Stringers
supercli-site/awesome-coding-plan · supercli-site

A 200-day-old AI plan-comparison repo with 8 stars: everyone forking it signed up last month

What it is An open-source comparison table of AI coding plans and API prices: 113 plans and 51 API price quotes, each with an official link and a verification date, plus a comparison site priced in RMB.

Why it blew up It didn't, really. The star curve was most likely pushed up by the author using three of their own repos to boost each other. They also rode the name of a 2.9k-star top repo, awesome-coding-plan, to climb GitHub search and topic rankings and drive traffic to their comparison site.

DHow you could ride this Don't build another AI coding plan comparison. The real stars in this niche were taken by 3 to May. You could take its "every entry has a source and a verification date, and publishing is blocked when data goes stale" approach to a category where prices are changing and nobody has organized them yet, and post in the first few weeks while the topic is heating up.

D grade: its stars mostly aren't from real people, and there's no growth method here worth copying. The niche is already held by mahonzhan (2.9k), wmpeng (1.3k), and others, and the topic has passed its peak. What's worth learning is the data-credibility approach and how to spot inflated stars. Few people know about it yet
Data (updated daily) · Repo created 2026-11-03 · 205 stars · 7 days +1
GitHub · stars
010020011-0411-0811-1211-1611-20205
Data 2026-11-04 → 2026-11-20 (17 days, 17 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

TokenPlan (on GitHub as supercli-site/awesome-coding-plan) is an open-source dataset that compiles AI coding plans, API unit prices, and agent task costs from China and abroad. Each entry is tagged with its official source and verification date, and it comes with a price-comparison site. It went from 8 to 202 stars in a short time, which looks like a new project catching fire. Digging deeper, almost all the accounts that forked it are empty shells registered in October, and two other repos under the same org gained stars the same week. No real person outside the site seems to be talking about it. The data itself is carefully done, but those 200 stars most likely didn't come from the data (inference). If you want to build an open-source project that earns stars, there's nothing here to copy. It's worth a look because it shows how to spot bought stars, and who has already taken the real stars in this niche.

The four forks as of 21:8 PM, Nov 3

On Nov 3 2026 17:39 (Beijing time) the repo was created, and 18:12 pushed the first commit: "feat: TokenPlan comparison site, plans, API unit prices, and model value-for-money". Within seconds, over the next two-plus hours, 20:21:49 to 20:22:18,29 forks came in 4 times.

Of these 4 accounts, every one now returns an error (GitHub returns 404, meaning they were banned or deleted). By 11-04 at 01:22 AM, 3 more showed up, and 2 of those are already 404. After that it's roughly one batch a day, each batch bunched into a minute or two, and one batch of 3 accounts was created just 10 seconds apart (GitHub API, 11-11).

The accounts still around look like this:

  • uibessredwood, bio reads "Max Ivanov, Software engineer @JetBrains". Registered 10-26, 0 followers, 5 repos: a forked TypeScript project, this comparison repo, plus empty "adjective-noun" repos like fond-shaper-lite and pure-yarn-tool.
  • guo-kaihs, bio reads "Ryan Qi, Tech lead @calcom". Registered 10-26, 0 followers.
  • buildsbrigettetgz, bio reads "Rune Ye, Frontend developer, Singapore". Registered 10-25, 0 followers.

The username, name, and bio don't match each other, and all the registration dates cluster between 10-25 and 11-02. The 14 I spot-checked all looked like this.

What it is

The author is the GitHub organization supercli-site, created 11-01, with a bio reading "SuperCLI tool matrix: CLI directory and discovery site · project quality Scorecard · beginner tutorials, newbie dashboard". Nearly all commits come from webkubor, a developer in Hangzhou with 29 followers; the rest are daily automatic syncs from GitHub Actions.

