Stringers
threerocks/article2media · threerocks

A "publish once, everywhere" Skill got 429 stars in 10 days: the feature isn't new, the new part is the "traffic code" line

What it is An open-source Claude Code / Codex Skill that takes an article or video link and automatically produces RedNote image posts, mobile long images, WeChat Official Account layouts, and vertical talking-head videos with voiceover and subtitles, with each output automatically checked.

Why it took off The author is a Chinese AI content creator with 21k X followers who already has two repos with a thousand stars. A long X post explaining the features brought only about twenty-odd stars. Then a short "traffic code" post, using his own traffic screenshots as proof, added 228 stars over two days, plus a GitHubDaily recommendation and automatic reposts by a circle of aggregator sites.

CHow you could follow it Stop building Chinese all-in-one suites. If you already have your own audience, pick one piece and go deep, or build the English version that's still empty (LinkedIn carousels, X threads, talking-head videos). Put a one-line selling point like this one on the first screen of your README, and when promoting, use "grows traffic" plus your own traffic screenshots as the hook.

Grade C: the promotion playbook is complete and worth studying, but the stars mostly came from the author's followers and referral relationships, which followers can't replicate. The Chinese-language niche is already saturated by baoyu-skills (27k stars) and others. Those who already have a content audience in the tens of thousands, or who can build an English version, could rate this a B. Chatter in small circles
Data (updated daily) · Repo created 2026-11-08 · 440 stars · 7 days +153
GitHub · stars
020040011-1011-1211-1511-1811-20440
Data 2026-11-10 → 2026-11-20 (11 days, 11 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

article2media turns an article or video into RedNote cards, mobile long images, WeChat Official Account layouts and AI-voiced talking-head videos. It's an open-source Claude Code / Codex Skill that got 429 stars in 10 days. What it does, baoyu-skills (27k stars) and Guizang's card skill (8 thousand stars) had already mostly done. It worked because of the author's own 21k-follower X account, plus one short post that switched the pitch from "saves time" to "grows traffic". Below: which days the stars came in, which posts brought them, which tactics you can copy, and which you can't.

The short post from the evening of Nov 11

On 2026 Nov 11, Beijing time 18:28, "Boss Liu" (@leo_xiaolei) on X posted a short post:

[Traffic code] If you don't know what to post, just grab some videos or articles and convert them in one click with this Skill. Publish across platforms and keep the content flowing. Link: github.com/threerocks/article2media

The attached image showed traffic data for content he made with the tool. He added a comment under it: "Whether the output gets traffic, you can check my homepage, the facts speak for themselves." That night he followed up with: "No! Everyone! Still don't get it!? It's the traffic code! The long-image summary gets almost the same traffic as the original article!! Terrifying!!" with four screenshots.

The main post ended with 21k views, 304 likes and 473 bookmarks (X, 11-18). The repo's stars rose by 116 the next day and by another 112 the day after, the two steepest days on the whole curve.

This wasn't the first time he'd promoted this repo. Two days earlier he had posted a long X article explaining the features, titled "I open-sourced a Skill that can run all my self-media: drop in articles and videos, and it produces image posts and talking-head videos". It got 69k views, more than twice the short post, and 734 bookmarks. But by the next day the repo had only 24 stars in total.

Explaining clearly what it can do reached 69k people and brought in about twenty-odd stars. Saying that using it grows traffic, and showing his own traffic numbers, reached 20k people and brought in over two hundred stars in two days.

What it is

A set of Skills that work with both Claude Code and Codex, also packaged as a Claude Code plugin, installed with two commands. You give it a link (an X long post, a webpage, a WeChat article, or a Douyin, Bilibili or YouTube video), and it outputs:

  • RedNote 3:4-image card sets (also 4:5, 1:1, 9:16), long images readable directly on a phone, and layouts you can paste straight into the WeChat Official Account backend;
  • vertical talking-head videos with local AI voiceover and word-by-word subtitles;
  • full Chinese translations, whiteboard argument diagrams, black-and-white comics, and several alternative covers.

