# How does a list get 1,900 stars in 24 days: awesome-short-drama-skills tests for you which skill to install · GitHub open-source project

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

> Short-drama skills are too numerous to sift through, and this is the first list dedicated to them. The blind AI benchmark with a control gave big accounts a ready-made conclusion to repost. The three star-growth peaks each line up with a wave of big-account posts.

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

UnderleveledBuilder/awesome-short-drama-skills · UnderleveledBuilder

# How does a list get 1,900 stars in 24 days: awesome-short-drama-skills tests for you which skill to install

**What it is** An open-source list of agent skills for AI short dramas and comic dramas. It collects 83 skills, organized by creation stage, and uses controlled comparison tests to tell you which ones to install for scriptwriting, topic selection, and season outlines.

**Why it took off** There are too many short-drama skills to sort through, and this is the first list dedicated to them. Its AI blind-test results against a control group give big accounts a ready-made conclusion to repost. Its three star-growth peaks line up one by one with three waves of big-account posts.

**B** **How you can follow it** Don't build another short-drama list. Pick a hot category where skills already proliferate but no list exists yet: test first, then list. Put a "which one to install" table on the first screen of the README, send it to Chinese open-source recommendation accounts, and let the authors who get named repost it for you.

Grade B: the playbook is clear and doesn't depend on the author's own traffic, so it can be copied. But this niche is already taken, and when the author tried to replicate the approach in another category on their own, they only got tens to about two hundred stars. That means you still need to bet on a category that's blowing up and on big accounts reposting. It's being passed around in small circles

Data (updated daily) · Repo created 2026-10-13 · 1,958 stars · 7 days +425

GitHub · stars

Data 2026-10-16 → 2026-11-08 (24 days, 22 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

GitHub has countless awesome lists, and most have only a few dozen stars. This one collects agent skills for AI short dramas and comic dramas. The author's account has only 42 followers, yet the repo gained about 1,900 stars within 24 days of being created (GitHub, 11-06). It does one thing other lists don't: it runs every skill through a controlled comparison against a control group. This piece covers what it is, which posts brought the stars, why this one in particular worked, and what you can take from it if you want to build an open-source project that attracts stars.

### "33 scripts out of 15 had a character named Xiaoman pop up"

At 1:30 PM Beijing time on Oct 20, 2026, "Programmer Yupi", a major Chinese-language X account that recommends open-source projects, posted:

> There are so many screenwriting Skills for AI that it's dizzying. Which one should you install? This open-source list collects 83 short-drama and comic-drama Skills. The author picked 10 screenwriting ones and ran them against Claude Code without any Skill on the same 60-episode vertical short-drama task, 3 runs each, then had AI judge them blind. ... Surprisingly, in 33 of the scripts, 15 had a character named "Xiaoman" appear, and almost no Skill could avoid it.

This post got roughly 8,300 views and 88 likes, not a huge number, but the repo gained 309 stars that day (per Trendshift daily tallies), its biggest single-day gain so far. One reply underneath said: "This comparison method is worth learning from. Running the same task three times and then blind-judging beats the author bragging about it being reliable."

The repo was only a week old at that point. It was created on Oct 13, and got almost no views in its first five days. On Oct 17 the author published a post on Juejin titled "33 open-source skills adapted for Seedance", and on Oct 18 the first round of test results went out, and the stars started moving.

### What it is

A single Markdown list, plus a bilingual website at shortdramaskills.com.

- It collects 83 short-drama skills and 16 related open-source platforms (README; metadata updated 11-02). You can browse by genre, creation stage (novel adaptation, episode outlines, hooks and paywall moments, script review, storyboarding, video prompts), or the video model they fit (Seedance, Kling, MiniMax);

- It explains how to install each skill in agents such as Qwen, Doubao, OpenClaw, WorkBuddy, Claude Code, and Codex;

- When you open the README, it isn't a table of contents. It's three "which one to install" tables: **which to install for scriptwriting, which for topic selection, and which for full-season outlines** . Each table lists the few skills that beat the control group, with one column for "best for" and one for "watch out for". For example: "short lines, each character has a distinct voice, emotion carried by objects / understated episode-ending hooks";

- The list is itself a skill. Install it with `npx skills add UnderleveledBuilder/awesome-short-drama-skills` and let the agent pick for you;

- The README is generated by a script from a data file, star counts refresh automatically every week, and "candidate project" issues are opened automatically.

