# "One person running an account, and the day is gone": Zhejiang and Peking University's social-media agent Easel rode the Chinese AI self-media circle, 70 days, 4,400 stars · GitHub open-source project

> Source: https://stringers.io/en/github/gh_ZJU-REAL__Easel.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

> In September, a GitHubDaily line, "Running one account for a day wipes you out," lit the fuse; New Intelligence then ran a press release framed as "Zhejiang x Peking University," and AI-tool bloggers on Xiaohongshu, Zhihu, and Weibo followed up with hands-on tests, riding the "let OpenClaw do my Xiaohongshu" wave.

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

ZJU-REAL/Easel · ZJU-REAL

# "One person running an account, and the day is gone": Zhejiang and Peking University's social-media agent Easel rode the Chinese AI self-media circle, 70 days, 4,400 stars

**What it is** An open-source social-media ops workbench running on OpenClaw: 113 skills that find trending topics, write copy, make images and edit video, publish to 10 platforms including Xiaohongshu and Douyin, then pull back data for review and keep a profile for each account.

**Why it took off** In September, GitHubDaily's line "one person running an account and the day is gone" lit the fuse. New Intelligence then ran a press release framed as "Zhejiang x Peking University", and AI-tool bloggers on Xiaohongshu, Zhihu and Weibo followed with hands-on tests, riding the wave of "let OpenClaw run my Xiaohongshu".

**C** **How you could follow** Don't build another Chinese Easel. Copy its launch playbook: write the selling point as a user's day, make a vertical bundle on the hottest agent runtime, put real outputs and risks in the README, and pick users who will post about it themselves. If you do build, consider an English platform version or fill its fact-checking gap.

Grade C: the stars mostly came from university endorsement and press coverage, which followers can't replicate. Generation and distribution already have several older projects with tens of thousands of stars, and the best same-category skill packs in the same period reached 575 stars at most. What's worth learning is the distribution playbook. Spreading in small circles

Data (updated daily) · Repo created 2026-08-28 · 4,490 stars · 7 days +247

GitHub · stars

Data 2026-08-31 → 2026-11-08 (70 days, 32 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Easel is a "social-media ops agent" open-sourced by Zhejiang University and Peking University labs. It runs on OpenClaw and chains five steps into one pipeline: finding trends, picking topics, making images and video, publishing to platforms, and collecting data for review. It didn't invent any new capability; each feature has older projects with more stars. It took off because of a relatable pain point, a "Zhejiang x Peking" label, and a group of users who already post a lot. Later same-category skill packs reached 575 stars at most. Anyone building a similar project should study how it got spread.

### Sep 8 morning tweet

On 2026 Sep 8 7:30 (UTC), the Chinese open-source recommender GitHubDaily posted on X, opening with:

> One person runs an account: in the morning, scan the hot list for topics; at noon, write copy and make covers; in the evening, edit video. After publishing to 6 platforms, you still have to check each one's data, and the day is gone.

It continues: made by Zhejiang and Peking University labs, running on OpenClaw, with 112 skills that are "all scripts that actually run", a profile for each account, and published data fed back into the profile. The last line: "The README itself warns that Xiaohongshu checks automation closely and recommends previewing then posting manually, which is refreshingly honest." The post got 36.3K views (X, November).

At that point the repo was only 11 days old. Created on Aug 28, v0.1.0 released on Aug 31, 65 stars that day; about 275 in the first week. By Nov 6 it had 4,440 stars and 735 forks (GitHub data, 11-06).

### What it is

A local workbench with a web UI that manages sessions, assets, accounts, a content library and a calendar. Underneath are 113 skills (the November README number; it was 112 in September), covering copy, Xiaohongshu knowledge cards, posters, infographics, voiceover, voice cloning, subtitles, editing and AI short dramas. The hot-topic radar aggregates trending lists from Weibo, Douyin, Zhihu, Bilibili, Baidu and Toutiao. Publishing has expanded from 6 platforms in September to 10 now, adding WeChat Official Accounts, X, Juejin and Toutiao.

What it really sells is the "account profile": six dimensions of positioning, style, audience, platforms, preferences and red lines, plus long-term memory. Post-publish data flows back into it, so the next generation fits the account better.