The whole thing splits into 10 independent capabilities, so if you only want to format a WeChat article you can use that alone. Videos are first converted to a word-by-word transcript, then follow the same path as articles.

Compared with similar tools, it does two extra things:

  1. Original text, summary and translation are three separate paths. Most similar tools default to compressing the article into a summary. This one can only typeset the text without changing a word, or do a full careful translation.
  2. Every output is checked by a script. Original-text layouts are read back and compared block by block, and changing a single character raises an error. Translations are checked for omissions and numbers. Videos are checked frame by frame for dropped subtitles and loudness. Anything that fails exits with an error. GitHubDaily, when recommending it, said "What surprised me most is its validation."

Rendering is handled by scripts, not an image-generation model, so Chinese characters don't blur and numbers aren't made up. The only dependencies are Node 22 and Chrome.

Who the author is

On GitHub he goes by threerocks, on X by "Boss Liu", and his bio reads "AI R&D at a leading security company | Instructor", with an "AI top-up shop" and a PPT-making system linked below. His X account was only created in April 2026, has already posted over 9,800 times, and has 21k followers.

On GitHub he has built the same kind of thing every time: hand-drawn-styles (June, 2,726 stars), the hand-drawn video tool sweety-handdraw-video (October, 1,248 stars), with roughly 4,600 stars across the accounts combined. GitHubDaily first recommended his hand-drawn video repo in November; this is the second time they've featured him.

Before creating the repo he spent about four days on it (the long post says "the past four days"). He created the repo on Nov 8, with the last push on the 11, committing 9 times and releasing from 1.5.0 to 2.3.1. The day after it was created, someone tested it on Windows and opened 6 issues about things that wouldn't run, also suggesting fixes.

Where the stars came from

Time (Beijing)What happenedStars that day
11-08Repo created; Korea's GitHub aggregator site indexed it the same day—
11-09The author posted an X long article: 69k views, 734 bookmarksTotal through 11-10: 24
11-11 18:28The "traffic code" short post with the link: 21k views; two more posts followed that night and the next day11-11 total 59
11-12Chinese site-owner sites began reposting it; @GitTrend0x (28k followers) mentioned it in a skill roundup+116
11-13More aggregator reposts+112
11-14 to 11-15Included in a "16 open-source Skills for self-media" list; an English Instagram repost account put it in a "3 trending repos this week" roundup; @Ryrenz posted "a one-ingredient-many-dishes tool for self-media creators", 2,909 views+17, +9
11-16 15:30A dedicated @GitHub_Daily recommendation post: 12k views, 214 likes+33
11-17 to 11-18Second wave+51, +32

(GitHub daily stars; view counts are the values shown on the X post page 11-18)

There's also a whole other path: it's packaged as a Claude Code plugin, installable with one command from the plugin marketplace. Sites and bots like ClaudePluginHub, Context7, and agents.nix pick it up and package it automatically. The repo is new, yet aggregator pages in Chinese, English, Japanese, Korean, and Russian already exist. 64 forks, and the fork rate is 15%, higher than typical. Part of that comes from people installing the plugin and bots packaging it, so not all of it is real usage.

Not seen: Hacker News, Product Hunt, paid promotion, none of it. Whether anyone is pushing it on Xiaohongshu or WeChat Official Accounts, I couldn't find out. Within the 200 stars from 11-12 and 11-13, I also can't split out how many came from short posts, aggregator sites, and WeChat reposts.

Why it works

Ranked by volume:

  1. The author brings his own distribution. 21k followers, two repos with a thousand-plus stars each, and GitHubDaily has already promoted him once. This isn't a cold-start repo.
  2. The topic is exactly his followers' daily grind. His readers are mostly Chinese AI content creators and side-hustlers. Turning one article into RedNote, WeChat, and video scripts is the manual work they do every day. The bookmark counts on his three main posts are 1.5 to 2 times the likes (734/370, 473/304, 271/214), so people are hoarding it as a tool.
  3. The hook is "traffic," not "features." The comparison above already shows it: long posts that explain features get more views, yet far fewer stars than short posts that talk about traffic.
  4. The engineering details are easy for a reposter to summarize in one line. "Changing one character throws an error," "zero dependencies," "one-line install": people can copy lines like these straight into their recommendations. GitHubDaily did exactly that.
  5. Automatic distribution through plugin marketplaces and aggregator sites means the long-tail exposure comes almost for free.