The repo holds four tables: domestic plans, overseas plans, API unit prices, and a value-for-money ranking. In total: 113 plan entries, 51 API price quotes, and 137 model tiers, each with an official link and verification date, released under CC-BY-4.0. API prices list cache-hit pricing and peak/off-peak rates. The value-for-money ranking combines public agent benchmark scores with per-task cost. Prices are converted to RMB, and the homepage says "Exchange rate baseline 2026-11-10 · 1 USD ≈ ¥6.7167 · Quotes are each dated by verification date".

The companion site plan.supercli.site opens with a faux terminal line, $ tokenplan compare --all, followed by "TokenPlan official picks": DeepSeek V4 Flash for pay-per-call usage, MiniMax M plans for personal agents, Zhipu GLM for domestic coding, and "Claude → Codex → Grok" for overseas.

Engineering-wise it's built like a data product: data validation, alerts when a snapshot is 7 days old, publishing blocked at 30 days, README generated from the data, and a separate document for manual review. The entries already cover the current generation (GLM-5.3, Kimi K3), making it the newest of its kind.

How the 200 stars came in

Datestarfork
11-04211
11-06568
11-0710213
11-0919015
11-1120215

Source: GitHub daily star snapshots; 11-11 is the GitHub API. The biggest single-day gain was 11-09, +88.

The fork accounts are covered above. The star accounts can't be checked directly this time: GitHub's stargazers endpoint returns 404 for this repo, so the list isn't available. The claim that the stars were inflated is an inference, based on three points:

First, forks. Nearly all the visible fork accounts are the same batch of empty accounts, arriving in clusters within seconds. A repo only 3 hours old with no public awareness got its first forks from 4 accounts registered less than two weeks earlier.

Second, sibling repos are rising together. xiaobai-kanban from the same org (a Chinese AI-coding beginner tutorial) only hit 21 stars in mid-October. 11-04 it jumped to 120 in 23,11-06, and now it's at 286. Its forks since 11-02 also come from accounts like pmmichaelcooper and devmaryannalattimore. Another repo, scorecard, 9 in October, and its forks come from achun9501, long90019501, and longlong9501, all with the suffix "9501", which looks like the author's own alt accounts. Since 11-03, the same batch of empty accounts has been showing up. Three repos under one org rising the same week doesn't look like one article driving one repo.

Third, it's quiet outside GitHub. Searches on X for supercli, awesome-coding-plan, and TokenPlan over the past ~30 days turned up nothing. linux.do, V2EX, and Juejin have no posts introducing it either. The only hits are auto-indexing pages like trendshift and thedailyrepo that list new GitHub repos. It hasn't appeared on GitHub Trending, and plan.supercli.site doesn't show up in the global top 1M websites.

A project people genuinely like usually has a post, a tweet, or a repost behind its star curve. Nothing here matches.

So what's the payoff?

Not the stars themselves, but the positions the stars can buy (inference).

The name was borrowed. The repo is called awesome-coding-plan, identical to mahonzhan/awesome-coding-plan, which took off in March and now has 2,929 stars. The license is also CC-BY-4.0. Searching that term on GitHub or a search engine brings up both.

Stars buy ranking. GitHub search and topic pages both weigh stars. It already ranks #3 under the coding-plan topic, and aggregator sites like trendshift index it automatically. Stars cost a few cents to a few tenths of a yuan each, so the math works.

Rankings buy site traffic. Commit 11-09 is titled "fix(site): render crawlable pricing snapshot", specifically so search crawlers can read the prices. The repo carries 12 topics, and CI auto-commits once a day to keep showing "recently updated". The README's first screen links to the comparison site. This is part of the same play as the "tool matrix" in the organization bio: CLI directory, project quality checks, beginner tutorials, and the comparison site, all feeding traffic to each other.