Setup is not light: it needs Python, Node, FFmpeg and Playwright, plus OpenClaw. Even so, the fork-to-star ratio is 16.6%, higher than the median (11%) for new repos with 5,000+ stars in the same period, which suggests many people actually pulled it down and installed it.

### Where the stars came from: a relay chain in Chinese self-media

| Timeline | What happened | Stars around |
|---|---|---|
| 08-31 | v0.1.0 released | 65 |
| 09-06 | The Chinese recommendation account @QingQ77 posted on X, 2,818 views | — |
| 09-08 | GitHubDaily post, 36.3K views | about 275 in the first week |
| 09-09 | The first wave of introductions appeared on Weibo | — |
| 09-16 | New Intelligence published a press release, "Self-evolving social-media agent! Zhejiang x Peking University", and Tencent News, Sina, Sohu and BAAI Community reposted it | About 940 in September |
| 09-24 | v0.2.1 released | About 1,700 |
| 09-25 to 09-30 | Dozens of notes appeared on Xiaohongshu, Zhihu ran a test "1 hours to make 7 pieces of content", and Bilibili UPs followed | about 09-30 2,500 |
| 10-01 | AI blogger @张岱樾 with 380K followers reposted on Weibo: "It's like giving a one-person self-media account a team of AI content staff that reviews its own work", 175 likes, 124 reposts | — |
| From ¥10-03 | Word of mouth split: a Xiaohongshu "AI guide" post documented a failure (without web search enabled, the agent made up facts), with 111 likes and 343 saves | about 10-09 3,280 |
| From October to now | Aggregator sites, search and Zhihu teardown articles brought a long tail, 30–50 a day | 11-06 4,440 |

(September's stars are reconstructed from third-party snapshots and reposts, since GitHub doesn't give per-star timestamps for this repo, so only rough estimates are possible. From October on, the data is recorded daily.)

The fastest rise was the week of Sep 25 to Oct 1, not the day of GitHubDaily. That week, posts on Xiaohongshu clustered together, like the "AIJJ productivity tool" post with 116 likes and 447 saves, pitching "112 composable skills, one-click distribution to seven platforms." These posts appeared right around the Sep 24 release. One reposted piece said "half promotion, half real testing," so the team may have seeded a lot of it themselves, but no partnership disclosure was found, **unverified** .

Also worth noting, the paths not taken: 36 氪 and 钛媒体 never covered it, the English-language scene has only scattered LinkedIn posts and aggregator sites, nothing on HN or Reddit, and I saw no paid ads. It also never hit GitHub trending. When it blew up in September the repo was under 60 days old, so by convention it couldn't chart. The whole chain played out almost entirely inside Chinese AI self-media circles.

### Why this crowd was willing to share

**The pain point was written as one person's day.** "Morning topic hunting, noon cover-making, evening video editing": anyone running self-media sees themselves in it. What got shared was a complete story, not a feature list.

**The brand helps.** The README's first line shows both the Zhejiang University and Peking University logos, and the 新智元 press release headline is literally "ZJU x PKU." The phrase "985 Lab open-sources" carries built-in reach in Chinese tech media. Other repos from the same lab get 1,370 stars at most. The lab had no audience of its own before, so this brand is the traffic.

**It rode the OpenClaw wave.** OpenClaw has 390K stars (GitHub, November), and throughout 2026 the Chinese sphere was full of tutorials on "let OpenClaw run my Xiaohongshu". Easel was a ready-made bundle for that crowd.

**The users are posters themselves.** Those who tried it are self-media creators, and after trying it they write reviews. Failure posts also drive traffic.

**It looks like a real product.** It has a project homepage, promo videos and posters, and the README shows lots of real outputs (paper-reading cards, novels, short dramas), with a release every two weeks. Explicitly writing "post to Xiaohongshu with caution" made recommendation accounts see it as something other than yet another marketing-account tool.

