A repo with no code has 9.8k stars: a Nanjing University professor's "Research Getting-Started Guide," with Douyin and Xiaohongshu resource accounts doing the star-gathering for it

A Ph.D. supervisor at Nanjing University posted a "Research Starter Guide" on GitHub in September. The repo was just a README, with the full text on Feishu. 69 days later it had 9,835 stars. Line up the daily star counts against the posts on Douyin and Xiaohongshu and the pattern is plain: every jump follows a repost with thousands to tens of thousands of likes, and almost none of those posts were made by the author. Below: how it took off, what the reposters get out of it, and which steps of this playbook you can copy for your own open-source project and which you can't.

Oct 3: a note with 21k saves

On Oct 3 of 2026, a Xiaohongshu account called "Xiaomaolü" posted a note titled "Strongly recommend the research starter guide compiled by a Nanjing University professor," with a body that said "posted it on Douyin yesterday." The note ended with 13,070 likes and 21,074 saves. Saves outnumbering likes is the classic "bookmark it for later" pattern.

That day the LAMDA-NeSy/Research-Starter-Kit repo added 556 stars, and 898 the next day. It ranked #10 and #6 on trendshift's daily GitHub overall chart (trendshift daily data). Before this it had roughly two to three thousand stars (estimated), and for more than a week afterward it kept gaining three to four hundred a day.

同一个国庆假期,抖音上「小年」(1,358 个粉丝)10-04 发的同题视频有 7,366 个赞;小红书「胖胖的科研笔记」10-06 的标题是「南大教授蒸馏的这个skills」,1,760 赞 / 3,822 藏。

What it actually is: a table of contents with a dozen Feishu links

The repo was created in Aug 29, has 7 commits in total, and its last commit was Sep 1. Nothing has changed since. There's no code and no license. The README is a table of contents:

• How to do research: finding papers, reading papers, generating ideas, academic presentations, meetings, good habits
• How to write papers: drawing figures, then section by section through Abstract, Introduction, Related Works, Methods, Experiments, References, and finally Rebuttal
• Misc: conference attendance guide, scholarships you can apply for, tips on using AI for research

Each entry links to a Feishu wiki. The author, Guo Lanzhe, is a tenured associate professor and Ph.D. supervisor at the School of Intelligence Science and Technology at Nanjing University, with his group under Zhou Zhihua's LAMDA lab. The README explains why it exists: every year he has to repeat the same basics to incoming students, and "some students may be left to fend for themselves, without systematic guidance." It closes with: "If you find this project helpful, please star it in the top right corner!"

The content itself isn't new. The acknowledgments cite Zhou Zhihua's "Doing Research and Writing Papers," Peng Sida of Zhejiang University's "My Research Experience," and Wei Xiucan of Southeast University's article on writing rebuttals. The repo has 368 forks (about 3.7%), because there was nothing in it to fork. GitHub did one job here: it was a credible-looking place to click star.

Where the stars came from: a wave of posts, then a wave of stars

Lining up trendshift's daily star counts against the Douyin and Xiaohongshu posts findable by date:

Two things stand out. First, the author's own post brought only a few hundred stars, and afterward it was a dozen a day for a week. Without someone else reposting it, it would likely have stayed under a thousand. Second, GitHub's own recommendation surfaced the repo for only two days, on 10-03 and 10-04. That was a result of off-platform traffic, not its cause. There's no trace of it on X, Hacker News, or tech media.

The best control is in the same lab: LAMDA-NeSy/Lab-RoadMap is a lab onboarding guide from the same professor, published in August 2025. The content is similar, nobody reposted it, and it has 373 stars today.

What the reposters get out of it

这些抖音、小红书账号的昵称很有特点:「滋.料在主页」「资源看简介」「资料👀简介」「籽料.在主业」。小红书上「南大教授科研Skill」这个标题出现了十几次,话题标签也是同一套(#howto入门codex #人类高质量科研工具 #科研skill)。评论区几乎全是「求」「已关,求」。

This is a mature playbook: use a free guide from a prestigious university as bait, get people to follow and comment, then route them to DMs or a private channel, where the money comes from the account's paid resource packs, courses, or ads. (The 938-like note from Miracle had 782 comments, and some of them were people selling A100 / H100 compute rentals.) Douyin's "Zhang Yu's AI Research Corner" put it bluntly in the title: "A 300 research course, and it still doesn't cover everything in this free guide?" The author got nothing, and I found no sign he has any arrangement with these accounts. Whether they're the same operation can't be confirmed. The only evidence is highly similar titles and tags.