Three rounds of tests went live on 10-18, 10-20, and 10-31. Each round included a control group with no skill installed, the same task was run 3 times, and two AI judges, Opus and GPT, compared the results blind. The limitations are stated candidly: "on episode-ending hooks, Opus thinks the control group is often stronger, while GPT sees it the other way", "only one model was tested", and "the two judges disagreed on 63 of 198 conclusions".

### Which posts the stars came from

| Timeline | Stars gained that day | What happened |
|---|---|---|
| 10-13 to 10-17 | About 0 | Repo created; bilingual site launched and submitted to Baidu and Toutiao webmaster platforms; the author published an article on Juejin |
| 10-18, 10-19 | +26, +136 | First round of test results published; "which to install for scriptwriting" added to the top of the README |
| 10-20 | **+309** | Programmer Yuji posted; the second round of tests went live the same day |
| 10-21 to 10-23 | +104, +45, +22 | Aftermath |
| 10-24 to 10-26 | +91, +161, +117 | GitHubDaily published a long post, wlzh followed up ("live only 11 days and already 661 stars"); featured authors reposted; it reached the Trendshift HTML ranking, best position #7 |
| 10-29, 10-30 | +118, +208 | The Weibo AI blogger "Aikeke-Aishenghuo" posted; Trendshift HTML ranking #5 |
| 11-02 to 11-06 | About 45 to 90 per day | Third round of tests went live; Taiwanese Facebook blogger Ho Chia-hsun posted image-text and tutorial videos; Sina, CSDN, and Tencent Cloud community reposted it |

(Daily star gains are from Trendshift, UTC dates.)

The three peaks line up with three waves of big-account posts, down to the hour. Over the last 7 days it gained 590, and the first 7 days contributed 624, and it hasn't dropped off: every few days a new account picks it up. The fork-to-star ratio is about 12%, similar to comparable new repos, so there's no sign of star farming.

The big accounts reposted the same line. GitHubDaily's original text:

> Searching GitHub for short-drama Skills turns up dozens, and it's hard to tell which ones can actually write good scripts. ... What's rare is that the author ran two rounds of tests themselves, 10 script Skills against a control group with no Skill installed ... The limitations are stated honestly too, for example on episode-ending hooks, the control group often wrote better ones. I value this kind of control-group evaluation more than a list that just names tools.

There's also a thread through the people who got named. In the test tables, the recommended jtydhr88 reposted on X: "My screenwriting skills were placed among the four top picks." An old rule of awesome lists: authors who get included and praised will spread it for you.

There is no evidence either way on whether the author actively reached out to these big accounts or paid them.

### Why this one

**The space is hot, but nobody is helping you pick.** 2026 6 through October, short-drama skills started piling up on GitHub, and a single repo hit thousands of stars: eternityspring/shuohao-skills at about 6,000, zenstory-ai/drama-skills at about 3,800 (GitHub, 11-06). Anyone wanting to make AI short dramas now faces dozens of options, and GitHubDaily's line "hard to tell which one can actually write a good script" is a user's own complaint. As of now, GitHub still has no second list dedicated to short-drama skills.

**It gives the people sharing it a conclusion, not just a list of names.** An ordinary awesome list can only say "collects N"; this one can say "5 beat the control group in all three runs," and it comes with fun details like "Xiaoman." The copy from Yuji, GitHubDaily, and Aikeke centers on this testing.

**The author treats distribution as a craft.** The bilingual site is connected to Baidu, Toutiao webmaster tools, and IndexNow, and it ships an llms.txt for AI crawlers. The README explicitly adds the names of domestic agents like "Qianwen, Doubao, Longxia, WorkBuddy," and the commit is literally titled "add domestic agent keywords." The author also maintains two other lists about SEO / GEO.