### Who it's competing with

| Project | What it does | Stars (November) |
|---|---|---|
| gitroomhq/postiz-app | Open-source alternative to Buffer for scheduled overseas publishing | 40,965 |
| yikart/AiToEarn | Multi-platform auto-publishing, pitched as "making money with AI" | 27,308 |
| xpzouying/xiaohongshu-mcp | A single-platform API for Xiaohongshu only | 16,936 |
| dreammis/social-auto-upload | Batch video upload to Douyin, Xiaohongshu and WeChat Channels | 16,150 |
| zhouxiaoka/autoclip | Long video cut into short clips | 10,008 |
| TeamWiseFlow/xiaobei | Self-media lead generation for SMEs | 8,687 |
| ddean2009/MoneyPrinterPlus | Batch remixing for volume posting | 7,309 |
| HisMax/RedInk | Generates Xiaohongshu image-text posts only | 5,691 |
| Xiangyu-CAS/xiaohongshu-ops-skill | Also on OpenClaw, but Xiaohongshu only | 2,513 |

Distribution, single-platform APIs and single-point generation have all been done before, and those projects have more stars. A Zhihu teardown put it bluntly: "Easel doesn't build an agent core itself; it's an integration layer." Its position is packaging these capabilities on OpenClaw and adding profiles and review.

Copycats came too, but none were big: cv-cat/catbus (created 9-27, 575 stars), taishang-laojun (10-22,495), BullFlow (10-14,335), AttentionOS (10-15,247), all same-category skill packs, with nobody naming Easel as the benchmark. Another "full-chain" project, tsingyuai/growth-lab, is stuck at 1,971 stars and hasn't grown in nearly 7 days.

### It's easy to share, but whether it keeps people is unknown

There's no public long-term test yet to back the "profiles get more accurate with use" claim. User feedback runs "somewhat useful but not by much" and "the cost may not even pay back": token use is high and setup is hard. After mid-October, the daily growth of 30–50 comes mostly from search and aggregator sites, not word of mouth compounding. So it spread because the pitch was easy to repeat; whether it's actually good after installing has no evidence yet.

### If it were you

**Verdict** : Don't build another Chinese Easel, but do learn its launch method. Most of its stars came from the "Zhejiang x Peking" label and the New Intelligence press release, which followers can't get. The feature layer is already crowded with older projects at tens of thousands of stars, and later same-category skill packs top out at a few hundred stars.

**Not a fit** : anyone without institutional backing or an audience of their own, planning a "more complete Chinese social-media agent" and hoping it gains stars on its own. Most likely outcome: the catbus and BullFlow tier.

**A fit** : people or teams with an application-type open-source project whose target users are content creators, especially university labs. A few moves are worth copying:

1. **Write the selling point as a user's day.** The first line isn't "we support 113 skills", it's "and the day is gone".

2. **Bet on the hottest agent runtime and build a vertical bundle on top of it.** Don't build your own core; package the full workflow for one audience.

3. **Put real outputs in the README and be upfront about risks.** Admitting "RedNote may flag this" got it a separate shout-out from a recommender account.

4. **Pick users who post on their own.** Build for content creators, designers, and tutorial bloggers. Users who try it will write reviews for you, and even bad reviews are exposure.

If you really want to work in this space, there are two open slots, both small. One is an English-platform version: overseas there are scheduling tools like postiz and research-oriented skill packs (the ScrapeCreators set, 5,770 stars), but few agents that run from topic selection through post-mortem. The other is what it does poorly, such as reliable fact-checking (the failure posts were about exactly this) and measurable persona results.

**How to get users** : follow this chain. First, get Chinese open-source recommender accounts (GitHubDaily, Geek Lite and similar) to post on X. If a university or company backs it, pair that with a research press release in the style of 新智元 or 机器之心. Within a week of a release, get AI tool bloggers on Xiaohongshu and Zhihu to publish hands-on tests. For an English version, switch to English developer accounts on X and Reddit.

**When the window closes** : the OpenClaw skill-pack wave is still running, but similar packs appear every week. To ride the "OpenClaw + platform operations" framing for stars, the window probably closes around early 2027 (estimate).

**What would show I got it wrong** : if an independent project with no institutional backing reaches 2,000 to 3,000 stars through recommender accounts and hands-on posts alone, the demand is big enough to work without the brand. Then "swap the platform or audience and build another one" is worth reconsidering. Conversely, if Easel's daily star gain drops to single digits and issues go unanswered, this kind of integration layer doesn't retain users, and copying its launch playbook should be discounted.

