Athenaion: an Athens built from code, 684 stars, the ledger is worth more than the visuals
What it is An open-source showcase repo: a Greek world built by Claude Code with three.js, pure code and no external assets, with reference images, the full prompts and a cost ledger.
Why it took off Opus 5.5 had just launched, and "building worlds with code" was a hot pattern on X. The author has 33k X followers, and the launch video got 250k views. Two weeks later a long methodology post used it as a case study and pushed it again. The public "$2,000" ledger gave Chinese accounts and media something to talk about.
CHow you can follow it Don't build another city. Copy how it's packaged: zero assets, a ledger on the first screen of the README, reference image comparisons and full prompts. Apply that to a subject nobody has packaged, or to a different tool ecosystem. Ideally, launch a tool with a spectacular showcase, rather than adding the tool after the showcase goes viral.
- Data (updated daily) · Repo created 2026-10-17 · 688 stars · 7 days +374
GitHub · stars Data 2026-10-19 → 2026-11-08 (21 days, 21 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
Athenaion is an open-source repo containing a whole ancient Greek world built by Claude Code with pure three.js code: the Acropolis, temples, the statue of Athena, 57 waterfalls, 149k trees, with not a single external texture or model. It was created in Oct 17, and the code hasn't changed since. By Nov 6 it had 684 stars and 87 forks. Below: what it is, how the stars came in two waves, how crowded the theme is, and whether and how to build a project that can earn stars like this.
A 54-second video, and the line "no assets used at all"
2026 Oct 17 evening (UTC 22:16), Leon Lin, a developer in Munich (X: @LexnLin, 33k followers), posted a tweet:
I asked Opus 5.5 to build an ancient Greek world as detailed as possible… no assets were used!
The video is a 54-second 4K cut: the camera drifts over an acropolis on a limestone cliff, a Doric temple with gold trim, sea of clouds and snowy peaks. By Nov 6 the tweet had 250k views, 3,434 likes and 1,443 bookmarks (X, 11-06). He created the repo 26 minutes after posting.
The video itself isn't rare. In October 2026, "Opus 5.5 + three.js, building a world in code" was one of the hottest plays on X: cyberpunk cities, castles in the clouds, Middle-earth, one after another. daily.dev even ran a post titled "Opus 5.5 plus Three.js is being called AGI by people who never learned Blender" (10-22). What sets Athenaion apart from them is in a table in the README.
What it actually is: a finished piece, plus a ledger
The repo holds about 45k lines of JavaScript across 168 modules. It runs locally, and you switch between nine camera angles with 1 through 9. But most people who click in are looking at these things:
- A ledger: about 45 hours of total time (from 10-16 a.m. to 10-18), 9 rounds of multi-agent iteration, 55 sub-agents (up to 12 running in parallel), about 11M output tokens and 3.8B cache-read tokens. At Opus 5.5 API prices that comes to roughly $1,800 to $2,300 (in practice it ran on a Claude Max subscription). The final 4K cut took 780 hours to render frame by frame on a laptop's integrated GPU (Radeon 13M).
- Nine reference and comparison images: the whole world was "reconstructed" from 9 AI-generated reference images, reference on the left, finished render on the right.
- The full prompt (docs/PROMPT.md): the author condensed his own remarks across the session into one prompt. The core rules: "zero external assets", "match the reference images as closely as possible, 99.9% target, especially colors", "iterate as long as it takes, don't stop until it matches", "render still frames for me first, don't render video until I approve".
- Method notes: first Claude Code wrote the rendering engine, then a art bible and module conventions. Terrain, water, sky, buildings, statues, trees and the city were then split across 12 specialized agents running in parallel, each rendering, comparing against the reference, and revising on its own.
It's a showpiece, not a tool. Nobody is going to use it for a project, but plenty of people download it to study: the fork-to-star ratio is 12.7%, while the author's other library, earshot, sits at 6.4%. CSDN and GitCode quickly published tutorials like "How to run Athenaion in 10 minutes" and "A full breakdown of Athenaion's architecture geometry".
The stars came in two waves
| Date | Cumulative stars | What happened |
|---|---|---|
| 10-17 | 0 | Launch tweet; repo created 26 minutes later |
| 10-19 | 84 | Chinese account @sundyme reposted it as a long thread, 10-18 |
| 10-22 | 267 | Aftershocks of the launch tweet |
| 11-01 | 314 | Nearly flat over the last ten days |
| 11-02 | 510 | +196 in a single day, the biggest of the whole run |
| 11-03 | 659 | Aftershocks of the second wave |
| 11-06 | 684 | Fell back to single digits per day |
(GitHub stars per day, as of 11-06; the daily figures for 10-17 and 10-18 are missing.)