**The users have changed.** Skills are spreading from Claude Code to domestic agents such as Qianwen Office and Doubao desktop, and short-drama practitioners who don't write code have started installing skills. They need someone to tell them directly which one to install even more than programmers do.

### Who it's competing with

| Project | Scale (GitHub, 11-06) | How it differs |
|---|---|---|
| bestaidrama/awesome-ai-drama-generators | 74 stars | Collections of short-drama generation tools, not skills, with no testing |
| zhaotoday/awesome-ai-video | 46 stars | General AI video indexes where short drama is just one section |
| The awesome-seedance series (ZeroLu, LearnPrompt, YouMind) | Each 2.2k to 2.8k | Only collect Seedance prompts; no scripts, adaptation, or storyboarding |
| General skill directories (skillsllm, skills-hub.ai, sickn33/agentic-awesome-skills, etc.) | Collect thousands | Keyword listings without stage classification or effectiveness testing |
| Short-drama platforms like Toonflow, huobao-drama, waoowaoo | 15k to 20k | Finished apps, not skills you install into an agent |

The short-drama skill list niche currently has only this one. Users can, of course, bypass the list and install the few skills with thousands of stars directly.

### Can the playbook be copied: the author tried it themselves

The best counterexample is the author. Same playbook (vertical list, generator, bilingual site, keywords), moved to other niches: awesome-social-media-skills about 200 stars, awesome-novel-skills and awesome-seo-geo-skills each about 50 stars, awesome-story-skills 8 stars. All far behind the short-drama one.

Method alone isn't enough. Two things have to line up: the niche is hot and there are so many skills that users can't pick among them, and there's a conclusion that big accounts will repost. The short-drama list hits all three; the others miss the last two.

Barely spread in English. Only one English AI news post cited its test results, plus a Portuguese news piece (inference: the English side is still open, and the author already has an English README and site, so they can fill it in anytime).

### If it were you

**B. The star playbook is clear and copyable, but the short-drama slot is already taken, and the author's own repeats in other niches only got a few dozen to a couple hundred stars. What's worth doing is a different niche that's booming, and it must include real test results.**

**Who it's for** : developers following a particular agent skill niche who are willing to spend a few days actually running a dozen-plus skills; people who write Chinese and know who the recommenders in the Chinese AI open-source scene are.

**Who it's not for** : anyone wanting to build another short-drama skill list (the top one already has 1,900 stars and is still publishing weekly test updates); anyone who only wants to list and not test (the author's untested lists in other niches are the outcome); niches with only two or three dozen skills (users don't have a 'can't choose' pain, so a list adds no value).

**How to do it:**

1. Find a vertical niche where skills are numerous enough that people can't choose. Signals: dozens of similar skills on GitHub, a few single repos with over a thousand stars, and no dedicated list yet. Directions to watch: novel and web-fiction writing, Xiaohongshu and Douyin operations, e-commerce product pages and sales scripts, legal documents, thesis writing, PPT and poster design.

2. Test before listing. Pick about 10, set up a control group without the skill, run the same question 3 times, have two different AI judges compare blind, publish the whole process, and state the limitations up front. The first screen of the README should have an 'X: which one to install' table with a fit column and a caution column.

3. During testing, note one counterintuitive detail (for this one it was 'all 15 drafts came out with a Xiaoman'). That's the line people love to quote when they repost.

4. Make the list itself a skill, with the README generated by a script and the data refreshed automatically every week.

**How to get users** : your own account probably has no audience (this author has only 42 followers), so stars have to come from others reposting. On launch day, write a long post on Juejin and Zhihu; condense the test conclusions into a paragraph people can copy directly and send it to Chinese X open-source recommenders and GitHub-digest accounts; open issues or @ the authors of the skills you name, and send them a 'listed' badge so they repost it for you. Run a new round of tests every few days; each round is a reason for big accounts to repost again.

**Deadline** : the two or three months when the niche is hottest and skills are multiplying fastest. The short-drama skills appeared from 6 through October, and the list came in October, just in time; once someone in the niche has published a tested list, you're stuck as the second.