[GitHub repo ↗](https://github.com/ZJU-REAL/Easel)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| yikart/AiToEarn | Open source | 2025-02 | 27,308 stars | Focuses on multi-platform auto-publishing and AI-driven earnings. Easel covers the full chain from topic selection to post-mortem, plus account personas. |
| dreammis/social-auto-upload | Open source | 2023-12 | 16,150 stars | Only handles video upload. Easel also covers creation and post-mortem. |
| xpzouying/xiaohongshu-mcp | Open source | 2025-08 | 16,936 stars | A single-platform capability API for Xiaohongshu. Easel is the multi-platform workbench above it. |
| Xiangyu-CAS/xiaohongshu-ops-skill | Open source | 2026-02 | 2,513 stars | Also built on OpenClaw, but only one Xiaohongshu skill. Easel has 113 skills, a web workbench, and 10 platforms. |
| gitroomhq/postiz-app | Open source | 2023-07 | 40,965 stars | An overseas scheduling and publishing tool. Easel targets Chinese platforms and includes generation and post-mortem. |
| TeamWiseFlow/xiaobei | Open source | 2024-04 | 8,687 stars | Leans toward SME lead generation. Easel leans toward content production. |
| ddean2009/MoneyPrinterPlus | Open source | 2024-05 | 7,309 stars | Batch-remixes content and scales distribution. Easel leans toward fine-grained operation of a single account. |
| HisMax/RedInk | Open source | 2025-11 | 5,691 stars | Generates Xiaohongshu image-text posts only |
| Yixiaoer / Yimei Assistant / Xinbang Xiaodouya | Other | Years | Domestic commercial SaaS, scale unverified | Paid multi-account distribution and analytics dashboards, with no agent-based generation or persona loop |
| Official AI creation assistants from each platform | Big-company features | 2024- | Full user base | Serves only its own platform. Easel works across platforms. |

Where users come from

- **Verified** **X Chinese open-source recommender accounts** : 09-06 @QingQ77 2,818 views; 09-08 @GitHub_Daily 36.3k views

- **Verified** **Tech media press release (新智元)** : 09-16 新智元 "Self-evolving social media agent! Zhejiang U x Peking U", syndicated to Tencent News, Sina, Sohu, and the BAAI community

- **Verified** **Hands-on posts from AI tool bloggers on Xiaohongshu, Zhihu, and Bilibili** : 09-25 to 09-30, a dozen-plus Xiaohongshu notes (AIJJ productivity tool with 116 likes and 447 saves), 09-27 Zhihu hands-on test, 09-29 on Bilibili. Over the same period, stars rose from about 1,700 to 2,500.

- **Verified** **Weibo AI influencers** : 10-01 张岱樾 (380k followers) reposted it as a recommendation, with 175 likes and 124 reposts

- **Verified** **University backing** : the README leads with the Zhejiang U and Peking U logos, and the press headline is "Zhejiang U x Peking U"

- Inferred **Team seeding on Xiaohongshu** : the notes clustered after the 9-24 release. 什么值得买 says "half of it is promotion," but no sponsorship disclosure turns up.

- **Verified** **Aggregator sites and long-tail search** : OSCHINA, pyshine, agentskillshub, Trendshift badges, and Zhihu teardown articles. From October onward, 30-50 stars per day.

## Why ZJU-REAL/Easel took off

> Research date 2026-11-06. Data comes from our GitHub database (repo_profile / repo_day), the GitHub API, Jina search, and page text. The GitHub stargazer detail endpoint returned 404 for this repo, so per-star timestamps are unavailable. The September curve was reconstructed from third-party pages.

### One line

Easel is an open-source "social media operations agent workbench" built jointly by the Zhejiang University REAL Lab and the Peking University OpenDCAI Lab. It runs on OpenClaw and chains five steps into one pipeline: hotspot discovery, topic planning, image-text and audio-video creation, publishing to 10 Chinese and overseas platforms, and post-publish data review. It also builds and continually updates a persona for each account.

It took off through **relay sharing across the Chinese AI self-media scene** : GitHubDaily posted on X, 新智元 issued a press release, and AI tool bloggers on Xiaohongshu, Zhihu, and Weibo followed with hands-on posts. What got shared was "a university lab product where one agent runs an entire account's operations," which rode the 2026 wave of "install skills on OpenClaw and let it do my Xiaohongshu." Stars grew fastest in the week from 25 to 1 (roughly 1,700 to 2,500), then settled into organic growth of 30–50 per day.