Wave one: the author's own tweet, plus a Chinese long thread. The launch tweet was the main engine. The next day, Chinese AI account @sundyme (21k followers) posted a long thread titled "This is insane!", restating the 45 hours, 55 agents and $2,000 point by point. It got 48k views, with the repo link in the replies. By Oct 22 the repo had reached 267 stars, then stalled, gaining only 10 over 47 days.
Wave two: the author presented the same work again. On Oct 31, Leon Lin posted a long X article, "How I prompt AI to build good looking demos", on how he gets agents to produce good-looking demos by using "9 to 12 reference images + a 99.9% target + still frames before video". Athenaion was the headline example. The same day he turned the method into an agent skill called cook. The article got 148k views and 1,078 bookmarks. Then, from Nov 1 to 2, tech bloggers on Sina turned the article into two Chinese write-ups, and sites like aiproducthub and jxxy also reposted it. The repo gained 196 stars in a single day on Nov 2, more than in the launch wave. How much of that came from the article versus the Chinese coverage can't be separated from the outside.
Running in the background: reposter accounts and aggregator sites. At least 3 Instagram reels carried the caption "This isn't a drone shot, it's code". YouTube Shorts, Threads and LinkedIn all reposted it, and English aggregators (ohmyopus, chen.media, and pagepop's "Eight creative Opus 5.5 examples") indexed it. How many stars they brought can't be verified.
Absent: Hacker News and Product Hunt never featured it, and no paid promotion turned up.
Same author, same kind of video, why is the star count ten times lower?
The clearest comparison is the author's own work. Leon Lin has a dozen or more "scenes built in pure code" repos, all using the same formula:
| Project | Tweet performance | Repo stars (11-06) |
|---|---|---|
| verdant-forest (September) | 3,971 likes / 1.08M views, his most viral tweet | 52 |
| golden-temple | — | 141 |
| napali / open-water | — | 134 / 139 |
| athenaion | 3,372 likes / 240k views | 684 |
(Tweet data is from the demo list on the author's personal site; star counts are from GitHub, 11-06.)
Verdant forest got more than four times Athenaion's video exposure, yet its repo has only 52 stars. The difference isn't the visuals, it's what the repo contains. Athenaion's README is a ledger plus a homework assignment: how long it took, how many agents, how much it cost, which images it was matched against, and the full prompt, all laid out. The video makes people say "wow". The ledger gives them something to discuss (nearly every Chinese write-up argued over whether $2,000 was worth it). The prompt gives them a reason to open the repo and leave a star.
Looking further out, unknown authors doing the same thing show the pattern more clearly:
| Project | Scale (stars) | How it differs from Athenaion |
|---|---|---|
| xr843/insect-world, StarKnightt/jungle-trail, and others | 300–808 | Same zero-asset, procedural approach, but mostly small interactive scenes with no "multiple agents, how much it cost" story |
| rawprogress/fable-cities, jbang2004/long-wind | 86–114 | Three.js games built by AI agents; the authors had no following and it didn't spread |
| img2threejs/img2threejs | 20,341 | A tool: it turns objects from an image into pure-code three.js models, rather than being a showcase piece |
| viettranx/3dviz-pro-max and similar | 262–1,040 | Skill packs that teach agents to build 3D scenes |
| All the "Opus 5.5 builds an X city" posts on X | Not verified | Mostly just video posts, closed source, no process explained |
In short: for the showpiece slot, the ceiling depends on whether the author can put the video in front of hundreds of thousands of people. The finished repo on its own can pull a few hundred stars, but only once that exposure already exists. The ones that reach tens of thousands of stars are tools (img2threejs), not showpieces.
Copycats have already arrived
On Nov 5, agihunt published "Rebuilding the ancient Greek world with free Hy4, zero prefab assets", which recreates the same subject with Tencent's Hunyuan (I only saw the headline; details unverified). On Oct 27 someone on dev.to wrote "I built all of Middle-earth in three.js with Opus 5.5". The author hasn't stopped either: on Nov 2 and 3 he posted two new scenes, halo-above-the-clouds and casa-ocaso.
