# The ninth official-document skill in a crowded niche: they pitted 8 rivals head to head on benchmarks · 22 days, 300 stars · GitHub open source project

> Source: https://stringers.io/en/github/gh_duyunzhoudyz__dyz-gongwen-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

> Year-end is exactly the season for summaries and reports. Among many single-purpose peers it is the most complete, and its README leads with head-to-head benchmark results against 8 competitors. The author also kept submitting it to skill aggregators and awesome lists, with no big account promoting it, and it climbed steadily.

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

duyunzhoudyz/dyz-gongwen-skills · duyunzhoudyz

# The ninth official-document skill in a crowded niche: they pitted 8 rivals head to head on benchmarks · 22 days, 300 stars

**What it is** A Chinese official-document writing toolkit you install into Claude Code, Codex, or Cursor. It covers picking the document type, drafting, removing AI-sounding phrasing, proofreading, and formatting a red-header Word file to national standards. Industry packs for hospitals, universities, and state-owned enterprises are included. Its selling points: no made-up data, and it respects confidentiality.

**Why it took off** Year-end is peak season for summary reports and work documents. In a field full of single-purpose competitors, it's the most complete, and its README leads with benchmark results against 8 rivals. The author also submitted it to skill aggregators and awesome lists one by one. No big account pushed it, but it climbed steadily anyway.

**B** **How you could follow this** Don't build another official-document skill. Pick a Chinese writing scenario with national standards and checkable outputs (bid documents, thesis formatting rules, patent filings, legal documents). Build it as a toolkit plus industry packs, put a rival benchmark at the top of the README, submit to skill directories and awesome lists the week you launch, then find a Chinese recommendation account to relay it once.

Grade B: the approach is clear and cheap. The author used the same 5-day playbook to build another repo that hit 200-plus stars, which shows it's repeatable. But official documents are already saturated, the vertical is narrower, and there has been no big-account reshare, so the ceiling is probably a few hundred stars. Chatter in small circles

Data (updated daily) · Repo created 2026-10-15 · 381 stars · 7 days +177

GitHub · stars

Data 2026-10-27 → 2026-11-08 (13 days, 13 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

An AI-drafted district education bureau notice contained the line "Parents will be handed over to the public security organs for detention without exception." A district education bureau has no punitive power, so sending that out is an incident. This is from the README of dyz-gongwen-skills, which puts it up front: an AI official-document writing kit for people who write materials at government agencies and state-owned enterprises. It was created in Oct 2026 and had about 15 stars by Nov 6 300. At least a dozen official-document skills already exist on GitHub. It didn't surface because it arrived early. It surfaced because it benchmarked itself against rivals, and then the author spread it across every skill directory.

### What it is

A set of "skills" (think of them as specialized operating manuals for AI) that you install into AI coding tools like Claude Code, Codex, or Cursor, dedicated to Chinese official documents. 17 core skills cover the full workflow: first deciding whether a notice, request, or letter is the right type, then drafting (formal documents, year-end summaries, leadership speeches, party-building materials, meeting minutes, rectification logs), then revising, removing AI tone, proofreading, and finally formatting a red-header Word file to the GB/T 9704 national standard. There are also 5 industry packs for hospitals, universities, primary and secondary schools, state-owned enterprises, and townships. The hospital pack, for example, writes DRG and CMI metrics into grade-evaluation self-assessments and department reports.

The README's first screen sells three things: no fabrication, confidentiality, one-click formatting. Missing data is not invented; it's marked "【待补】" (to be filled in). Confidential materials get a reminder not to paste them into online AI, and an offline prompt pack is provided for in-house models. People who don't use the command line can copy the prompts straight into the web versions of DeepSeek, Doubao, or Kimi.

The author's account was registered in March 2026, has 0 followers, and had no other projects before this. The commit history shows 15 skills generated in one day in Oct 16, then more than seventy commits over two days in October 20 to 22. This looks like batch generation with an AI agent (inference).