So differentiation explains only half of it. It keeps this repo from being dismissed as "just another" among all the RedNote skills, but baoyu-skills, which is also an all-in-one suite, was already there with heavy feature overlap. Someone without a following who published the same code would most likely stall at a few dozen stars (inference). Supporting evidence is the author's own repos: each one spikes in the days after a post and falls back within days. The hand-drawn video repo now gains only single-digit stars a day.

Who it's competing with

CompetitorsScale (GitHub stars)How it differs
JimLiu/baoyu-skills (Baoyu)27,63920+ skills covering web-to-text, translation, RedNote images, WeChat formatting and publishing, comics, and infographics. Heavy overlap, and they can publish directly. Images go through an image-generation model. No local voiceover talking-head videos and no per-item acceptance checks.
op7418/guizang-social-card-skill (Guizang)8,148Only RedNote carousels and WeChat cover images, with 28 layout styles and strong design. Doesn't handle video, article body text, or talking-head videos.
isjiamu/gzh-design-skill4,397Only converts Markdown to WeChat formatting.
md2wechat-skill, wewrite, and others.Around 3,700Full WeChat workflow: writing, formatting, and saving to drafts. No RedNote or video.
XiaohongshuSkills, Auto-Redbook-Skills, and others.2,000 to 3,600Writes RedNote notes, generates images, and auto-publishes. Operations-heavy. Doesn't turn existing articles into finished products.
autoclip, MoneyPrinterTurbo.10k / 140kStandalone video editing and generation apps, not skills. No image-text output.
CapCut image-to-video, RedNote Red Skill.Big-company featuresBuilt-in platform features. Not a local pipeline that turns one external article into finished output for multiple platforms.

(GitHub,11-18)

This track has been the most crowded corner of Chinese skills since 2026. article2media is a latecomer: each individual capability has earlier, larger repos. We haven't seen any new repo copying it yet.

The English side is much emptier. Searching GitHub for "content repurposing," the biggest one built as a Claude Code skill, AgriciDaniel/claude-repurpose, has only 176 stars. The rest range from single digits to a few dozen (11-17, rough search).

If it were you

Verdict: C. You can learn a very complete playbook for how an open-source tool gets promoted on X, but its stars mostly come from things you can't get: the author's 21k-follower account, followers built from two repos with a thousand-plus stars, and GitHubDaily's recommendation. The Chinese track is already packed, and baoyu's single repo has 27k stars. Someone without an audience who builds one of these would most likely get a few dozen stars. If you already have a tens-of-thousands-follower content audience, you can treat this as a B.

Who it's for: developers who already have a creator or AI side-hustle audience (X, Jike, WeChat Official Accounts, RedNote, tens of thousands of followers or more) and want to use an open-source repo to grow their following and sell courses or communities. Or people with real reach in English creator circles who can build an English version, with target platforms swapped to LinkedIn carousels, X threads, Instagram cards, and TikTok voiceovers.

Who it's not for: anyone with no posting channels who expects organic GitHub traffic (repos under 60 days old don't make trending, and this one never has). Anyone who wants to build another Chinese RedNote or WeChat skill: those spots are already taken. Anyone who wants to make money from it directly: the repo itself doesn't earn anything. The author's income is in the tip shop and courses linked in his profile.

How to do it:

  1. Stop building all-in-one suites. Pick a segment that this repo and baoyu both do poorly and go deep, or switch the language or platform. For English, the one-article-to-LinkedIn-carousel-plus-X-thread-plus-voiceover idea has the largest existing repo at only 176 stars.
  2. Copy its verification approach: put engineering selling points that fit in one sentence, like "changing one character throws an error" and "every number needs a source," on the first screen of the README, so people can repeat them when recommending it.
  3. Ship it as a Claude Code plugin and list it in the plugin marketplace to collect the free long-tail from aggregators and packaging bots.
  4. On the first screen of the README, add a collage like "twelve outputs from the same article," plus a sample video.