**How we could be wrong** : ① one of the author's other niche lists also topped a thousand stars after adding tests, which suggests the playbook is what matters and the niche is less critical, so the bar is lower than written here; ② someone starts a rival short-drama skill list with harder testing (multiple models, human judges, ten-episode serialized writing) and passes it quickly, which means the slot isn't locked and B should go up; ③ Chinese open-source recommenders collectively stop reposting list-type projects, which cuts off the free distribution path.

**Data and sources** : star and fork counts come from GitHub (11-06) and daily snapshots; daily star gains come from Trendshift (trendshift.io/repositories/310209); Yupi's post and replies come from X (10-20), the GitHubDaily original from X (10-24); test tables and limitations come from the repo README (read 11-06); competitor star counts come from GitHub, 11-06. Wherever we write 'inference', it is an inference.

[GitHub repo ↗](https://github.com/UnderleveledBuilder/awesome-short-drama-skills)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| bestaidrama/awesome-ai-drama-generators | Open source | 2026-09-26 | 74 stars | Collects AI short-drama generation tools, not agent skills, and has no tests |
| zhaotoday/awesome-ai-video | Open source | 2026-09-15 | 46 stars | A general AI video index; short drama is just one section; no tests |
| The awesome-seedance series (ZeroLu / LearnPrompt / YouMind) | Open source | 2026-02 to 2026-08 | About 2.2k to 2.8k stars each | Only collects Seedance prompts; doesn't cover the full pipeline of script, adaptation, and storyboarding |
| General skill directories (skillsllm, skills-hub.ai, skillsdirectory, typingmind, sickn33/agentic-awesome-skills) | Web | 2026 | Collect thousands of skills | List by keyword without splitting by short-drama creation stage, and don't test effectiveness |
| eternityspring/shuohao-skills | Open source | 2026-08-06 | 6,075 stars | Is itself a set of short-drama skills (listed and recommended by the list), not a selection index |
| zenstory-ai/drama-skills | Open source | 2026-07-16 | 3,825 stars | A full-pipeline skill from a comic-drama studio, covering only its own set |
| Short-drama platforms such as Toonflow / huobao-drama / waoowaoo | Open source | 2026-01 | 15k to 20k stars | Are full applications, not skills you install into an agent |
| Script-to-storyboard agents built into platforms such as Jimeng and Seko | Big-company features | 2026 | Not verified | Serve only their own platform and don't compare third-party skills side by side |

Where users come from

- **Verified** **X big account: Yupi (程序员鱼皮)** : posted 10-20 05:31 UTC, 8.3K views, 88 likes; that day Trendshift +309, the full-cycle peak

- **Verified** **X big accounts: GitHubDaily, wlzh** : posted 10-24, highlighting 'tests with a control group'; 10-25 +161, reached #7 on the Trendshift HTML chart

- **Verified** **Weibo: 爱可可-爱生活 (Ai Keke)** : posted 10-30 20:49 Beijing time; that day +208, #5 on the Trendshift HTML chart

- **Verified** **Reposted by a listed author** : jtydhr88 reposted on X 10-24: 'My screenwriting skills were put in one of the four top picks'

- **Verified** **Author's own SEO / content** : 10-17 long post on Juejin (self-described maintainer); bilingual site connected to Baidu / Toutiao webmaster tools, IndexNow, llms.txt; README deliberately stacks Chinese agent keywords

- **Verified** **Reposts from Chinese aggregators and content farms** : Zhihu, duan.cn, ai211, Sumike, Sina, CSDN, Tencent Cloud Community, a Telegram channel, Taiwanese Facebook blogger Ho Chia-Hsun (何佳勳)

- Inference **Whether the author actively sought out KOLs or paid for promotion** : no evidence found

## Why UnderleveledBuilder/awesome-short-drama-skills took off

> Channel: GitHub · created 2026-10-13 · 2026-11-06 GitHub API shows 1,895 stars / 229 forks · license CC0 · companion site shortdramaskills.com

### Bottom line

This is an awesome list of AI short-drama / comic-drama agent skills. It gained traction through three things: it picked a niche that was already hot (from June, short-drama skills and platform repos sprouted on GitHub, several thousand stars for a single repo, but no proper list existed); it added a layer beyond a typical awesome list, namely blind AI comparison tests with control groups, giving KOLs a ready reason to repost; and the author treats SEO / GEO / Chinese self-media distribution as one system. The direct cause of the stars was a relay of reposts by big accounts in the Chinese AI scene: Yupi (X) → GitHubDaily / wlzh (X) → 爱可可-爱生活 (Weibo), plus reposts from a batch of Chinese aggregator sites.