### How the stars grew

| Date | star | Gained that day |
|---|---|---|
| 10-27 | 58 |  |
| 10-28 | 111 | +53 |
| 10-31 | 172 | +44 |
| 11-01 | 204 | +32 |
| 11-02 ~ 11-04 | 209 → 216 | Barely moving |
| 11-05 | 239 | +23 |
| 11-06 | 284 (around 300 later that day) | +45 |

(GitHub daily star records)

It never appeared on GitHub Trending, and no big account wrote about it. Searches on X, Zhihu, Xiaohongshu, and V2EX turn up no posts dedicated to it. I couldn't trace the source of the +53 jump in Oct 28. It may have come from places that aren't searchable, like WeChat groups or Official Accounts (inference). The sources I could find fall into three groups:

**Skill directories and aggregators picked it up automatically.** hotskills.dev's "24-hour growth chart" listed it around Oct 30 (at 123 stars then). ProSkillsMD opened a listing candidate for each of its dozen-plus sub-skills. ClaudePluginHub, SkillsLLM, and the Diedong AI navigation site (11-05) all have listing pages. A Korean Facebook daily, "PIELAB 스킬 데일리", ranked it #3 in its office-themed issue of Nov 3.

**The author submitted it personally.** On Nov 4 they applied for inclusion in awesome-claude-code, and on Nov 6 they applied to awesome-claude-skills. In the same two days they forked 6 more awesome lists, apparently preparing more PRs. The timing lines up with the +23 and +45 star jumps on 11-05 and 11-06.

**The packaging was overhauled the same day.** On Nov 4 the repo was renamed (it was gongwen-skills, now prefixed with the author's initials, dyz-), a GitHub Pages landing page went up, along with a social preview image and a before-and-after comparison image for the top of the README. 12 topic tags were added (claude-skills, codex, cursor, deepseek, gongwen, and so on), and both the opening and closing of the README say "Give it a Star."

### Why click this one out of all the similar repos

The official-document skill niche is crowded, and the leaders are an order of magnitude bigger:

| Project | star(11-06) | What it does |
|---|---|---|
| lieflat-less-ai-tone | 5,241 | General de-AI-ification, not just official documents |
| lieflat-gongwen (official-document writing DNA) | 2,378 | A writing style distilled from a 1.02M-character corpus; GitHubDaily promoted it on X |
| gongwen-gbt9704-skill | 848 | Only handles national-standard formatting; a Zhihu article covered it |
| official-document-skill | 431 | A single writing skill distilled from People's Daily corpus |
| KaguraNanaga's writing and formatting libraries | 309 / 238 | Earliest to start (January), no longer growing much |
| dyz-gongwen-skills | About 300 | Writing, revision, formatting, and checking, plus industry packs |

There are also a dozen or so single-purpose tools with a few dozen stars each, paid web and Office plugins like Gongwen Bao and Secretary Materials Star, and the built-in official-document features in WPS AI and Doubao.

From Oct 28 to Nov 6 it gained 173 stars, ranking third in the official-document category, behind only the two lieflat repos (+731 and +851). The others in the category all stayed within +24 per day (GitHub daily star records). It won on two things:

- **It's the most complete.** Rivals either only write, only format, or only remove AI tone. This one covers everything and splits it into industry packs. Among a set of similar-sounding repos, "complete" is the easiest reason to get picked.

- **It benchmarked rivals and put the results up front.** It ran 8 comparable official-document skills from GitHub, plus a "no skill installed" control group, on the same model with the same test questions, 276 times, with two large models acting as judges. The results sit at the top of the README: on national-standard formatting audits it scored 97.2 points, with the runner-up at 79.6. On confidential-material test questions with classification markings, its checkpoint pass rate was 91.7%, versus 0% for all rivals. It also honestly notes caveats like "the judges are LLMs, not human reviewers," which makes the results more believable.

It's also riding the season: every year 11, December is peak season for year-end reviews, performance write-ups, and next year's goal plans, and the lieflat category rises in the same window, so the whole niche is heating up.

But the ceiling is also clear: lieflat-gongwen did no benchmarking and got over two thousand stars from one GitHubDaily post. In this niche, whether someone vouches for you matters more for star count than how good your work is. Benchmarks and "most complete" decide whether people who arrive click through. They don't decide whether people arrive.