One more detail worth noting: the cook skill the author released on Oct 31 was meant to channel the work's traffic into a tool that could keep gaining stars over time. By Nov 6 it had only 40 stars (GitHub, 11-06). The second wave's 196 stars landed on Athenaion and never reached cook. The people who came for the spectacle bookmarked the finished piece and the ledger, not the method.
If it were you
A verdict: the packaging here, "large scene + zero assets + a ledger + reference comparison + full prompt", is worth learning from. But building yet another "AI built a city in code" right now isn't worth it. The subject is already saturated since October, and unknown authors doing the same thing usually stall at a hundred or two hundred stars. This time it also burns through a Max subscription's quota for a day or two.
Who it suits: people who already have a few thousand to tens of thousands of followers on X and are already posting AI demos. It also suits anyone building a tool that matches images with agents, since Athenaion is ready-made proof of demand.
Who it doesn't suit: people with no distribution channel who expect the repo to be found on its own, and anyone hoping a one-off piece will keep gaining stars over the long run. Between the two waves it gained only 47 stars over a full ten days.
How to do it:
- Stop doing Greece, Japanese temples and Middle-earth. Pick a subject nobody has packaged as "zero assets + ledger" yet: a real city block, a mechanical watch you can take apart, a famous movie scene. Or keep the same method but change the model or the ecosystem (Godot, Blender Python, WebGPU). A "same subject, how much it cost with a different tool" comparison is also a conversation people will join.
- The ledger is a must-have, not an appendix. Time spent, agent count, tokens, the dollar equivalent and failure count (Athenaion lists two blue-screen crashes) all belong on the first screen of the README. The numbers themselves get repeated, and the controversy brings a second round.
- Build the repo within a few dozen minutes of the tweet, and put the link in the first reply. Then two weeks later, retell the same work in a long methodology article.
- If you want stars that last, make the showpiece a demo for a tool, not the other way around: build the tool first, then launch it with a spectacular piece. cook has only 40 stars, which shows that adding the tool after the showpiece went viral didn't carry the traffic over.
How to get users: pair the X launch with a striking finished cut. Prepare a Chinese long thread, or have a Chinese AI account repost it (@sundyme's single post alone got nearly 50k views, and Sina and CSDN then picked it up). Reposter accounts on Instagram and YouTube Shorts will come on their own.
When to move: the "new model + build a world in code" wave rides on model launches. The round that Opus 5.5 kicked off is still warm into November. The next window worth using is the first two to three weeks after the next major model release.
What would mean we're reading it wrong: if over the next month or two, unknown authors' similar "ledger + prompt" repos also reliably pull in a few hundred to a thousand stars, that would show the packaging itself can drive growth, more than the author's following, and the direction should be upgraded. Conversely, if even Leon Lin's new scene library (halo-above-the-clouds, casa-ocaso) stalls at a hundred or two hundred stars, the subject has already worn out its appeal.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one The author's own golden-temple / napali / open-water / verdant-forest and others Open source 2026-09~10 52~141 stars; tweets with 520~3,971 likes Same zero-asset template, but smaller scenes, with no cost table and no full prompt. The Verdant forest tweet got 1.08M views, yet the repo has only 52 stars img2threejs/img2threejs Open source 2026-07-15 20,341 stars A tool skill (turns reference images into pure-code three.js models), not a showpiece viettranx/3dviz-pro-max, Moh4696/open-3d-skills, sloemo01/diorama Open source 2026-09~10 262~1,040 stars Skill packs that teach agents to build 3D scenes. Athenaion is one finished session plus its prompt. xr843/insect-world, StarKnightt/jungle-trail, Kenton-GMI/sakura-crossing Open source 2026-07~08 300~808 stars Also zero assets and procedural, but smaller. Mostly small interactive scenes, with no multi-agent or cost angle. rawprogress/fable-cities, jbang2004/long-wind Open source 2026-09 86~114 stars Three.js games built by AI agents. The author had no distribution. All the "Opus 5.5 + three.js world" posts on X Other 2026-10 Not verified Most only post a video. No open source, no process shared. - Where users come from
- Verified Author's original X post: @LexnLin 2026-10-17 22:16 UTC, 3,434 likes, 1,443 bookmarks, 250,596 views. Repo created 26 minutes later.
- Verified Relayed by Chinese AI accounts: @sundyme posted a long thread 10-18, 546 likes, 48,392 views, with the repo link in replies. @growaiskills published an X article 10-19 (758 views), and 11-01 got picked up by jxxy.