Where the real stars went

ProjectScaleHow it differs from TokenPlan
mahonzhan/awesome-coding-plan2,929 stars, created 3-30, growing 0~4 a day since OctoberA same-named original, focused on quota multipliers and subjective buying advice; its data stops at the previous model generation, and it doesn't cover API unit prices or agent rankings
wmpeng/codingplan (with web page)1,263 stars, built 3-06, +3~8 per daycovers 31 platforms in China, has a review web page; the curve looks fairly natural
anomalyco/models.dev7,597 star / 2,028 forkonly tracks API prices and model metadata; many tools use it directly as a data source; doesn't cover subscription plans
codingplan.fyiWeb, launched around Junesame idea: uses agent benchmarks to compute task cost and rank value-for-money, nearly identical approach
coding.mcppla.net, codingplan.org, codepick.dev, etc.Webin China, ranks by requests per yuan; overseas, English price comparisons and SEO articles
CreditsPlanweb + open source, 0 GitHub starsalso an indie-dev plan comparison site that never took off
Zhihu, Juejin, Bilibili reviewsposted monthly since March, a few thousand to tens of thousands of reads per postone-off content that goes stale

The real stars in this niche clustered from 2026 3 to May: every vendor's Coding Plan was getting launched and sold out, and whoever made a comparison table then picked up thousands of stars. By October, the leader mahonzhan was only gaining 0 to 4 per day; the topic had already peaked.

TokenPlan is newer and more complete than these: China plans, overseas subscriptions, API unit prices including cached and peak/off-peak pricing, and agent task costs, all in one place, which none of the others have. But these strengths haven't earned real human attention; what's growing is something else.

One more small detail: the daily auto-commit message reads "ranking snapshot 2026-11-04", and from 11-04 to 11-10 the date stayed the same, only the exchange rate changed. So the ranking table hasn't actually been refreshed this week, and the author 11-11 has just set the staleness rule to alert after 7 days and block after 30 days. "Updated daily" is only true right now for prices and exchange rates.

If it were you

One-line verdict: Don't build the Nth AI coding plan comparison repo, and don't copy its star-chasing. The real stars in this niche were taken between 3 and May. Building it now, the best outcome is what TokenPlan has: excellent data, and nobody comes.

Who it suits: people who want to build an open-source data project around "fast-changing prices you need to compare before buying." Its approach is worth copying: every data point carries an official link and a verification date, goes on alert when stale and blocks publishing, the README is generated from the data, and PRs that correct prices are accepted. Apply it to a category nobody has organized yet that's actively changing, such as cloud GPU rental prices, quotas and rate limits across agent tools, or a new category just starting a price war.

Who it doesn't suit: people hoping to win stars with a roundup tied to a hot topic. These repos only gain stars during the hottest few weeks; once the leader holds the spot, latecomers can only follow with similar names and free-ride, and real people won't click a second time.

How to do it:

  1. Pick a category where prices or rules change every week, where people keep asking in groups "which one is worth it right now," and where GitHub has no decent roundup yet.
  2. First make the data credible: a source link and verification date on every entry, with stale data flagged automatically. This is what TokenPlan does best, and it's the key to readers trusting you.
  3. Ship a v1 that is a directly viewable web page, not just a README table.

How to get users: post where the topic is already being discussed: under the threads on linux.do, V2EX, and X that kicked off the comparison discussions, with a comparison table people can screenshot and repost. Post an update the day a vendor changes prices, limits, or sells out, so "what changed" becomes a reason to come back each time. TokenPlan did none of this, and it's the thing it's really missing.

When to get in: for a roundup-style project, the window is the first few weeks of rising interest. The AI coding plan topic is already more than half a year old; for a new niche, you need to move before its first "awesome" list appears.