### Can this playbook be reused?

Yes, and the author already did it once. On Nov 1 they used the same template to build dyz-med-paper-skills, a set of 32 skills for doctors writing medical papers, which reached about 240 stars in 5 days. Same person, same packaging, same promotion, and the two repos together reached nearly 500 stars. That suggests the playbook worked, not a single lucky break.

The README also includes an enterprise WeChat discussion group and a payment QR code. The license was switched in Oct 22 from MIT to non-commercial only, and commercial use requires separate authorization from the author. The author's profile links to a service that helps universities, research institutes, and companies integrate large-model APIs. With one repo aimed at government and state enterprises and the other at doctors, the author may be pulling in institutional clients (inference, not verified).

What I still can't see: the repo has only 16 forks and almost no issues, so it's unclear how many people actually use it.

### If it were you

**Verdict** : Building another official-document skill is pointless. That spot is full. What's worth copying is the playbook: pick a Chinese vertical writing scenario that has national standards and checkable outputs, use AI to batch-generate a "toolkit plus industry packs," put a rival benchmark at the top of the README, and submit to every skill directory yourself. That reliably gets a few hundred stars. Reaching a thousand requires someone else to amplify it.

**Good fit** : Developers who can use AI agents to batch-produce work and who understand the ins and outs of a particular industry's documents (or are willing to spend a few days mastering the standards and model documents). Also people who want to build a reputation and portfolio through a series of projects.

**Poor fit** : Anyone who expects a single project to reach several thousand stars. Vertical skills like this have a narrow audience ceiling and no big-account reshares, so you'll likely land around three to five hundred.

**How to do it:**

1. **Picking a scenario** : Chinese, must-have, has standards, verifiable. Official documents and medical papers are already taken. Worth testing: bid and tender documents, thesis formatting (each university's rules), patent application documents, legal documents (complaints, defense statements), financial report notes, teaching plans and lesson-explanation scripts, construction supervision logs (inference, I haven't checked each one against rivals). First search GitHub for similar repos, count how many there are, and see how many stars they have.

2. **Build a toolkit** : Cover the whole chain, from "deciding which document type to use" to "a Word file in standard format," then split it into a few packs by agency or industry. Being the most complete is the reason you get chosen in a crowded niche.

3. **Benchmark** : Pull in every similar repo you can find, run them on the same model with the same questions, put the results in a table at the top of the README, and state the limitations clearly. Ideally include one dimension where everyone else scores 0 (their confidentiality check is that dimension).

4. **Reach people who don't use the command line** : Your target users are probably not programmers, so you need a web prompt version they can copy and paste.

**How to get users** : The week you launch, submit to skill directories like hotskills.dev, ProSkillsMD, and ClaudePluginHub, and open a PR on every awesome-claude-skills / awesome-agent-skills list. Rename it with a personal prefix to build a series. Put a before-and-after image and "Give it a Star" at the top. On top of all that, find a Chinese GitHub recommendation account that will relay it once (something like GitHubDaily), or industry Official Accounts and WeChat groups. That step decides whether you land at three hundred or three thousand.

**Before when** : skill directories and awesome lists are still in their hot inclusion phase, so new projects get listed easily. At the current pace of new entries, any scenario that gets three or five copies will be crowded out fast. The chosen scenario should launch before its peak season: official documents peak at year-end, theses in graduation season, bid documents have no clear seasonality. Roughly before the first half of 2027 is the best window (inferred).

**What would show you misread it** : if you build a new scenario with this playbook, submit it to directories and lists, and it still hasn't reached 150 stars after a month, that means this repo mainly rode the broad demand for official documents plus the year-end season, and the playbook alone can't carry a small scenario.