How to get users: use it yourself for a week before launch, and collect the content and traffic data you produce with it. The first post can be a long explanatory piece, but the one that actually drives stars is the second: short, with a link, selling "more traffic / more money," and backed by your own traffic screenshot. Afterward, send the link to accounts like GitHubDaily that specialize in promoting open-source repos. One recommendation can be worth several of your own posts. Release several repos on the same theme in a row, and each new one can tap into the followers the previous repos built.

When to move: the Chinese window has basically closed. The English one is worth moving on early: once big accounts start mass-producing "content repurposing skills," the gap will be gone, probably within a few months (inference).

What would show we're wrong: ① article2media keeps gaining more than 2027 stars a day into January 20 even after the author stops posting. That would mean the product is spreading by word of mouth without relying on the author, and the grade should go up. ② Someone without an existing following, relying on a single Hacker News or Reddit post, gets an English clone of the skill to 1,000+ stars. That would mean the English gap is real and doesn't depend on who you are. ③ In reverse, if English repos like claude-repurpose are still at two or three hundred stars three months from now, that means English creators aren't using Claude Code for this, and the English path should be downgraded too.

Data and sources

GitHub: threerocks/article2media repo page, README, commit history, issues, daily stars and forks (11-10 to 11-18), the author's other repos and similar repos' star counts (11-18), English-language similar-repo searches (11-17). X: @leo_xiaolei's long post and 11-11 three-post thread 11-09, posts by @GitHub_Daily 11-16, @GitTrend0x 11-12, @Ryrenz 11-15; views, likes and bookmarks are the values shown on each post page 11-18. Aggregator sites: 3kjs, airadar, hotin, allagent, ClaudePluginHub, SkillFoxx and other listing pages; agents.nix PRs #124753 and #128532. An Instagram repost account, neura-byte, posted the Reel. Anywhere marked 「推断」 is an inference.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
JimLiu/baoyu-skillsOpen source2026-01-1327,639 star20+ skills covering URL-to-markdown, YouTube subtitles, translation, RedNote images, WeChat formatting and publishing, comics, and infographics. Heavy overlap with it. Images go through an image-generation model. No local voiceover talking-head videos and no per-item acceptance checks.
op7418/guizang-social-card-skill (Guizang)Open source2026-05-278,148 starOnly RedNote carousels and WeChat cover images, with strong design. Doesn't handle video, WeChat article body text, or talking-head videos.
isjiamu/gzh-design-skillOpen source2026-07-014,397 starOnly converts Markdown to WeChat HTML (6 themes + validation). A single step.
md2wechat-skill / wewrite / wechat-article-skillsOpen source2026-01 to 033,805 / 3,726 / 752 starFull WeChat workflow for writing, formatting, and saving to drafts. No RedNote cards or video.
Auto-Redbook-Skills / XiaohongshuSkills / xiaohongshu-skillsOpen source2026-01 to 032,484 / 3,614 / 2,049 starWrites RedNote notes, generates images, auto-publishes, and handles comments. Operations-heavy. Doesn't turn existing articles into multi-platform output.
chengyi-ai/native-subtitle-quote-imageOpen source2026-08-162,887 starPulls frames from video to make 3:4 long images. A single-point capability.
zhouxiaoka/autoclip / MoneyPrinterTurboOpen source2025-07 / 2024-0310,666 / 144,191 starStandalone video generation and editing app, not an agent skill. No image-text or WeChat output.
FredTalk / eian-riso-broll / dashi-animationstyleOpen source2026-10100 to 300 starsOnly turns a transcript into one animated talking-head segment.
doocs/md, mdnice, 135 editors, Xiumi.WebYearsdoocs/md has 13.5k stars; the rest are mature commercial products.A manual WeChat formatter, not one-click AI.
CapCut image-to-video / RedNote Red Skill.Big-company featuresLiveBigIn-platform features that assemble videos from asset libraries or offer in-platform AI skills. Not a local pipeline that turns one external article into multi-platform output.
Where users come from
  • Verified the author's own X long post: @leo_xiaolei (20,987 followers) published an X Article titled "I open-sourced a skill that can run all my media accounts" on 11-09, with 68,834 views, 370 likes, and 734 bookmarks. As of 11-10 it had only 24 stars.
  • Verified the author's "traffic secrets" short post (using traffic as proof): 11-11 18:28 with a GitHub link and a traffic screenshot, 20,894 views, 304 likes, 473 bookmarks, plus two follow-up posts. Stars rose by +116 the next day and +112 the day after.
  • Verified reposts from Chinese tech accounts on X: GitHubDaily's dedicated post got 12,105 views and 214 likes 11-16, and stars rose +51 the next day. GitTrend0x's compilation got 1,740 views. Ryrenz got 2,909 views. ModengSir and x_sanjin's alt accounts also passed it along.
  • Verified automatic reposts on webmaster sites and skill aggregators: 3kjs (11-12), airadar (11-13), hotin, allagent, Bitmiao list (11-14), Guobi (11-16), Zhiyong (11-17), full-text reposts on a good-articles collection, Russian SkillFoxx, Korean gittrendkr, and a Japanese podcast.
  • Verified English repost accounts: Instagram neura-byte's "3 GitHub repos blowing up this week" ranked #3. A skill collection from @shushant_l on Threads / X.
  • Verified Claude Code plugin marketplace and packaging bots: one-line marketplace install in the README; listed on ClaudePluginHub and Context7; agents.nix bot auto-packaged via PRs #124753 / #128532.
  • Verified the author's existing followers and past hits: hand-drawn-styles has 2,726 stars, sweety-handdraw-video has 1,248 stars; GitHubDaily previously promoted sweety-handdraw-video.
  • Inference RedNote / WeChat / Douyin: found one WeChat article, but the page requires a CAPTCHA. Couldn't check RedNote or Douyin (the TikHub legacy account ran out of balance).

