Sold over 4000 in one day: how a video-channel selling Skill hit 781 stars in 11 days

On Oct 27 at 8 p.m., a Chinese X blogger with just over 5k followers posted a long article titled "Officially open-sourced! Using this skill to fully automate WeChat Channels selling, I sold over 4000 in one day." The next day, his GitHub repo, created only the day before, had gained 264 stars. This article explains what the Skill is, where the stars came from, why comparable projects with the same tech got only a fifth of its stars, and which parts to learn if you want your own open-source project to take off the same way.

One sentence to the Agent, out comes a Jianying draft

story-studio is a Chinese Skill for Claude Code, Codex, and Cursor. The usage the author demonstrates in the long post is this one sentence: "Use $story-studio, write only a spoken-narration script about the medical pioneer Lin Qiaozhi... First verify the key dates and figures of her life, then give me 3 different golden opening hooks to choose from." The Agent pulls several openings from the built-in hook library, such as the "consecutive numbers" type: "Lost her mother at 5, refused marriage at 18, practiced medicine for 60 years, personally delivered 50k+ babies..."

Once you pick an opening, it runs the rest of the pipeline: it writes the script, generates the voiceover with Doubao speech synthesis, aligns subtitles to the audio word by word locally with sherpa-onnx, splits the storyboard and generates images in one of three art styles (historical realism, thick-paint comic, black-and-white documentary), and finally exports a Jianying Pro draft through the open-source library pyJianYingDraft, with a slow push-in keyframe already applied to every image. You open Jianying to fine-tune and export.

It targets a very specific niche: spoken-narration accounts on WeChat Channels that cover history and biography, with a middle-aged and older audience and book affiliate links under the videos. In the long post the author says biography and social-science books sell for ¥49 to ¥88, with commissions of 30% to 50% (the author's own claim, which I couldn't verify).

The stars came from one long post

Source: Trendshift daily table; GitHub. 11-06 is 781 stars and 171 forks. It has never made GitHub's main trending page.

The first long post reached about 506k views and 2787 likes by 11-06, roughly 100 times the author's follower count. The sequel got 138k views. In Nov 1, another Chinese AI blogger, @DDJCXX, posted a roundup titled "8 Skills that let Claude Code make videos," with 107k views, which Trendshift counted as one mention of this repo. Beyond that are secondary write-ups on blogs, several navigation sites, and a YouTube explainer video. There's nothing on HN, Reddit, English-language communities, or mainstream media.

The fork rate is 22%, much higher than comparable projects. The reason isn't that people are modifying it: the README tells you to "have your Agent clone it into the skills directory," so many people forked it as soon as they installed it. I spot-checked 5 forks, and none had added a single commit beyond the original (GitHub, 11-06).

Why only a fifth of the stars, for a project with the same tech

The pipeline itself is nothing new. Doubao TTS, local alignment, image generation, and pyJianYingDraft for draft output: InsightCut, LingtiStudio, otopilot, and image-story-video-wizard all do this, with stars between 120 and 400 (GitHub, 11-06). The bigger general-purpose project MoneyPrinterTurbo has 140k stars, Pixelle-Video has nearly 30k, and CapCut's official product has an AI assistant and "AI text-to-video."

It adds two things.

First, the corpus. The repo includes the verbatim scripts from 5 top WeChat Channels accounts: 80 posts from 人间人物志, 80 from 闲读人间往事, 64 from 红颜旧事志, 37 from 通史观人物, 35 from 策山读书, plus 57 more history-narrative posts with hook annotations, and 12 style reference images. Readers aren't installing an "AI video tool" so much as a way to copy accounts that already work.

Second, a money story. Similar READMEs describe features. This one's long-form post opens its first section with "Results speak for themselves": view counts and sales screenshots before any tool talk. That's how it got out of the developer circle and into the side-hustle crowd, with 500k views far beyond what the tech community alone would have produced.

The author's X bio reads "Former big-tech growth lead | Million-GMV matrix player | Video sales steadily 2000+ per day | Xianyu/Xiaohongshu 3C category monthly profit 5w+ (paid teaching, full accompaniment)." In the long post he says the Skill "costs not a cent... and there's no high-priced paid group to fleece people," yet 10-29 he added a tip QR code and contact info to the repo. The repo itself doesn't earn money; what it brings is attention and trust, and the author's business lies outside it (inferred).

Did readers make it work?

Unknown. User reports are scattered: one says a video gained 200 followers, another says they got one working. A grounded comment in the long post's replies, from @YuanCongo6: "My method for history-figure accounts is almost identical to the blogger's... average views per video are only a few hundred on WeChat Channels and a few dozen on Douyin." So the tool can cut production time per video, but this niche is crowded, and the author himself writes: "It's not that I gave you a skill... that you can post randomly and it goes viral."

One more thing to watch: the corpus is verbatim transcripts from someone's WeChat Channels videos. The author changed the license to CC BY-NC-SA 4.0 (no commercial use) in Nov 3, but GitHub no longer detects a license type, and the earlier third-party check said MIT, so the records don't agree.

No one has produced a fork-level redesign of it. The similar projects all came earlier, so this isn't copying it.

If it were you

Bottom line: the pipeline can be copied. The stars came from "a ready-made corpus for one vertical" combined with "proof you made money in that vertical." Followers may not have the second, so it's worth trying, but first check whether you have results of your own.

Who it's for: people who have actually gotten numbers in a content niche (views, followers, or sales all count), who can build a pipeline with Claude Code or Codex, and who can publish long posts on Chinese X, Jike, or Xiaohongshu.

Who it's not for: people who can only build a pipeline with no niche results (you'd end up one of the 100 to 200-star clones); people who want a general "AI video generator" (MoneyPrinterTurbo and CapCut have that covered); and people who want another "history figure stories" project (the corpus and narrative are already taken).

How to do it: change the niche, keep the pipeline. Candidates: TCM wellness, personal finance education, legal awareness, English-language history accounts, or output formats like Xiaohongshu image-text posts or Douyin talking-head videos. Organize 5 top accounts' content in that niche into writing rules and a hook library (don't dump other people's full verbatim scripts; copyright is this project's biggest risk), add your own proven style or templates, and package it as a Skill you can install in one line.

How to get users: follow its sequence. Publish one long post; in section one, show your own results screenshot; section two covers the pain point; section three gives a copy-paste prompt; put the repo link at the top. Over the following week, repost user success reports or publish a follow-up every two or three days, and each one pulls stars back up a bit. Also find bloggers making "Claude Code video Skill" roundups and try to get included.

When to move: the "Agent Skill for content" wave started in October, and roundup posts are still coming out. Shipping before year-end is best. Once official tools like CapCut or Jimeng make "topic to finished video" a standard feature, the tool alone won't carry the story (inferred).

Signs you've misread it: ① a week after your long post you have fewer than 200 stars, meaning the hook isn't strong enough or nobody wants that niche; ② over the next month or two a batch of "[niche] Skill + make-money long post" repos appear, mostly stuck at a few hundred stars, meaning the formula is oversaturated; ③ story-studio gets taken down after a copyright complaint over its corpus, meaning "packaging other people's hits" doesn't work and you need your own corpus.