[GitHub repo ↗](https://github.com/duyunzhoudyz/dyz-gongwen-skills)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| larashero3-dotcom/lieflat-gongwen (Official Document Writing DNA) | Open source | 2026-09-02 | 2378 stars | Only does quantified writing style; was promoted by GitHubDaily. This project adds layout, classified-information handling, proofreading and industry packs |
| larashero3-dotcom/lieflat-less-ai-tone | Open source | 2026-08-20 | 5241 stars | General de-AI-tone tool; this project's humanize targets official documents only and keeps the legitimate conventions of official writing |
| mizzlelover/gongwen-gbt9704-skill | Open source | 2026-09-09 | 848 stars | Only does national-standard layout; this project treats layout as just one step in the full workflow |
| Liuxiangjian-ai/official-document-skill | Open source | 2026-06-07 | 431 stars | A single writing skill (trained on People's Daily corpus) |
| KaguraNanaga/official-document-writing-skill / document-format-skills | Open source | 2026-01 | 309 / 238 stars | Writing and layout split into two repos; started earliest, no longer growing much |
| linhut/gongwen-skill, imhuhe/gongwen-kit, Rimagination/gongwen-draft, etc. | Open source | 2026-01~10 | 14~103 stars | Single-point tools, small scale |
| Gongwenbao, Wanxiang Gongwen, Mishu Cailiao Xing, Zhipai Gongwen Zhushou, Xiaokonglong Gongwen Paiban | Other | Years | Commercial products | Paid SaaS or Office plugins; this project is a free skill you use inside an agent |
| WPS AI, Doubao, DeepSeek's built-in official-document writing | Big-company features | Already there | Largest user reach | General models make up data and can't produce national-standard layouts; this is the pain point the project's README leads with |

Where users come from

- **Verified** **skill aggregator sites and directory listings** : hotskills.dev's 24-hour growth chart listed it at about 10-30 (123 stars); ProSkillsMD opened Candidate issues for a dozen sub-skills about two weeks ago; ClaudePluginHub, SkillsLLM, TypingMind, Diedong Navigation (11-05)

- **Verified:** **the author pitched into awesome lists** : 11-04 hesreallyhim/awesome-claude-code #3467,11-06 ComposioHQ/awesome-claude-skills #2428; the same period saw 6 awesome-list forks; timing lines up with the +23 and +45 on 11-05 and 11-06

- **Verified** **rename, landing page, README conversion design** : renamed to dyz-gongwen-skills on 11-04, put up GitHub Pages and a social preview image, the first screen shows before/after comparisons and a competitor benchmark table, Star calls at both top and bottom, and 12 topics added

- **Verified** **overseas social daily roundup** : ranked #11-03 in the Korean Facebook daily 'PIELAB 스킬 데일리' issue 3

- Inferred **WeChat private community** : README includes a WeCom group and payment QR code; source of the +10-28 on 53 not found, suspected to be WeChat groups or an Official Account

- Inferred **X / big accounts / media** : both x-radar and X live search show 0 posts, and it has not been covered by GitHubDaily

## Why dyz-gongwen-skills (Official Document Writing Skills) took off

> Research day 2026-11-06. Repo [github.com/duyunzhoudyz/dyz-gongwen-skills](https://github.com/duyunzhoudyz/dyz-gongwen-skills) (originally gongwen-skills, renamed 2026-11-04).

### Bottom line

A Chinese official-document writing suite of 22 skills in a category already packed with players (the leader lieflat-gongwen has 2378 stars). It broke out not through big accounts reposting it, but through three things: **being the most 'complete' option, running a head-to-head benchmark against 8 same-name competitors and publishing it in the README, and the author steadily seeding it into skill aggregators and awesome lists** ; plus the timing of year-end summary and report writing season. Still small (about 300 stars), circulating within a niche circle.

### 1. Data

- Our GitHub repo data (repo_day): 10-27 58 stars → 10-28 111 (+53, highest single day) → 10-31 172 (+44) → 11-01 204 → 11-02~04 mostly flat (209/210/216) → 11-05 239 → 11-06 284. Live GitHub API: 300 stars, 16 forks, 18 watchers.

- Star growth comparison among similar repos from 10-28 to 11-06 (GitHub): lieflat-less-ai-tone +851, lieflat-gongwen +731, **this project +173** , Liuxiangjian-ai/official-document-skill +24, mizzlelover/gongwen-gbt9704-skill +19, kaogong-skill +15, linhut/gongwen-skill +12. So it's the third-fastest growing in the official-document category, and the only new project still growing recently; but it's still an order of magnitude behind the lieflat series (covered by GitHubDaily).