threerocks/article2media: why it reached 429 stars in 10 days

Data as of 2026-11-18. Star curves come from our GitHub repo_day data. X post figures come from the official X API (checked 2026-11-18). Other sources are linked inline.

Bottom line

An established Chinese AI Skill author (GitHub threerocks / X "刘老板" @leo_xiaolei, 21k followers; earlier hand-drawn-styles at 2.7k stars and sweety-handdraw-video at 1.25k stars) open-sourced the "one piece, many posts" toolkit he uses himself. It packs the whole chain, from article or video to Xiaohongshu cards, long images, WeChat Official Account layout, and AI-voiced narration videos, into a single Claude Code plugin, with a script check on every output. The launch was driven by the author posting three times on X. The key post was a "traffic formula" using the traffic data of content made with the tool as proof. On top of that came a GitHubDaily re-recommendation and auto-reposts by webmaster and aggregator sites. The niche is crowded: baoyu-skills has 27k stars, and 歸藏's card skill has 8 thousand. The stars come mainly from the author's distribution, plus the combination of "full chain + video + verification." Taken on features alone, it isn't new.

Key facts

itemsNumbersSource
Repo created2026-11-08 08:07 UTC (the author had used it himself for about 4 days before, and the X long post says "the past four days")GitHub API
star / fork429 / 64 (fork rate about 15%)GitHub repo 11-18
Daily stars11-10 24 → 11-11 59 → 11-12 175(+116) → 11-13 287(+112) → 11-14 304 → 11-15 313 → 11-16 346 → 11-17 397(+51) → 11-18 429github repo_day
Commits / releases9 commits, 1.5.0 → 2.3.1; last push 11-11GitHub API
Author's other reposhand-drawn-styles 2,726★(2026-06), sweety-handdraw-video 1,248★(2026-10), sweety-image-art-direction 98★ and more; combined across accounts about 4.6k★github repo owner table
author on X@leo_xiaolei, 20,987 followers, account registered 2026-04-17, 9,803 posts; bio links to "AI Recharge Shop" and "PPT Production System"X API

Corroboration: the author's X and GitHub are the same person. The Juejin account XiaoLei_Liu ("Senior Frontend / Architect") lists hand-drawn-styles on its profile, and the long post about the Geewu crayon-style state diagram there is the same one as the @leo_xiaolei post on X. The GitHub profile name is LeiLiu.