What would show I'm wrong: if TokenPlan gets passed around by real people on X or linux.do afterward, or its stars grow steadily day over day without being pushed, that would show "freshest and most complete data" alone can earn attention in a past-peak niche, and the niche isn't closed. If the author posts a real promotion thread that lines up with the two jumps on 11-06 and 11-09, then the "pushed up artificially" inference falls apart.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
mahonzhan/awesome-coding-planOpen source2026-03-302,928 stars / 41 forks, after October +0~4 per daysame-name predecessor, focused on quota multipliers plus subjective buying advice, data stuck on the previous model generation; doesn't do API unit prices or agent rankings
wmpeng/codingplan (includes GitHub Pages site)Open source2026-03-061,263 stars, +3~8 per day31 China platforms plus review web pages; TokenPlan adds overseas tool subscriptions, cached and peak/off-peak pricing, and daily automatic sync
anomalyco/models.devOpen source2025-067,597 star / 2,028 forkonly tracks API prices and model metadata, doesn't cover subscription plans
codingplan.fyiWebAbout 2026-06Not verifiedsame idea: uses AA benchmarks to compute task cost for value rankings, nearly identical approach
coding.mcppla.netWeb2026Not verifiedonly covers 22 China vendors, ranks by requests per yuan
codingplan.org / BenchLM / RankLLMs / codepick.devWeb2026-06~10Not verifiedEnglish overseas-oriented price comparisons and SEO pages; little coverage of Chinese cloud vendors
CreditsPlan(creditsplan.cn)Web2026-070 GitHub stars, Juejin article with 66 readsalso an indie-dev AI plan comparison and buying-guide site that never took off
Coding Plan review articles and videos on Zhihu, Juejin, and BilibiliOtherposted monthly since 2026-03a few thousand to tens of thousands of reads per articleone-off content that goes stale; TokenPlan updates daily and accepts PRs
Where users come from
  • Inference star/fork farming: 21 forks arrived in batches, within seconds of each other (e.g. 11-06 16:55:56–16:56:06 brought 3); spot-checked 14 accounts, all registered 10-25~11-02 with 0 followers, profile and login names don't match, 6 already 404; sibling repos xiaobai-kanban and scorecard got the same farm-account forks in the same period
  • Confirmed borrowing the top repo's name: the repo name awesome-coding-plan and CC-BY-4.0 license both match mahonzhan/awesome-coding-plan (2.9k stars); searching the term surfaces both together
  • Confirmed SEO/GEO and cross-linking the tool matrix: 11-09 submitted 'render crawlable pricing snapshot'; 12 topics; daily automated CI commits; the org profile states the CLI directory site, Scorecard, and the beginner kanban cross-link each other
  • Inference community and KOL promotion: x-radar has 0 posts; nothing found on linux.do, V2EX, or Juejin

Why supercli-site/awesome-coding-plan (the TokenPlan comparison site) didn't actually take off

Bottom line: it never really took off. The 8-day 200-star curve looks like one author (webkubor in Hangzhou, who runs the "SuperCLI tool matrix") boosting three of his own repos at once: nearly every fork account you can trace is a farm account registered in late October, and all three repos gained together, week after week. Almost no real people discuss it off-GitHub. The product itself is a carefully made AI coding plan / API price comparison dataset, and its data is the newest in its category. But the niche has been crowded since 2026 March (two GitHub repos with 2.9k and 1.3k stars, plus a pile of comparison sites and posts), so its stars didn't come from differentiation. The play: ride the same name as the top repo's awesome-coding-plan (2.9k stars), inflate GitHub heat with bought stars, then funnel traffic to plan.supercli.site for SEO.

1. What it is

  • Repo created 2026-11-03 09:39, by the organization supercli-site (org created 2026-11-01; its bio reads "SuperCLI tool matrix: CLI directory and discovery site · project quality Scorecard · beginner tutorial kanban"). Of 19 commits, 15 came from webkubor (bio "turning taste into verifiable engineering", Hangzhou, 29 followers), and the other 4 are daily syncs from github-actions. (Source: GitHub API, 2026-11-10/11)
  • Content: four tables covering China plans, overseas plans, API unit prices, and a value-for-money ranking. In total 113 plans, 51 API price quotes, and 137 model tiers; every entry has an official link and a verifiedAt verification date, under CC-BY-4.0. The ranking data is pulled once daily from ModelGo's public agent leaderboard (Artificial Analysis's Agentic Index plus per-task cost), with prices checked by hand. The companion site plan.supercli.site is hosted on Cloudflare Pages, with prices converted to RMB (1 USD≈¥6.7167), and the homepage has two blocks: "TokenPlan official recommendations" and "start from the cloud vendors' official entry points". (README and site copy, read via Jina, 2026-11-10)
  • Monetization: the "go directly" links on the site and in the README all point to official pricing pages, no invite codes or referral parameters seen. However, the copy keeps emphasizing "official cloud platform entry points" and "purchase navigation," so leaving room for cloud-vendor referral commissions later is plausible (speculation).