- The author's account was registered 2026-03, has 0 followers, and only two original repos: this one, and the sister repo **dyz-med-paper-skills** built by 11-01 (32 skills for doctors writing medical papers; our repo went 11-03 90 stars → 11-06 213 stars, API shows 239 stars). The same playbook produced a second repo with over two hundred stars within 5 days, suggesting the growth comes from a reusable playbook, not one-off luck.

- Commit history (latest 94 commits via API): 10-15 initial release, 10-16 added 15 skills in one day, 10-20~22 over two days with seventy-plus commits (multi-branch merges, 'round N fixes'), 11-04 renamed, shipped the Pages landing page, fixed the macOS install script, 'aligned promotional copy with benchmark data, removed absolute claims'. The rhythm is clearly batch-developed by an agent.

- Star timestamps not obtained (the stargazers endpoint requires auth, and our token quota ran out that day), so the daily numbers can only come from daily snapshots stored in the database.

### 2. Competitors (doing the same thing)

| Name | Where | Launched | Scale | How it differs from this project |
|---|---|---|---|---|
| larashero3-dotcom/lieflat-gongwen (Official Document Writing DNA) | Open source | 2026-09-02 | 2378 stars (11-06). | Its pitch is 'extracted from 1.02M characters of corpus, quantified style'; GitHubDaily wrote about it on X and Sina reposted it. Only covers writing style, not layout or classified-information handling |
| larashero3-dotcom/lieflat-less-ai-tone (remove AI tone) | Open source | 2026-08-20 | 5241 stars | General-purpose de-AI-tone tool, not just for official documents; this project's humanize line benchmarks against it |
| mizzlelover/gongwen-gbt9704-skill | Open source | 2026-09-09 | 848 stars | Only does GB/T 9704 layout; Zhihu has an article titled '700-star open-source Skill generates Word in one click per national standard' |
| Liuxiangjian-ai/official-document-skill | Open source | 2026-06-07 | 431 stars | A single writing skill distilled from People's Daily corpus |
| KaguraNanaga/official-document-writing-skill + document-format-skills | Open source | 2026-01 | 309 / 238 stars | Writing and layout split into two repos, started earliest |
| KeWang0622/kaogong-skill | Open source | 2026-04 | 223 stars | Aimed at civil-service exam prep, not in-job document writing |
| linhut/gongwen-skill, imhuhe/gongwen-kit, Rimagination/gongwen-draft, wzbwan/gongwen-format-skill, Amer-CN/super-official-writer, etc. | Open source | 2026-01~10 | 14~103 stars | Single-point tools (layout or writing), all small in scale |
| Gongwenbao, Wanxiang Gongwen, Mishu Cailiao Xing, Zhipai Gongwen Zhushou, Xiaokonglong Gongwen Paiban (Word/WPS plugins) | Web / Windows / plugin | Have been around for several years | Commercial products | Paid SaaS or Office plugins; this project is a free skill you use inside an agent |
| WPS AI, Doubao, DeepSeek's built-in official-document writing | Big-company features | Already there | Largest user reach | General models make up data and can't produce national-standard layouts; this is exactly the pain point the project README leads with |

Also, the blog 夜雨飘零博客 (blog.yeyupiaoling.cn) batch-indexed a dozen-plus 'official document writing skill' entries since October, and several are on Tencent SkillHub. This category is **already very crowded** .

### 3. Where users came from (growth tactics)

1. **Passive listings on skill aggregators and directories (confirmed)** : hotskills.dev's '24-hour growth chart' picked it up around 10-30 (123 stars then); ProSkillsMD/proskills opened a Candidate issue for each of its dozen-plus sub-skills about two weeks ago (#10032, #10033, #10046, #10048, #10193, #10194, #10206, #10286, etc.); ClaudePluginHub, SkillsLLM, TypingMind skills, RepoMasters, and Diedong AI Navigation (11-05) all have listing pages. The Korean Facebook 'PIELAB 스킬 데일리' 11-03 issue (docs/office theme) ranked it #3.