1. Competitors: who does the same thing

This space is crowded. "Claude Code / Codex Skill for Xiaohongshu / WeChat Official Account content" has been a hot topic since early 2026.

NameWhereLaunchedScaleHow article2media differs
JimLiu/baoyu-skills (Baoyu)Open source2026-01-1327,639★The biggest direct competitor. It has url-to-markdown, youtube-transcript, translate, xhs-images, markdown-to-html, post-to-wechat, comic, infographic and 20 more skills, with nearly identical coverage, and it can publish directly. Images go through an image-generation model; there is no narrated talking-head video and no item-by-item verification.
op7418/guizang-social-card-skill (Guizang)Open source2026-05-278,148★Only does Xiaohongshu carousel images and WeChat cover images: 28 layouts, 10 themes, HTML to PNG. Strong design sense, but no video, and it doesn't output WeChat article bodies or video.
isjiamu/gzh-design-skillOpen source2026-07-014,397★Only converts Markdown to WeChat HTML, 6 themes, with "two-stage gate validation." It only covers the WeChat segment.
geekjourneyx/md2wechat-skill, imraywang/wewrite, aiworkskills/wechat-article-skillsOpen source2026-01 to 033,805 / 3,726 / 752★Full WeChat workflow of writing, layout and publishing, leaning toward drafting and sending to the draft box. Does not do Xiaohongshu cards or video.
comeonzhj/Auto-Redbook-Skills, white0dew/XiaohongshuSkills, autoclaw-cc/xiaohongshu-skillsOpen source2026-01 to 032,484 / 3,614 / 2,049★Xiaohongshu note writing, image generation, auto-publishing and comment handling. The focus is operations and publishing, not turning an existing article into a finished product.
chengyi-ai/native-subtitle-quote-imageOpen source2026-08-162,887★Pulls frames from video into a 3:4 long image. A single-point capability.
helloianneo/ian-xhs-carousel, chengyi-ai/cy-carousel-skillOpen source2026-10312 / 442★Contemporary Xiaohongshu 3:4 layout skills. Small in scale, single output.
zhouxiaoka/autoclip, harry0703/MoneyPrinterTurboOpen source (desktop / web apps)2025-07 / 2024-0310,666 / 144,191★Cuts long videos into short ones, generates short videos from topics. These are standalone apps, not agent skills, and they don't output image-text posts or WeChat articles.
FredTalk, eian-riso-broll, dashi-animationstyle and other talking-head animation skillsOpen source2026-10100 to 300★Only covers the "transcript → talking-head animation" segment.
doocs/md, mdnice, 135 editors, Xiumi.WebYearsdoocs/md 13.5k★, the rest are mature commercial productsA hand-built WeChat layout tool, not one-click AI.
CapCut "image-text to video," Xiaohongshu's official Red SkillBig-company featuresLiveBigCapCut makes videos from copy but assembles scenes from a material library; Xiaohongshu's Red Skill is an in-platform AI capability store. Neither is a local pipeline of "one external article → multi-platform finished products."

Summary: taken individually, any single capability (cards, long images, WeChat layout, video-to-text, talking-head video) already has earlier and larger open-source libraries. article2media's position is chaining these into one pipeline with item-by-item verification. baoyu-skills is also an all-in-one suite, but it has no locally dubbed talking-head video and no item-by-item verification.

