chenbuildsai ("Comment TURBOFAN") and @synccreation 10-16 ("comment MACHINE"). Chen Media also wrote an email-unlock tutorial.
- Verified Skill directories and aggregators: listed on skills.sh, SkillsLLM, Skill Leaderboard, TypingMind, auto-packaged by agents.nix, skillfoxx and others.
- Inferred spillover from similar hits: teardown went from 11-02 to 11-05 stars over 353 to 1878, while machine-skill gained +47 in a day, and factory-skill also grew at the same time.
AirsupHQ/machine-skill: why it took off
Research date 2026-11-06. Data: our github database (repo_profile / repo_day, daily since 2026-10-16), the GitHub API, Trendshift, Jina/Exa/Brave search and page text.
One line
An open-source Claude Code skill: /machine a turbofan engine from a wide-body airliner. Claude first looks up real aircraft dimensions and physics, then builds a photorealistic 3D simulation in a single HTML file with three.js (exploded views, cutaways, X-ray, "break it"), with every readout computed by a small solver. It also includes a recording script that outputs a 28-second short video ready to post, in one command. The author is Airsup (Los Angeles), a digital twin startup for manufacturing, founded by Konstantin Saifoulline.
Bottom line: it is an open-sourced skill built from the portfolio that the founder accumulated through a month of Opus 5.5 "build a XX you can take apart" viral posts. It launched on the day the similar skills wheresryan22/anatomy exploded (25k stars), and "Claude builds a machine you can take apart from one sentence" became a hot topic. That heat, plus secondhand content creators (Chinese X accounts, Instagram "comment a keyword, get a DM link", aggregator and tutorial sites), pushed it up. Its differentiators (a real physics solver, photorealistic materials, built-in video recording) are real, but its scale is "niche": 586 stars, while the similar anatomy has 25k and teardown reached 2300 within a week. It works more as a lead-gen prop for Airsup's core business than as a money-maker itself.
Data
| Metric | Value | Source |
|---|---|---|
| Repo created | 2026-10-12 23:10 UTC | GitHub API |
| star / fork | 586 / 77 (fork/star about 13%) | Data card 2026-11-06 |
| Day one | +134 (10-13), then +60, and from +49,10-16 onward, 10–25 a day | Trendshift daily table |
| past 7 days | +142 (first 7 days +91); 11-03 → 11-05 three-day +22 / +47 / +31 | Data card / repo_day |
| Maintenance | Last commit 2026-10-13; 4 external PRs/issues (10-14, fixing render black blocks, headless GPU on Windows, opencode support) remain unmerged | GitHub API |
| Same author | airsup-lab 56 stars (created 9-28, a machine portfolio), factory-skill 45 stars (launched the same day as machine-skill), youbuild 1 stars | GitHub API |
1. Competitors
| Product | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| wheresryan22/anatomy | Open-source Agent Skill | built 2026-10-08, breakout 10-12 | 25,444 stars / 2,125 forks (10-11 has only 99 stars, 10-12 2,009,10-13 5,381; repo_day) | Isometric SVG illustrations with optional WebGL effects, explaining "how a machine or system is put together". No physics solver, not photorealistic. The topic is more recent and the scale is 40× bigger. Trendshift even mistakenly attached anatomy's description to the machine-skill page, so the two were treated as one category. |
| elvarofficial/teardown | Open-source Claude Code skill | 2026-10-30 | 2,346 stars / 361 forks (repo created 7 days ago; GitHub API) | Nearly the same thing: one reference image → a 3D simulator you can take apart + a video with sound. The author Elvar (about 37k followers on X) posted it, with 160k views and 2,519 likes (thakicloud.com 10-31 snapshot). It stresses "starting from a reference image", a light background that suits feeds, and one input driving everything. machine-skill stresses "starting from one sentence", a real solver, and photorealistic materials. |
| viettranx/3dviz-pro-max | Open-source skill | 2026-09-10 | 1,040 stars | General-purpose three.js/Blender creative 3D scenes, not specifically machines |
| lhlGitHub/threejs-architecture-effects | Open-source skill | 2026-09-22 | 703 stars | Self-assembling architecture 3D, not machines |
| mromidev/claude-animated-shorts, echris6/motion-video-kit, opus-video-skills and others | Open-source skill | Late September – November | 400–1,700 stars | The "code-to-video" category, which partly overlaps machine-skill's video-making function |
| ashemag/human-atlas | Open source | 2026-09-05 | 4,479 stars | A human-body 3D dissection browser. Same "take it apart and look" experience, but the subject is the human body |
| airsup-lab (in-house) | Open source / web airsup.ai/lab | 2026-09-28 | 56 stars | The company's own portfolio. machine-skill is the open-sourced method behind it |
| Directly prompting Opus 5.5 with one sentence | The large model itself | — | — | 9–October's X feed is full of "I asked Opus 5.5 to build a XX you can take apart" posts that are bare prompts. The skill's value is in locking down the aesthetics, materials, solver, critique rounds and recording into a repeatable pipeline. |
| SIMULIA and other professional simulation / CAD | Professional software | Already established | — | Real CFD/FEA is a different level entirely. machine-skill itself says its physics is simplified |
Landscape: In October, "Claude builds a machine you can take apart from one sentence" was already a crowded hot topic. The leaders are anatomy (25k) and teardown (2,300); machine-skill (586) sits in the third tier.