### 1. Data: what the curve looks like

| Date | star | Source |
|---|---|---|
| 2026-08-28 | 0 (repo created) | GitHub API |
| 08-31 | 65 | GitHub repo_day (Wayback snapshot); v0.1.0 released the same day |
| About 09-04 | 275 (first week after launch) | LinkedIn, post by Arthi R. |
| Around mid-September | 938 | OSCHINA reprint |
| About 09-24 | 1,697 (+441 that week, +168 that day) | RepoPulse snapshot |
| 09-30 | 2,497,fork 381 | 什么值得买 article "Easel hits 2.5k stars" |
| 10-09 | 3,277,fork 504 | github repo_day (deeply) |
| 10-13 | 3,554 (peak single-day gain during observation: +91) | GitHub repo |
| 11-06 | 4,440, forks 735; over 7 days +247, first 7 days +197 | github library repo_profile |

- From 9-07 to 11-06 total +4,091 (Trendshift).

- Fork-to-star ratio is 16.6%, above the median of 11% for newer repos with ≥5000 stars in the database. That suggests many people actually cloned and installed it (requirements include Python, Node, FFmpeg, Playwright, and OpenClaw, which is a high bar). The suspect list is empty.

- 454 commits in total, 5 releases (v0.1.0 08-31 → v0.3.0 10-12). The main committer lidingm has 138 commits, plus about a dozen other contributors. Issues and PRs total 65.

- Has not made GitHub Trending (trending_days_any = 0). The repo was under 60 days old when it took off in September, and GitHub normally excludes repos that young. It is now 70 days old and still hasn't made it. Trendshift has awarded Python daily and weekly chart badges.

- The largest other repo in the same org is ClawGUI (1,370 stars). Easel alone accounts for 9,714 of the org's total stars, 4,440, which means the lab did not have a traffic pool this big before.

### 2. Where users came from (timeline)

1. **09-06 X @QingQ77 (Geek Lite)** : a Chinese tech recommender account, post with 2,818 views. [Link](https://x.com/QingQ77/status/2096623205717021089)

2. **09-08 X @GitHub_Daily** : a top Chinese open-source recommender account, post with 36.3k views. The opener, "One person running an account: in the morning scanning trending lists for topics, at noon writing copy and making covers, in the evening editing video... and the whole day is gone," names the pain point directly. It also calls out the README's own admission of RedNote ban risk. [Link](https://x.com/GitHub_Daily/status/2097225972366114879)

3. **09-09 Weibo** : the first wave of introductions appeared (per the @AIGC·非著名程序员 Weibo post cited by a Stringers article).

4. **09-16 新智元** : published "Self-evolving social media agent! Zhejiang U x Peking U: from multimodal task orchestration to feedback-driven", syndicated at the same time to Tencent News, Sina, Sohu (an ITBEAR rewrite), the BAAI community, and AI news aggregators. This is a typical university-lab press release, framed in research language ("agent self-evolution", "feedback-driven"). [Tencent News](https://news.qq.com/rain/a/20260916A07J9U00), [BAAI](https://hub.baai.ac.cn/view/58037), [Sohu](https://www.sohu.com/a/1076775163_362225)

5. **09-25 to 09-30 Xiaohongshu saw a cluster of a dozen-plus notes** : for example, the "AIJJ productivity tool" note got 116 likes and 447 saves, pitched as "112 composable skills, one-click distribution to seven platforms." **09-27 Zhihu** "cost-effective life" published a full hands-on test ("1 hours to make 7 pieces of content"), and **09-29 Bilibili** also had UP creators follow. This was the week stars grew fastest, roughly from 1,700 to 2,500. Sources: [什么值得买 9-30](https://post.m.smzdm.com/p/a4qpo0dx/), [什么值得买 10-09](https://post.m.smzdm.com/p/al38dev8/)

6. **10-01 Weibo @张岱樾** (AI influencer with 380k followers): reposted it as a recommendation, saying "It's like giving a one-person self-media account an AI content team that reviews its own work," with 175 likes and 124 reposts.