### 1. What it is

- The README lists 83 short-drama skill projects and 16 related open-source platforms (metadata updated 2026-11-02). It indexes them three ways: by type / creation stage / compatible video model (Seedance, Kling, MiniMax H3), and explains how to install each skill in agents such as Qwen, Doubao, OpenClaw, WorkBuddy, Claude Code, and Codex.

- It has three public rounds of tests: 10-18 scripts, 10-20 topic selection, 10-31 full-series skeletons. Each round sets up a control group without the skill, runs the same question 3 times, has Opus and GPT judge blind comparisons, and lists limitations alongside. The full process is at shortdramaskills.com/evals/.

- The list itself is also packaged as a skill (short-drama-skill-finder). After installing with `npx skills add UnderleveledBuilder/awesome-short-drama-skills`, you can have an agent pick skills for you.

- The README is generated from data/ by scripts/build.py. GitHub Actions refreshes star counts and metadata weekly and automatically opens 'candidate project' issues.

(Source: repo README and commit history, read 2026-11-06)

### 2. Growth curve and the source of each inflection point

| Time (UTC) | Star gains (Trendshift, daily) | what happened |
|---|---|---|
| 10-13 to 10-17 | About 0 | Repo created; 10-15 transferred to the UnderleveledBuilder account; 10-16 bilingual Chinese/English site launched with IndexNow, Baidu / Toutiao webmaster verification, llms.txt; 10-17 author published '33 open-source skills adapted for Seedance' on Juejin, self-described as the list maintainer |
| 10-18 to 10-19 | +26, +136 | 10-18, 10-19 first round of tests published; 'which one to install for writing scripts' added at the top of the README |
| 10-20 | +309 (highest of the full cycle) | **Yupi (程序员鱼皮) @yupi996** posted 10-20 05:31 UTC, main post 8.3K views, 88 likes; a second round of tests launched the same day |
| 10-21 to 10-23 | +104, +45, +22 | Aftershocks; began to fall back |
| 10-24 to 10-26 | +91, +161, +117 | **GitHubDaily** posted a long post 10-24 04:00 UTC, stressing 'I respect tests with a control group'; **wlzh** posted 10-24 07:22 ('live only 11 days and already 661 stars'); a listed author jtydhr88 also reposted (10-24); Zhihu, duan.cn, ai211, Sumike, and a Telegram channel followed with reposts; from 10-25 it appeared on the Trendshift HTML chart, best #7 |
| 10-29 to 10-30 | +118, +208 | **Ai Keke (爱可可-爱生活)** , a Weibo AI blogger, posted 10-30; a mushroom.cv blog published an article 10-29; Trendshift HTML chart #5 |
| 11-02 to 11-06 | About 45 to 90 per day | 11-03 third round of tests launched; Taiwanese Facebook blogger Ho Chia-Hsun (skyman301) posted image-text and tutorial videos; Sina, Tencent Cloud Community, and CSDN reposted |

- Our database (github.repo_day) records Beijing-time snapshots: 10-19 is 36 stars, 10-20 is 112,10-21, 461,10-26 is 889,10-31, 1,462,11-06 is 1,845. The single-day peak of +349 was recorded on 10-21, corresponding to Yupi's wave at UTC 10-20. Over the last 7 days it gained +590, and over the prior 7 days +624; it hasn't dropped off, because a new big account picks it up every few days.

- Fork/star ratio is about 12%, close to the median for similarly aged new repos (about 11%), with no sign of star-farming. The stargazer breakdown API returned 404 this time, so it couldn't be verified.