2. Growth levers (where users come from)
- The founder's X groundwork (confirmed): from late September, @konstantinsaifo posted repeatedly "I asked Claude Opus 5.5 to explain how X works by building one you can take apart": a fusion reactor (9-2.74M views, 320 likes; whatpeoplemade.com), the Raptor 3, a five-axis CNC, and others. These posts were widely reposted by "Opus 5.5 case/prompt" aggregators (felo.ai, frontiernow, oneshotted, loveimage, progressing.tech and others), building up awareness of the topic. The same posts went out on LinkedIn.
- Launch post (confirmed): on 10-12 23:03 UTC, a long X thread "Name a machine. Claude builds the real thing. We open sourced the skill behind our lab: 50+ machines…" with a video, and the repo was created 8 minutes later. +134 stars on day one. Engagement numbers were not captured.
- Riding anatomy's breakout (timing confirmed, intent unverified): anatomy went from 99 stars to 2,009 on 10-12, then 5,381 the next day. machine-skill's last-minute commit was 10-12 22:52 "Rename to Machine Skill", live within an hour. factory-skill shipped the same day. Searches and shares for "one sentence to generate a machine" are at their peak.
- Chinese mega-account repost (confirmed): Yupi (程序员鱼皮 @yupi996, self-described as 2M followers across platforms) posted a long post on 10-15 titled roughly "Come see how real a 3D turbofan built by Claude from one sentence is", with the install command attached. It matches the single-day peak of +83 (10-15) in our database. Taiwan's Threads (@hsu_wenyen 10-13), Facebook fan pages and Alibaba Cloud's developer community (11-06) also relayed it.
- Instagram "comment a keyword, get the link by DM" (confirmed): @chenbuildsai (Chen Media) posted a Reel: "Comment 'TURBOFAN' and I'll DM you the repo link", and @synccreation 10-16 "comment 'MACHINE'". Chen Media also wrote 3 email-unlock tutorials (10-19, 10-23,11-04 for the one covering teardown), using the open-source skill as a hook for their own mailing list.
- Auto-listed in skill directories and aggregators (confirmed): the skills.sh ecosystem (
npx skills add), SkillsLLM, Skill Leaderboard, TypingMind, an auto-packaging PR from agents.nix, a candidate issue for ProSkills, skillfoxx (Russian), mindpattern, aiezz and others. This is long-tail passive traffic. - Secondary bump from 11-03 (inferred, not verified): 11-02 → 11-05 averaged +22 / +47 / +31 a day, while teardown went from 353 to 1,878 over the same period, and the in-house factory-skill also gained 12 stars. The most likely cause: after teardown took off, people searched for similar skills and articles compared them side by side (aaronorelup.com 10-31 tested teardown; most.tw and explainx 11-06 wrote teardown articles), which pulled machine-skill along. It doesn't look like the author pushed it again: the author's recent X posts promote Airsup's factory and machining digital-twin simulations.
- Not found: ads, HN, a dedicated Reddit thread, Product Hunt, or press coverage. Our x-radar local corpus also has no related posts.
3. Why it took off
- Topic timing: after Opus 5.5 launched, "have the model build a real machine you can take apart and run" became one of the most shareable demos on X. anatomy's surge on the same day shows the demand was there.
- Founder brings the material and the persona: over September, they built up 50 machines (airsup.ai/lab), and the README's first screen is a video, with every machine in the gallery runnable online. Hard-core details like "the real thrust of 340 kN is computed from the Brayton cycle" are built for sharing.
- Built-in shareable output: the skill itself outputs "a video ready to post" (
record.mjs, 1080×1350 vertical, with background music and anti-flicker frames). Users can post right after use, and every post is an ad for it. That's why secondhand creators (Yupi, the Instagram accounts) were willing to make content with it. - Low install barrier: a single
npx skills addline, compatible with Claude.ai uploads, Codex, Cursor and opencode. A fork rate of 13% shows many people actually took it and modified it (MushroomDAO, maiacity, LionDriver and others brought it into their projects via PRs; someone on dev.to tested it with Codex). - Why it stopped at 586: anatomy drew most of the attention that same day; a later teardown had a 37K-follower author and 160K views on a single post; after the machine-skill launch, 0 maintenance and unmerged PRs pulled the author's attention back to their main job (factory digital twins).