- Verified Author's methodology article + cook skill (second post): @LexnLin published the X Article "How I prompt AI to build good looking demos" at 10-31 22:44 UTC, with 1,078 bookmarks and 148,327 views, using Athenaion as the flagship example. GameSlash covered it 11-01. The repo went 11-01→11-02 stars, with a snapshot at 196.
- Verified Chinese media / SEO: Sina 11-02 articles (a tech blogger summarizing the author's long post), aiproducthub 11-01, jxxy 11-01, CSDN/GitCode tutorial posts, airebao 10-19
- Verified English aggregators and short-video reposts: ohmyopus, aicrier, aiidelist, usecases.sh, chen.media, pagepop. At least 3 Instagram reels, 2 YouTube Shorts, plus Threads and LinkedIn.
- Inferred HN / Product Hunt / paid promotion: No entry in our hackernews database. No evidence of paid promotion found.
Why Athenaion (Leonxlnx/athenaion) took off
In short: Claude Code (Opus 5.5) built an ancient Greek world in pure three.js code: an acropolis, Doric temples, a statue of Athena, 57 waterfalls and 149k trees, with no external models or textures. The repo has 45k lines of JS, 9 reference images, the full prompt (docs/PROMPT.md) and a cost table. It's a showpiece, not a tool: one commit on the day it was created (10-17), and nothing since.
Verdict: This is a content-driven hit: author X post → Chinese AI account recap → aggregator and short-video reposts, in two waves. Wave one: the author @LexnLin (33k followers) posted a 10-17-second finished-render post on 54, which got 250k views. Wave two: on 10-31 the author published the X article "How I prompt AI to build good looking demos", using Athenaion as the flagship example, alongside the cook skill launch (148k views). From 11-01 to 11-02, a batch of Chinese media outlets relayed it. Compared with the author's other scene repos, Athenaion is the largest and it published the numbers "45 hours / 55 sub-agents / about 2000 USD" along with the prompt. That made it something people could discuss and copy.
1. Data
- Source: our github database repo_profile / repo_day (as of 2026-11-06), GitHub API (2026-11-06). Created 2026-10-17 22:42 UTC. 684 stars, 87 forks (fork/star 12.7%). Last 7 days +371, first 7 days +40. Biggest single day +196 (11-02 snapshot). No suspect flag. pushed_at has stayed at 10-17 22:58 ever since.
- Curve (repo_day): 10-19 84 → 10-20 196 → 10-21 249 → 10-22 267 → 10-28 311 → 11-01 314 (mostly flat) → 11-02 510 → 11-03 659 → 11-06 684. Two steps: 10-18–10-21 (original post) and 11-02–11-03 (author article + Chinese re-coverage).
- Author Leon Lin (Munich, X @LexnLin, 33,255 followers): taste-skill 111k stars, earshot 4,054, unlazy 4,318, cinetic 1,053. Also a dozen "pure code scene" repos: golden-temple 141, open-water 139, napali 134, procedural-fire 66, procedural-alpine-lake 65, shanshui 62, verdant-forest 52 (GitHub API 2026-11-06).
- From the author's personal site data (data/demos.json in the Leonxlnx/leons-demos repo), which logs likes and views for the 75 X demos he posted: Athenaion got 3,372 likes / 239k views, ranking 2 among all his demos. Ranked 1 is September's Verdant forest (3,971 likes / 1.08M views), but the matching verdant-forest repo has only 52 stars.