2. **Author submitting to awesome lists (confirmed)** : 11-04 filed #3467,11-06 on hesreallyhim/awesome-claude-code and #2428 on ComposioHQ/awesome-claude-skills; 11-04 and 11-06 also forked 6 awesome-agent-skills / awesome-claude-skills lists (including a zh version) in the two days before, looks like preparation for PRs. Timing lines up with the +23 and +45 spikes on 11-05 and 11-06.

3. **Rename and packaging for search and conversion (confirmed)** : on 11-04 renamed to dyz-gongwen-skills (with a personal prefix, to build a series brand), added a GitHub Pages landing page, social preview image, a first-screen before/after comparison (P0 changed 5→0, P1 31→0), llms.txt, and 12 topics (claude-skills, codex, cursor, deepseek, gongwen...); both the README top and bottom carry a 'give it a Star' call to action.

4. **Easy to install, works for non-programmers** : Claude Code plugin marketplace in two commands, `npx skills add`, and a one-click script supporting 7 agents, plus a 'web prompt version' usable without installing (copy-paste into DeepSeek, Doubao, Kimi, or an internal government/enterprise model). The target users are people who write materials in government agencies and state-owned enterprises, and this decides whether visitors can actually use it.

5. **Private community and monetization entry points** : README includes a WeCom QR code ('join the official-document skill discussion group', commercial licensing also goes through here) and a WeChat payment QR code; license changed from 10-22 MIT to PolyForm Noncommercial, commercial use requires separate authorization. The link on the author's GitHub profile points to 1keyapi.com, a unified large-model API access and reconciliation service built for 'universities, research institutes, and enterprises'. I infer the two skill repos (official documents for government and SOEs, medical papers for doctors) are bringing the same batch of institutional customers to 1KeyAPI, **this has not been verified** .

6. **Not found** : no dedicated posts about it were found on X (both the local x-radar database and live search show 0 posts), GitHubDaily, Zhihu, Xiaohongshu, V2EX, or linux.do. The source of the +53 on 10-28 **could not be traced** , possibly WeChat groups or Official Accounts that search engines don't index. It has never appeared on GitHub Trending (card trending_days_any=0).

### 4. Why it took off

- **Demand and season line up** : official documents are a hard requirement in the public sector, and every year from 11 to December is the peak for year-end summaries, performance reports, and work-key-point documents (Smzdm 10-10 has a discussion titled 'Year-end material season kicks off, the pen-pushers' circle agrees on a consensus'). The lieflat series is also growing in the same window, which shows the whole category is heating up; this project is riding the tailwind.

- **Claiming the 'most complete' spot in a crowded category** : competitors either only write, only lay out, or only remove AI tone; this one packages drafting (statutory documents, comprehensive texts, speeches, party-building, meeting minutes, rectification reports...), revision quality checks, national-standard layout, and data chart generation into 17 core skills, plus 5 industry packs for hospitals, universities, primary and secondary schools, SOEs, and township governments, giving the widest coverage.

- **Naming competitors in the README for direct comparison** : ran a head-to-head with 8 GitHub competing skills plus a 'no-skill' control group, 276 times; layout audit score 97.2 vs the runner-up's 79.6, and classified-topic checkpoint pass rate 91.7% vs the rival's 0%. In a field of similar-sounding alternatives, this is the reason most likely to make people trust it and hit star. It also adds a 'few limitations' note (the judge is an LLM, not a human), which makes it look more credible.

- **The hook hits the 体制内 worry:** "anti-fabrication, keeps secrets" speaks to "AI writes documents and makes up data" and "classified material can't leave the network." (Sohu's November pieces had security-warning articles like "Would you dare use AI for official documents?"), so it shipped an "offline intranet package."

- **Distribution is manual and high-volume:** rename, landing page, submit to aggregator sites and awesome lists; its sister repo used the same playbook and hit 239 stars in 5 days, which shows the playbook is reusable.