4. Differentiation
- Real physics: each machine gets a small solver (Brayton cycle, rocket combustion chamber/nozzle, slider-crank, pump curves, etc.;
references/solvers.mdgives the real constants). Every number on screen is computed, and "How it is computed" shows the equations. anatomy is illustration, and teardown says "readouts and motion come from the same function," but machine-skill has the strongest engineering flavor (refinery pump API 610, compressor surge line), aimed at "factory training." - Photoreal materials + solid cross-sections: metal materials with measured reflectance, a real-time shader for solid cross-sections (no cap geometry needed), studio lighting. It looks like product photography.
- Starts from one sentence (teardown needs a reference image).
- Is running it the reason? It's why people shared it (the visuals and hard numbers), but not the main reason it took off. Traffic came mostly from topical timing and secondary creators. Within the same topic, anatomy is simpler (SVG) and got far more traction, which suggests that at this stage users wanted "a striking image from one sentence," not physical precision.
5. Visibility and copycats
- Visibility: circulating in small circles. No public revenue (MIT open source, backed by Airsup's B2B business). There are Chinese accounts with large followings (程序员鱼皮), Instagram traffic accounts, and a batch of tutorial and aggregator sites that wrote about it, but no mainstream outlet or top English X account has covered it specifically. The topic itself (anatomy, teardown) has been widely discussed.
- Copycats/borrowing: elvarofficial/teardown (10-30, same topic, now 4× its star count; whether it referenced this one is unverified); iTechwu/sensteed-machine-skill (ported to DeepSeek Harness); gbtweilai-star/gbt-machine-distill (explicitly "distills" machine-skill's approach); and a large number of direct forks (77).
6. What to borrow
- Spend a month posting "I had the model make XX" on X to gather topics and material, then open-source the workflow behind it as a skill, with the video and an online gallery in the first screen at launch.
- Make the skill's output shareable by default (built-in recording, vertical format, background music). Every use is a distribution event.
- Launch in the one or two days when a similar hit is peaking, when search and reshares are at their highest.
- Use the open-source skill as a lead-generation tool for the B2B company: Airsup sells factory digital twins, and the skill and lab serve as its showcase (factory-skill and the warehouse/machining factory simulations all point back to the core business).
- The downside: after launch, 0 maintenance and PRs don't line up, and when the second wave of buzz (teardowns) arrives you can't keep up.
7. Uncertainties
- The post's views/likes and the author's follower count were not retrieved (the X page requires login).
- No public data on whether, or how many, clients came to Airsup's core business from it.
- I couldn't pinpoint which post or article caused the small peak from 11-03 to 11-05.
- Whether the launch was deliberately timed to anatomy's breakout day: only a coincidence in timing (the rename commit landed on the day anatomy broke out), with no direct evidence.
- Whether teardown referenced machine-skill: the approaches are highly similar, but there is no evidence.
Sources
- github.com/AirsupHQ/machine-skill (README, commits, issues/PRs, forks)
- trendshift.io/repositories/301925 (daily star growth)
- x.com/konstantinsaifo/status/2109782159250608604 (launch post, 2026-10-12)
- www.whatpeoplemade.com/i/a-fusion-reactor-you-can-take-apart… (9-27 fusion reactor post: 43K views)
- x.com/yupi996/status/2110572118375309518 and x.com/yupi996/status/2110572118710882442 (程序员鱼皮 10-15)
- www.instagram.com/reel/De1EGZ-Dd1N/ (@chenbuildsai, "Comment TURBOFAN"), www.instagram.com/reel/Dek5l7xqatx/ (@synccreation)
- chen.media/guides/machine-skill-one-sentence-in-a-real-3d-je… (10-19)
- github.com/wheresryan22/anatomy; our GitHub repo_day (anatomy 10-11 99 → 10-13 5,381)
- github.com/elvarofficial/teardown; thakicloud.com/tech-blog/en/tutorials/teardown-3d-skill/ (Elvar post, 163K views)
- aaronorelup.com/ledger/teardown-lab/, dev.to/alicia_lu/i-tested-machine-skill-with-codex-two-three…
- github.com/gbtweilai-star/gbt-machine-distill, github.com/iTechwu/sensteed-machine-skill
- www.airsup.ai/, www.airsup.ai/lab/skill