2. Competitors (same idea: "AI builds an entire 3D scene/world from code" showcase repos)
| Name | Where | Launched | Scale | How it differs from Athenaion |
|---|---|---|---|---|
| The author's own golden-temple / napali / open-water / alpine-lake / procedural-fire / verdant-forest | Open source | 2026-09–10 | 52–141 stars; posts with 520–3,971 likes | Same zero-asset, pure-code template, but smaller scenes, and no cost table or full prompt in the README. Athenaion is the largest and best documented. |
| img2threejs/img2threejs | Open source (skill) | 2026-07-15 | 20,341 stars, +7 in 553 days | A tool that rebuilds objects from a reference image as pure-code three.js models. Athenaion is a finished showcase, not a tool. |
| viettranx/3dviz-pro-max, Moh4696/open-3d-skills, sloemo01/diorama | Open source (skill) | 2026-09–10 | 1,040 / 262 / 339 stars | A skill pack that teaches agents to build 3D scenes. Athenaion only published the prompt for one session. The author later packaged this method as the separate cook skill. |
| xr843/insect-world, StarKnightt/jungle-trail, Kenton-GMI/sakura-crossing, Token-Gremlin/natural-disasters | Open source | 2026-07–08 | 300–808 stars | Also procedural three.js scenes with no external assets, but mostly small interactive games or scenes. None of them tell a story about multiple agents, hours spent, or money spent. |
| rawprogress/fable-cities, bridge-mind/turbo-kart-rally, jbang2004/long-wind | Open source | 2026-09 | 46–114 stars | "Three.js games built by AI agents." The authors had no name recognition and it didn't spread. |
| Lots of "Opus 5.5 + three.js world" posts on X (cyberpunk cities, castles in the clouds, Middle-earth dev.to articles, and so on) | X / blogs | 2026-10 | Not verified | Mostly just video. Mostly not open source and the process isn't shared. |
Sources: our github database search (three.js / procedural / zero assets / entirely in code, 2026-11-06), Exa search. Verdict: this theme was crowded in October 2026, and daily.dev even had an article titled "Opus 5.5 plus Three.js is being called AGI by people who skipped Blender" (10-22). But Athenaion is nearly the only one that combines a city-scale scene, open source, and a full account of the process and cost.
3. Growth tactics (evidence)
- Author's original post (wave one, verified): 2026-10-17 22:16 UTC, "I asked Opus 5.5 to build an ancient Greek world as detailed as possible / no assets were used!", with a 54-second 4K render. 3,434 likes, 205 reposts, 1,443 bookmarks, 250,596 views, 191 replies (api.fxtwitter.com, 2026-11-06). The repo was created 26 minutes after the post.
- Relayed by a Chinese AI account (verified): @sundyme (21k followers) posted a long thread at 10-18 10:24 UTC, "This is insane!", retelling the method and numbers in detail. 546 likes, 576 bookmarks, 48,392 views, with the repo link in replies. Note: the recall rationale said sundyme posted this in 11-01, but the post was actually on 10-18, which corresponds to wave one, not 11-02. @growaiskills (542 followers) published the X article "AI built a city and the computer crashed first" at 10-19, with only 758 views. 11-01 was picked up by the AI knowledge base jxxy.net.
- 作者 X 长文 + cook 技能(第二波主因,已证实,因果是推断):2026-10-31 22:44 UTC 作者发 X Article《How I prompt AI to build good looking demos》,598 赞、1,078 收藏、148,327 浏览;拿 Athenaion 当招牌案例讲「9～12 张参考图 + 目标 99.9% + 先出静帧再渲视频」,同时发布 Leonxlnx/cook(10-31 建库)。GameSlash 11-01 报道(标题直接叫 leon-lin-cook-ai-visual-workflow-athenaion)。仓库 11-01 快照 314 星,11-02 快照 510 星,时间对得上。
- Concentrated Chinese media coverage (verified): 11-01 aiproducthub.cn (relaying a post by @Yooeec3, 5.1K reads) and jxxy.net. 11-02 Sina (Weibo tech blogger "Jia Fan – The Lost Garlic" turned the author's long post into two Chinese-language articles), and CSDN / GitCode SEO tutorial posts ("How to run Athenaion in 10 minutes" and "A full guide to Athenaion's architectural geometry"). Earlier, there was also airebao (10-19) and a related issue of Ruan Yifeng's weekly (10-23, not verified whether it mentions this).
- English aggregators / reposts (existence verified, scale not verified): ohmyopus, aicrier, aiidelist, mindpattern, usecases.sh, whatagentscando, artificialbuilds, pagepop ("5.5 creative Opus 8 examples"), chen.media (10-29 "The Prompt That Had Claude Build an Entire 3D World From Nine Pictures"). Instagram has at least 3 reels ("This isn't drone footage. It's code."), YouTube Shorts has 2, there are Russian-language reposts on Threads and Spanish-language reposts on LinkedIn, plus nitter / 24vids / openchamber video aggregators.
- Not found: HN (not in our hackernews database), Product Hunt, paid promotion. The repo is less than 60 days old, and GitHub trending doesn't cover it, so its absence there means nothing.
4. Why it took off
- Author's existing distribution: 33k X followers, plus the attention from taste-skill's 110k stars. Each of his posts reliably gets 20k–1M views, so cold-start cost is close to zero (demos.json).