Sources: Trendshift [trendshift.io/repositories/310209](https://trendshift.io/repositories/310209); Yupi [x.com/yupi996/status/2112416270142652848](https://x.com/yupi996/status/2112416270142652848); GitHubDaily [x.com/GitHub_Daily/status/2113842948945412240](https://x.com/GitHub_Daily/status/2113842948945412240); wlzh [x.com/wlzh/status/2113893886309585068](https://x.com/wlzh/status/2113893886309585068); jtydhr88 [x.com/jtydhr88/status/2113955676506522051](https://x.com/jtydhr88/status/2113955676506522051); Ai Keke [www.sina.cn/news/detail/5359821781997313.html](https://www.sina.cn/news/detail/5359821781997313.html); Juejin [juejin.cn/post/7697089423961243684](https://juejin.cn/post/7697089423961243684); Facebook [www.facebook.com/skyman301/](https://www.facebook.com/skyman301/).

### 3. Competitors: who else does the same thing

**Other 'short-drama skill list / index' projects** : no second dedicated awesome list for short-drama skills turns up on GitHub; this is the first and only one. Nearby lists include:

- bestaidrama/awesome-ai-drama-generators, created 2026-09-26, 74 stars. Collects AI short-drama generation tools, not agent skills, and has no tests.

- zhaotoday/awesome-ai-video, created 2026-09-15, 46 stars. A master index of AI video production resources; short drama is just one section.

- JuneYaooo/awesome-ai-media-cn, created 2026-04, 359 stars. A collection of AI video and self-media tools.

- The awesome-seedance series: ZeroLu 2.8k, LearnPrompt 2.45k, YouMind 2.2k. These are Seedance prompt libraries that only cover the "video prompt" stage, not scripts, adaptation, or storyboards.

- General skill directories: skillsllm, skills-hub.ai, skillsdirectory, typingmind skills, mcpmarket. The sites are large but only list by keyword, without sorting by creation stage and without testing effectiveness. sickn33/agentic-awesome-skills collects 2,900 skills, likewise untested.

**The skills it lists** (users can also bypass the list and search for these directly):

- eternityspring/shuohao-skills, 6,075 stars, created 2026-08.

- zenstory-ai/drama-skills, 3,825 stars, created 2026-07.

- lixiaoxiao9888-create/manju-laoli-skill, 1,927 stars.

- 0xsline/short-drama, 1,563 stars.

- liangdabiao/Seedance2-Storyboard-Generator, 2,773 stars.

**Complete platforms** (Toonflow 20k stars, huobao-drama 17k stars, waoowaoo 15k stars, Alibaba lumenx, Tencent workrally): these are finished applications, not skills, and the list files them under a "related open-source platforms" section.

Big-company offerings: products like Jimeng and Seko already ship agents that go from script to storyboard, but they only serve their own platforms and do not benchmark third-party skills against each other.

### 4. Growth tactics broken down

1. **KOL relay (confirmed)** : the three star-growth peaks line up hour-for-hour with three waves of posts from Yuwu (10-20), GitHubDaily and wlzh (10-24), and Aikeke (10-30), all top "open-source recommendation" accounts in the Chinese AI/open-source community. There is no evidence yet on whether the author pitched or contacted them. However, all three posts push the same point: "benchmarked testing" and "three straight wins against the control group." This shows the testing itself was built as the thing to spread.

2. **Author's own SEO and channel content (confirmed)** : 10-17 published a long post on Juejin and self-identified as a maintainer. The site has Baidu and Toutiao webmaster verification, IndexNow, and llms.txt, and it logs "AI referral" traffic. The author also has two other libraries, awesome-seo-geo-skills and awesome-geo-cn, both about GEO. The README deliberately includes keywords for Chinese agents like Qwen, Doubao, Lobster, and WorkBuddy; the commit 10-17 "add Chinese agent keywords" does this.

3. **Flattering the listed authors so they amplify it (partially confirmed)** : the benchmark names specific recommended skills, and there is a "listed badge" section. The recommended jtydhr88 actively reposted on X ("my screenwriting skill was placed among the four top picks"). This is the old awesome-list playbook: listed authors spread the list for you.

4. **Automatic reposting by Chinese aggregators and content farms (confirmed, passive)** : Zhihu, duan.cn, ai211, Sumiko, baoshare, Sina, CSDN (gitblog), Tencent Cloud Community, TG channels, and the Portuguese-language novidadesia. Most of these followed after the KOL posts.