7. **Word of mouth split starting 10-03** : a Xiaohongshu account, "AI Guide," posted a failure report (without WebSearch configured, the agent invented facts), with 111 likes and 343 saves. Comments started to include takes like "you may not even recoup the cost" and "useful, but not that useful."

8. **Then the long tail** : Zhihu teardown articles ([link](https://zhuanlan.zhihu.com/p/2094752180486124749)), OSCHINA, pyshine, agentskillshub, heyaican, and other English and Chinese aggregators picked it up automatically. The README includes a WeChat group QR code.

**没查到的渠道** :36 氪、钛媒体没有报道(什么值得买文章也这么说);x-radar 语料里搜 Easel 没有命中相关推文(本地语料覆盖有限,只能说明不是 X 英文大号带起来的);HN、Reddit 也没搜到。也没看到投广告。

**Was the Xiaohongshu wave organized?** : a 什么值得买 author wrote that "the buzz was concentrated in the hands of self-media tool accounts, half promotion and half hands-on tests." A dozen-plus posts clustered between 9-25 and 9-30, close to the timing of the 9-24 v0.2.1 release and the 9-30 overnight commits, so a team may have seeded them. But there is no direct evidence (no sponsorship disclosure, and no shared accounts found), so this is **unverified** . Note also that the two 什么值得买 pieces themselves read like AI-generated content-farm articles, so their secondhand figures are only a reference.

### 3. Competitors

| Product | Where | Launched | Scale | How Easel differs |
|---|---|---|---|---|
| yikart/AiToEarn | Open source (Electron desktop app) | 2025-02 | 27,308 stars | Focuses on multi-platform auto-publishing and "AI earnings." Easel focuses on the full chain from topic selection to post-mortem, plus account personas. |
| dreammis/social-auto-upload | Open source | 2023-12 | 16,150 stars | Just upload your video to Douyin, Xiaohongshu, WeChat Channels, and more; Easel handles creation and review too |
| xpzouying/xiaohongshu-mcp | Open source (MCP) | 2025-08 | 16,936 stars | Only a single-platform Xiaohongshu interface; Easel is the upper-layer workbench that can plug in tools like this |
| Xiangyu-CAS/xiaohongshu-ops-skill | Open source (OpenClaw skill) | 2026-02 | 2,513 stars | Also built on OpenClaw, but covers only Xiaohongshu and is a single skill; Easel has 113 skills and a Web workbench covering 10 platforms |
| gitroomhq/postiz-app | Open source / SaaS | 2023-07 | 40,965 stars | An overseas scheduled-publishing tool (an open-source alternative to Buffer); Easel targets Chinese platforms and includes generation and review |
| TeamWiseFlow/xiaobei | Open source | 2024-04 | 8,687 stars | Positioned as an "AI agent for SMB self-media lead generation", focused on lead gen; Easel focuses on content production |
| ddean2009/MoneyPrinterPlus | Open source | 2024-05 | 7,309 stars | Batch-remixes short videos for redistribution, aimed at volume; Easel focuses on careful single-account operations |
| HisMax/RedInk | Open source | 2025-11 | 5,691 stars | Only generates Xiaohongshu image-text posts; Easel is end-to-end |
| zhouxiaoka/autoclip | Open source | 2025-07 | 10,008 stars | Cuts long videos into short ones; clipping is just one of Easel's skills |
| ScrapeCreators/social-media-research-skills | Open source | 2026-06 | 5,770 stars | A research-focused skill for overseas platforms; Easel targets Chinese platforms and handles execution |
| tsingyuai/growth-lab | Open source | 2026-07 | 1,971 stars, last 7 days 0 | An end-to-end tool for product growth, same "full-chain" pitch, but growth has stalled |
| Yixiaoer, Yimei Assistant, Newrank Xiaodou Sprout, and others | Domestic SaaS / desktop apps | Years | Commercial products, fairly large scale (unverified) | Paid multi-account one-click distribution and data dashboards; no agent-driven generation or persona feedback loop |
| Official AI creation assistants from each platform | Big-company features | 2024– | Full user base | Serves only its own platform. Easel works across platforms. |

Takeaway: multi-platform distribution, single-platform MCP, and single-point generation are all already done by others, and all with more stars than Easel. **Easel didn't invent new capabilities** . What it does is integrate these capabilities on top of OpenClaw into one workbench with account profiles and review feedback. A Zhihu teardown put it this way: "Easel isn't building an agent core itself; it's an integration layer." Among social-media repos newly created since 2026 June, Easel ranks second in stars and first in the Chinese-platform category (GitHub repo query).