### 5. Differentiation, and whether that's why it took off

There are three differentiators: (1) a suite: write, revise, format, and check, plus industry packs, all in one; (2) it follows the rules: when material is missing it marks [TO BE FILLED] instead of inventing, it checks classification levels, and when a font is missing it says so instead of silently swapping it; (3) it proves itself with a head-to-head competitor eval, and it ships a prompt version for web LLMs and intranet models. **Is that the reason?** Partly. Differentiation decides whether people who land on it choose it, and the eval table is strong conversion material. But it has only about 300 stars in total, far below lieflat-gongwen, which has over two thousand stars with no eval and just one GitHubDaily post. That suggests in this niche, **exposure channels matter more than product quality for star count** . Its growth comes mainly from seasonal demand plus the author's own distribution work; differentiation governs conversion.

### 6. How visible it is

Word of mouth: it was only picked up by skill aggregators, navigation sites, and a Korean Facebook daily. No big accounts or media wrote about it, and revenue is not public. The niche itself was already popularized by GitHubDaily, and there are many competitors.

### 7. Copycats

I found no clear copycat. But new projects in the same niche keep appearing, such as ianlinkstudy/chinese-formal-document-format (created 10-25, 20 stars, formatting only), and most of the dozen-plus official-document skills collected by the Night Rain Drifting blog were launched in October.

### 8. What others can learn

- Pick a rigid-demand Chinese vertical with national standards that can be measured objectively (official documents, medical papers, etc.), use agents to batch-build a "suite" plus "industry packs," and claim the "most complete" position.

- Run head-to-head benchmarks against same-name competitors, put the comparison table at the top of the README (with a scope caveat). Among many similar projects, this converts best.

- One author runs a series of repos (dyz- prefix) with the same template and the same distribution (aggregators, awesome lists, Pages, social preview images), cloning a new one each week.

- Prepare a "copy-paste prompt version" and an offline intranet package for users who don't use the CLI, catching non-programmers.

- Use a non-commercial license plus WeCom groups plus commercial licensing to turn stars into institutional client leads.

### 9. Still unclear

- The two jumps, 10-28 (+53) and 10-31 (+44): source not found, possibly WeChat groups or an Official Account.

- Whether the stars are real and how many people actually use it: it has only 16 forks and no issue discussions, so actual usage is impossible to tell.

- The commercial relationship with 1KeyAPI is only an inference.

- Whether the awesome-list PRs and issues were merged.

### Sources

- GitHub API: repos/duyunzhoudyz/dyz-gongwen-skills, users/duyunzhoudyz/repos, commits (pulled 2026-11-06)

- Our GitHub repo repo_day (2026-10-27 to 11-06)

- README: [github.com/duyunzhoudyz/dyz-gongwen-skills](https://github.com/duyunzhoudyz/dyz-gongwen-skills); eval table evals/results/compete-v7-tables.md

- [hotskills.dev/skill/duyunzhoudyz-dyz-gongwen-skills/](https://hotskills.dev/skill/duyunzhoudyz-dyz-gongwen-skills/) ;[hotskills.dev/trending/](https://hotskills.dev/trending/)

- [github.com/ProSkillsMD/proskills/issues/10032](https://github.com/ProSkillsMD/proskills/issues/10032), etc.

- [github.com/hesreallyhim/awesome-claude-code/issues/3467](https://github.com/hesreallyhim/awesome-claude-code/issues/3467) ;[github.com/ComposioHQ/awesome-claude-skills/issues/2428](https://github.com/ComposioHQ/awesome-claude-skills/issues/2428)

- [www.facebook.com/oopchoi/posts/10246891906453640/](https://www.facebook.com/oopchoi/posts/10246891906453640/) (PIELAB 스킬 데일리,11-03)

- [www.diedong.com/duyunzhoudyz-dyz-gongwen-skills.html](https://www.diedong.com/duyunzhoudyz-dyz-gongwen-skills.html) (11-05)

- [www.1keyapi.com/home](https://www.1keyapi.com/home)

- Competitors: [x.com/GitHub_Daily/status/2097988479280160850](https://x.com/GitHub_Daily/status/2097988479280160850); [zhuanlan.zhihu.com/p/2085769570321748989](https://zhuanlan.zhihu.com/p/2085769570321748989); [post.smzdm.com/p/ae6q9973/](https://post.smzdm.com/p/ae6q9973/)

- X: x-radar search, x live search "gongwen-skills" (0 results, 2026-11-06)

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