5. **Keeps shipping new content so distribution never stalls (confirmed)** : three rounds of benchmarks went live at 10-18, 10-20, and 10-31 to 11-03, each giving people a new reason to share. A bot also refreshes the data weekly, and the README has a "new this week" section.

6. **Channels not found** : Hacker News, Reddit, Product Hunt, ads. These terms returned zero hits in the local x-radar corpus (coverage is limited).

### 5. Why it took off

- **Hot track** : 2026 saw a wave of short-drama / comic-drama repos on GitHub, then 6 to October brought a batch of "short-drama skills," single repos at 1K to 6K stars, and Hongguo, Seedance 2.x and MiniMax H3 are all pushing the same thing. The real user pain is in GitHubDaily's own words: "dozens of results, hard to tell which one can actually write a good script." The list speaks directly to that need.

- **The differentiator is testing, not just listing** : a typical awesome list only lists names. This one has a control group, two judges, 3 runs, full public process, and an honest section on limitations. Nearly every KOL repost quotes this part.

- **The author has a distribution playbook** : a bilingual site, webmaster platforms, self-published Juejin posts, keyword stuffing for Chinese agents, and turning the list itself into a skill. An account with 42 followers reached 1.9k stars in 24 days.

- **Timing** : the skill ecosystem is spreading from Claude Code to Chinese agents (Qwen Office, Doubao desktop, WorkBuddy, OpenClaw), and short-drama practitioners with no coding background are starting to install skills. They need "just tell me which one to install" more than programmers do.

### 6. Differentiation, and whether that's why it broke out

Its differentiation comes down to three things:

1. Narrows down to short drama and comic drama, and indexes by creation stage and video model;

2. Runs controlled, blind AI-judged benchmarks and draws conclusions about which skills work;

3. The list itself can be installed into an agent as a skill.

Point 2 is why it took off. Every big account's repost copy uses the benchmark as the selling point. But the benchmark used only Opus 5.5, and the judges are also AI, so the conclusions are only so firm. The author states this in the limitations section.

### 7. Visibility and copycats

- Visibility: it has already been written about by top accounts in the Chinese AI open-source community (Yuwu, GitHubDaily, Aikeke), reached Trendshift, and was reposted by dozens of aggregators, so it is not obscure. It has barely spread in the English-speaking world; only varg.ai cited its benchmark conclusions, plus one Portuguese news article. No revenue is involved; this is a CC0 open-source list.

- Copycats: no one has yet been spotted launching a separate short-drama skill list. The most obvious copies come from the author, who applied the same playbook to other categories: awesome-social-media-skills (about 200 stars), awesome-novel-skills (about 50 stars), awesome-seo-geo-skills (about 50 stars), awesome-story-skills (8 stars), and awesome-geo-cn. All of these are far behind the short-drama list, which suggests the playbook alone is not enough: without KOL reposts and a hot enough category, it does not take off.

### 8. What others can learn

- Enter a hot vertical where "there are already too many skills to sift through," and build a list with controlled benchmarks. Compared with lists that only enumerate, it gives a shareable conclusion, and it gives the named authors a reason to repost.

- Use a README generator with Actions for weekly data refreshes, plus a bilingual static site, webmaster platforms, and llms.txt, to turn the list into an SEO/GEO asset.

- Package the list itself as a skill, so agents can discover it and install it.

- Publish a new round of benchmarks every few days to give KOLs fresh reasons to share, so growth never stalls.

### 9. Still unclear

- No evidence found on whether the author proactively contacted Yuwu, GitHubDaily, or Aikeke (paid promotion or organic discovery).

- The stargazer detail API returned 404, so account quality could not be verified, though the fork ratio looks normal.

- Actual usage is unknown: skill-finder install counts and site traffic are not public.

- The author's identity is unclear (X @UnderleveledDev, account registered 2016 years ago). Judging by the SEO/GEO playbook and running several lists at once, this looks like an operator systematically building content assets, and quite possibly an agent is running it too.

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