II. Growth: where users came from (by time)

Time (Beijing)EventDataStar reaction
11-08Repo created; Korean aggregator sites gittrendkr and AgentLensHQ indexed it the same day——
11-09 15:31The author posted a long-form X article (Article) titled "I open-sourced a Skill that can run all my self-media: drop in articles and videos, and get image-text posts and talking-head videos out"68,834 impressions, 370 likes, 734 bookmarks, 55 repostsBy the 11-10 snapshot it had only 24★. Long X articles keep people reading inside the app, so few click out; bookmarks are 2× likes, a typical "save for later" pattern
11-09 to 11-13Reposts and quotes from a small account and an acquaintance: @ModengSir (6k followers), two posts from @x_sanjinHundreds of impressions eachVery small
11-11 18:28The author posted "[Traffic Secret] If you don't know what to post, just grab some videos/articles and convert them with this Skill in one click; multi-platform publishing, endless content," linked the GitHub repo directly, and used content made with it as the image as traffic proof; self-assessment: "how well the output performs, check my profile." That night and the next day he posted two more: "Traffic secret! The long-image summary version gets almost the same traffic as the original."The main post: 20,894 impressions, 304 likes, 473 bookmarks11-12 +116,11-13 +112, the steepest two days on the curve
11-12 to 11-17Auto-reposted by site and aggregator sites: Kejishi 3kjs (11-12), AI Radar airadar (11-13), hotin, allagent.wiki, the Bitmiao list "16 open-source Skills for self-media" (11-14), Guobi AI Insights (11-16), Zhiyong AI (11-17), plus a Russian SkillFoxx post, a Japanese podcast Kana & Mari, and full-text reposts of the X long article on a good-articles siteNo traffic numberslong tail
11-12 19:30@GitTrend0x (28k followers) compilation post "AI Agent Skills is blowing up again," mentioning "threerocks's one-article-multi-post"1,740 impressionsSmall
11-13 to 11-15English reposts: Instagram neura-byte "3 GitHub repos blowing up this week" ranked #3; Threads / X @shushant_l's skill roundupNo data / 948 impressionsSmall, but it got out of the Chinese-speaking circle
11-15@Ryrenz: "A self-media creator's gadget for getting more out of one piece of content"2,909 impressions, 59 likesSmall
11-16 15:30@GitHub_Daily gave it a dedicated recommendation post, singling out: "What surprised me most is its verification; change one character and it errors out"12,105 impressions, 214 likes, 271 bookmarks, 41 reposts11-17 +51, 11-18 +32, second wave
OngoingDistribution through the Claude Code plugin marketplace (claude plugin marketplace add threerocks/article2media), ClaudePluginHub, Context7 indexed it; agents.nix bot auto-packaged it (PR #124753, #128532)—People who install it fork it, which is one reason the fork ratio is high

Not found: HN (our hackernews database doesn't turn it up), Product Hunt, TikTok, paid promotion; no evidence of any of these. Whether anyone pushed it on Xiaohongshu or WeChat couldn't be verified: WeChat article pages require a captcha, and the TikHub old account balance ran out.

Correction to the recall card: the card said the author's "traffic secret" post had "about 197k views" (the card's figure was 11-11). X API measurement shows 20,894 impressions, an order of magnitude less. The highest-impression main post among the author's few is the X long article at 11-09, 68,834.

3. Why it took off

  1. The author has an existing audience, and the same playbook has already worked twice. hand-drawn-styles (2.7k★) and sweety-handdraw-video (1.25k★) are both his. GitHubDaily promoted his sweety-handdraw-video in early November (x.com/GitHub_Daily/status/2119437327663784409), and this is the second time they've pushed him. The new repo didn't start from zero.
  2. The topic was chosen where the audience feels the most pain and most wants to monetize. Boss Liu's X followers are mostly Chinese AI self-media creators and side-hustlers, and his bio links to an "AI Recharge Shop." "Turn one article into Xiaohongshu + WeChat + talking-head video" is the repetitive daily labor this group faces. Bookmarks run 1.5 to 2× likes (734/370, 473/304, 271/214), suggesting people treat it as a "tool I'll need later."
  3. It hooked on "traffic secrets" and its own traffic screenshots, not on feature descriptions. The feature-explainer long post got 69k impressions and only about 24★. What actually steepened the curve was the short linked post on 11-11, which shifted the pitch from "saves time" to "grows traffic," backed by the self-proof "long-image summary version nearly matches the original's traffic," and then +228★ over the next two days.
  4. The engineering is solid enough that reposters had things to talk about. Scripted verification (the source is read back block by block and compared, one changed character triggers an error, translation omissions are checked, and video is checked frame by frame for dropped subtitles and loudness), local VoxCPM2 dubbing with Whisper alignment, zero npm dependencies, one-line plugin install, and regression tests on model swaps. All of this can go straight into a repost. GitHubDaily picked out the verification part specifically.
  5. It rode the automatic distribution of the Claude Code plugin marketplace and skill aggregator sites. In the second half of 2026 a large batch of site and aggregator sites began auto-crawling new skills (3kjs, airadar, hotin, allagent, claudepluginhub, agents.nix). The new repo was spread across Chinese, English, Japanese, Korean and Russian pages as soon as it appeared, so long-tail exposure came for free.