- Timing: Opus 5.5 just dropped, and "Opus 5.5 + three.js builds a world from code" was a hot X trend in October. People were looking for ways to show off how far the model could go. Athenaion is the biggest entry in that wave (a whole city, 149k trees, 57 waterfalls).
- Numbers spread better than visuals: the README and replies on the tweet publicly listed 45 hours, 55 sub-agents, 45k lines of code, 3.8B cached read tokens, an API-equivalent cost of $1,800–$2,300, 13 hours of rendering on integrated graphics, and two blue screens. Chinese coverage revolved around the "$2,000" figure and the "55 sub-agents" (jxxy, sundyme, Sina), and the comment sections fought over whether the money was worth it.
- Easy to copy: 9 reference images, comparison shots of reference vs. final render, and the full prompt in PROMPT.md are all public. Forks/stars are at 12.7%, well above the author's other repos (earshot at 6.4%), which suggests plenty of people downloaded it to study or replicate it. Tutorials like "running in 10 minutes" on CSDN and GitCode came out of that.
- A same-template comparison shows the point: the author's September Verdant forest tweet did better (1.08M views), yet that repo has only 52 stars; golden-temple and napali each got about 130k stars. The difference is that Athenaion's tweet and README packaged the repo as "tutorial + ledger," and 10-31 the author then pushed it again with a long methodology essay, which brought a second wave.
- One project, used twice: the first launch rode on the video; two weeks later, when the author published a long methodology essay and a cook skill, he pulled it out again as the flagship case study. That produced a second burst of 11-02, and this one was bigger than the first (+196/day).
5. Differentiation
Other "AI writes pure-code three.js scenes" projects are either small scenes or minigames, or they post only a video and never open-source. Athenaion stands out on three points: city-scale, zero assets, and the entire production process made transparent (reference images, prompts, multi-agent division of labor, tokens and cost). The first two make the video hit hard; the third turns it from a "pretty demo" into a case study people can discuss and replicate, which gets them to click through to the repo and star it. Differentiation is part of why it took off, but not all of it. The same project from an unknown author would likely stall at a hundred or two hundred stars; compare fable-cities (86 stars) and long-wind (114 stars). The decisive factor is still the author's distribution ability, plus his reuse of his own work.
6. Visibility and copycats
- Visibility: it spread in small circles. The launch tweet had 250k views on AI-coding X, and the methodology essay had 150k views; Chinese AI accounts, Sina tech bloggers, jxxy, aiproducthub, and CSDN have all written about it. English coverage is mostly aggregator sites and repost accounts. It wasn't covered on HN or Product Hunt, and no mainstream tech outlet reported on it. It's open source and free, so revenue isn't a question.
- Copycats: agihunt 11-05 "Rebuilding the ancient Greek world with the free Hy4, zero prefab assets, results beyond expectations…" (recreates the same theme with Tencent's Hunyuan Hy4; only the headline was visible, the page showed a captcha, details unverified); dev.to 10-27 "I Used Opus 5.5 to Build the Entire Middle-earth in Three.js"; a batch of X posts along the lines of "Opus 5.5 + three.js builds X city." The author is also extending the same formula himself: halo-above-the-clouds (11-02) and casa-ocaso (11-03).
7. Takeaways
- The template "large scene + zero assets + a ledger (time, agent count, tokens, dollars) + reference comparison + full prompt" can be reused over and over with different subjects and models. The numbers themselves are the topic, and the controversy (is $2,000 worth it?) brings secondary spread.
- Build the repo within a few dozen minutes of posting, routing tweet traffic to GitHub; then two weeks later re-promote the same work with a methodology essay and a companion skill (cook).
- Turn a one-off piece of work into an installable skill (cook), so the traffic from the work can carry over to a tool library that keeps gaining stars over time.
- Chinese secondary distribution is worth a lot: one long sundyme post got nearly 50k views, and the write-ups from Sina and jxxy keep bringing SEO traffic over time.
8. Uncertain
- It's impossible to separate how much of the 11-02 jump came from the author's essay versus Chinese media; we don't have a breakdown of stargazers by time or region.
- We don't have daily star counts for 10-17–10-18 in our database (tracking starts at 10-19).
- We haven't verified the view counts on Instagram and YouTube repost accounts or how many stars they brought in; no evidence of paid promotion was found.
- The figures in the README are all self-reported by the author and haven't been independently verified; it's also unclear who the "sponsors" the author mentioned actually are.