There's also a side benefit they rode: in 2026, "research Skills" were trending on Xiaohongshu (a "Codex skill summary" post had 10,506 likes), so a Feishu doc got rebranded as "skills distilled by a Nanjing professor." It isn't really a Skill and can't be loaded into Codex or Claude.

For resource accounts, this guide is close to perfect material:

• Ready-made title: "Nanjing," "Ph.D. supervisor," and "LAMDA" up front is enough. Nearly every viral post is written this way.
• Timing is ready-made: September intake and the National Day holiday. Students 0 and first-years are right in the middle of looking for answers to "what if nobody guides me."
• Screenshot-friendly and save-worthy: one question per page, from finding papers through Rebuttal to scholarships. Screenshot it as a nine-grid and you have a post.
• Free, no git needed: open the Feishu link and read. GitHub only serves to say "this really was posted by a Nanjing professor."

Who's following, and how they're doing

After it blew up, a batch of follow-ups appeared. The numbers are small (GitHub, 11-06):

Looking further back, the largest in the category is Zhejiang University's pengsida/learning_research (2023 years old, 15.9k stars), then hzwer/WritingAIPaper (English, covers only writing AI conference papers, 4.2k), and research_tao (NLP research intro, 2.1k). The topic of a research starter guide has been done many times already. The ones that collect lots of stars all carry a name from a prestigious university, where the author supervises students and publishes at top venues.

The best follow-up got just over two hundred stars, for a simple reason: the reposting accounts need the four characters "Nanjing professor." A student's account writing a similar guide, or someone's rewritten Skill version, has no such title to use.

If it were you

One-line verdict: this project shows a full playbook for getting stars through off-platform promotion. But its key piece, the name of a sitting Ph.D. supervisor at a famous university, is out of reach for followers, so copying a same-topic guide isn't worth it. Grade C.

Who it suits: people who already have a title or an audience, such as university faculty, well-known doctoral students, or tech bloggers with tens of thousands of followers in a field. Also people who hold a list others genuinely want to save and can find a credentialed person to co-sign it.

Who it doesn't suit: ordinary developers hoping to get stars by "writing another research starter guide too." The topic is already overdone, reposting accounts won't pick up a version without a name, and the best follow-up got only 214 stars.

If you do want to build this, how:

• Pick a timely need that nobody has covered thoroughly yet, instead of another research starter guide: graduate school summer camps and early admissions (5–July each year), retake prep for the graduate exam interview (March), the month before a first top-conference submission, PhD cold-emailing, and international students hunting for internships. Each has a new batch of people every year asking the same questions.
• Copy the format: the GitHub repo holds only a table of contents and a line saying "star it"; the body lives in Feishu or Notion, one question per page, laid out so it screenshots well as a nine-grid.
• Put a real, credible person or institution on the byline. If you don't have one, find a professor or lab willing to co-sign.
• Post the first item on Xiaohongshu yourself, then make it easy for resource accounts to repost it: put the institution's name in the title, use the hot term of the moment (right now that's Skills / Codex), and set no gate and no QR code, so reposting carries no risk for them.
• If your strength is code rather than reputation, fill the gap this project left: turn guides like this into research Skills that really load into Claude or Codex, with original content of your own, not just a rewrite.

When to move before: this kind of content follows the academic calendar. The next big window is September 2027; other workable moments are the graduate exam retake season in March and the early admissions season in 5–July. Publish two to three weeks before each one to give reposters time.

What would show you're wrong: if the next same-topic guide without a prestigious byline also gets reposted widely and reaches several thousand stars, then the name doesn't matter as much as I think, and this idea is worth more than a C. Conversely, if your repo has been out for two weeks with no off-platform post from anyone but you and the stars stall in the hundreds, stop adding content. The problem is distribution, not the content.