IV. Differentiation, and whether it explains the growth

The differentiation has three layers:

  • End-to-end, one stop: the input accepts both articles and videos (Douyin, Bilibili, YouTube, converted to transcripts first), and the output includes both image-text and video. Most competitors cover only one segment.
  • Three separate paths: "original / summary / translation": competitors default to compressing articles into summaries. This one can typeset without changing a word, or do a full faithful translation, and uses scripts to guarantee nothing was altered or dropped.
  • Rendering is handled by scripts rather than an image model, and every output is verified automatically: Chinese characters don't stick together, numbers aren't made up, and anything that fails exits with a non-zero code.

Why it took off: half of it is the framing. The differentiation gave people something to say about it. GitHubDaily named the acceptance criteria, and that kept it from reading as just another Xiaohongshu card skill in a crowded niche. But the similar all-in-one repo baoyu-skills already existed, and the feature overlap is high. What decided 10-day star growth against 400 was the author's own 20k-follower distribution, plus the "traffic secrets" pitch that tied the tool to making money. Without this author's account, these features would most likely have stalled at a few dozen stars. Another hint: this author's repos all spike the day of the post and fall off within days. sweety-handdraw-video is now gaining only single-digit stars a day.

5. Visibility and copycats

  • Visibility: circulating in small circles. Chinese AI self-media X accounts (the author, GitHubDaily, GitTrend0x, Ryrenz, x_sanjin) and a batch of auto-aggregator sites shared it. On the English side, only Instagram and Threads repost accounts. No media coverage, no Hacker News or Product Hunt, and it did not reach GitHub trending (the repo was not yet 60 days old, so it could not qualify). It is open source and free, and the author does not earn from it directly. Monetization lives in the "AI top-up shop" and courses linked from the X bio.
  • Copycats: I found no new repo copying it (checking GitHub repos created after 11-09 for "multi-post / article-to-Xiaohongshu / narration" style projects turned up only mdldm-maitian, a multi-platform publishing hub with a different focus). Instead, it followed into a niche that already had plenty of similar tools.

6. What to borrow

  • Pushing an open-source tool on X: long posts that explain it clearly, short posts with the link, the selling point reframed as "grow traffic / make money," plus screenshots of traffic the author earned using the tool. In this case, the short post brought in far more stars than the long article.
  • Put engineering selling points that a reposter can turn into a single sentence (acceptance criteria, zero dependencies, one-line install) on the first screen of the README, so accounts like GitHubDaily can recommend it again.
  • Package it as a Claude Code plugin and list it in the plugin marketplace. That gets it picked up automatically by aggregators and bots like agents.nix.
  • The same author keeps releasing skills on one theme (hand-drawn, video, content production). Each new repo inherits the followers and recommendation relationships built by the previous ones.

VII. Uncertain

  • The source of the +112★ surge from 11-12 to 11-13 can't be broken down precisely. Was it the aftershock of the "traffic secrets" post, aggregators, or reposts on WeChat / Xiaohongshu? GitHub does not expose stargazer timestamps to us (the API returns 404), and WeChat article pages require a captcha.
  • I couldn't find whether anyone promoted it on Xiaohongshu or Douyin, or whether the author used it to grow a Xiaohongshu account (the TikHub legacy account balance ran out).
  • Actual usage is unknown. Part of the 64 forks come from plugin installs and packaging bots, so they can't be read as active users.
  • Whether the author and GitHubDaily have a formal partnership (they recommended it twice) is unverified.

Sources

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