### 4. Why it took off

1. **The pain-point story travels well** : "Running one account for a day wipes you out" → "One agent handles all five steps," which AI self-media creators can turn into a post as is. What gets shared is a complete story, not any single feature.

2. **University backing plus press coverage** : the Zhejiang University and Peking University logos go on the first line of the README, and the 新智元 press release gets syndicated to Tencent, Sina, Sohu and 智源. The "985 Lab open-sources it" label performs naturally in Chinese self-media.

3. **Rode the OpenClaw and skill wave** : OpenClaw has 390k stars (GitHub repo), and throughout 2026 the Chinese web was full of "OpenClaw automates Xiaohongshu" tutorials (CSDN, Zhihu, Tencent Cloud all carried them). Easel was effectively a ready-made bundle for that crowd: 113 skills supporting 10 platforms.

4. **Polish and presentation** : a Web workbench, a project homepage, a promo video, posters, and the README shows plenty of real outputs (paper-reading cards, novels, short dramas). It looks like a product, not paper code. A new version ships every two weeks, with continuous commits since 9-30.

5. **Stating risks candidly actually helped** : the README itself warns about "being cautious with automatic publishing to Xiaohongshu". GitHubDaily and Smzdm both singled this out, which lowered the impression of "yet another marketing-account tool".

6. **The audience are content producers themselves** : users are self-media creators, so after trying it they post their own notes and test write-ups, creating secondary spread. Even the "it broke for me" posts brought in traffic.

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

Differentiation comes down to three things: (a) it covers the full chain from topic selection to review, rather than only distribution or only generation; (b) each account gets a six-dimension profile (positioning, style, audience, platform, preferences and red lines, long-term memory), and publishing data feeds back into it; (c) it targets Chinese platforms, with many skills, and they're runnable scripts.

These differentiators are **selling points that spread** , but not verified use value. "The profile gets more accurate the more you use it" has no public long-term ROI measurement to date; real user feedback is "somewhat useful but not by much," with high token costs and hard setup. So: the differentiators make it **easy to share** , which is why it took off; whether it can **keep people** is unproven. The steady 30–50 stars a day mainly come from search and aggregator long-tail traffic.

### 6. Visibility and copycats

- **能见度** :中文 AI 自媒体圈已经传遍了(GitHubDaily、新智元、38 万粉微博号、小红书十几篇),但没有进主流科技媒体(36 氪、钛媒体没写),英文圈只有零星 LinkedIn 和聚合站。不涉及收入。

- **Copycats** : found no repo directly replicating Easel. Several similar-direction skill packs appeared around the same time, such as cv-cat/catbus (09-27,575 stars, one command to operate Xiaohongshu, Douyin, and other platforms), taishang-laojun (10-22,495 stars), BullFlow (10-14,335 stars), and AttentionOS (10-15,247 stars). This shows the category is heating up, but none of them named Easel as a benchmark.

### 7. What to borrow

- When a university lab or team releases applied open source, the combo of "research-narrative press release (New Intelligence, Machine Heart), a Chinese open-source recommendation post (GitHubDaily), and Xiaohongshu/Zhihu hands-on posts" can reach 2,000+ stars in two weeks.

- Write the pitch as a user's day, not a feature list; put real outputs and risk warnings in the README.

- Bet on the hottest current agent runtime (OpenClaw) and build a vertical bundle on top of it, rather than building your own core.

- A tool for content creators is naturally shareable.

### 8. Uncertain

- Daily star counts for September can't be obtained (the stargazer API 404), so we can't precisely match how many stars each post brought in; how many came from GitHubDaily (9-08) and New Intelligence (9-16) can only be roughly estimated.

- The clustering of Xiaohongshu notes from 9-25 to 9-30 — whether it was team-organized — is unverified.

- How many real active users there are and how retention looks is unknown; the 735 forks and 65 issues/PRs are the only usage signals.

- The REAL Lab lead and team background were not dug into.

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