II. Growth data and whether it's real

Datestarfork
11-04211
11-06568
11-0710213
11-0919015
11-1120115

(github repo_day / data card; max single-day +88, on 11-09)

Fork accounts are farm accounts (GitHub API, 2026-11-10):

  • Listed in time order, 21 forks, arriving in batches, with batch members seconds apart: 11-03 12:21:49–12:22:18 brought 4 (less than 3 hours after creation), 11-03 17:22 brought 3, 11-04 22:27:01/02 brought 2, 11-05 19:36–19:41 brought 3, 11-06 16:55:56–16:56:06 brought 3, 11-07 17:42–17:43 brought 2.
  • The 14 accounts spot-checked all registered between 2026-10-25 and 11-02, with 0 followers and 0 following, each with 4–6 repos. The repos follow one pattern: a big forked project (flink, mkcert, TypeScript, llvm-project, starship) plus an empty "adjective-noun-lite" repo.
  • Login names and profiles don't match at all: for example buildsbrigettetgz's profile reads "Rune Ye, Frontend developer, Singapore", uibessredwood reads "Software engineer @JetBrains", and guo-kaihs reads "Tech lead @calcom".
  • Another 6 fork accounts (bernettacarlingayb, werner-althamrfg, jamedavisonvvk, beaconjeraldatu, adinaetchellsyun, jiang-gangumae) have since been 404, banned or deleted.
  • GitHub's stargazers endpoint returned no list this time, through both REST and GraphQL (REST 404, GraphQL returned 0), so the star accounts couldn't be checked directly. That the stars were bought is an inference from the fork patterns and the sibling repos below.

Two other repos in the same org rose together in the same period, same pattern:

  • xiaobai-kanban (Chinese AI coding beginner tutorial): only 10-17 21 stars; 11-04 went 23,11-06 to 120,11-11 to 284 (GitHub repo). 28 of its 11-02 forks came after 26, and the account names look like pmmichaelcooper, devmaryannalattimore, jacquelinappleyard26.
  • scorecard: forks from 9 to October came from achun9501, long90019501, xiadsadaso9501, longlong9501, achun2200, all with the same "9501" suffix, looking like the author's own alt accounts; from 11-03 onward, farm accounts appear too.
  • In other words, within one week the whole org was pushed up together, not one piece of content making one repo go viral. The data card's claim that fork/star counts being higher than 15/201 "shows people really want to take the data" doesn't hold: those forks are exactly what was manufactured.

Almost nobody outside GitHub mentions it:

  • X: x-radar searched supercli / awesome-coding-plan / TokenPlan / coding plan price comparisons. 30 posts in the last few days, all 0.
  • Chinese communities: Brave and Jina searched linux.do, V2EX, and Juejin. No posts introducing this project turned up. The only hits were auto-indexing pages like trendshift.io/repositories/341064 and thedailyrepo.com.
  • Never charted on GitHub Trending (data card trending_days_any=0). The database has under 60 days of history, so it couldn't have made it in anyway.
  • supercli.site is not in the Tranco top 1M (traffic database).

3. Competitors (a crowded niche)

NameWhereLaunchedScaleHow it differs
mahonzhan/awesome-coding-planOpen source + GitHub Pages2026-03-302,928 stars / 41 forks. Basically flat since October (about +0 to +4 a day).Same name, same CC-BY-4.0 license. It's clearly the prototype this one copied. mahonzhan focuses on "quota multipliers" plus subjective buying advice. Its data is less current (it still lists kimi-k2.6 and GLM 5.1). It doesn't do API unit pricing or agent rankings.
wmpeng/codingplan + wmpeng.github.io/codingplanOpen source + web2026-03-061,263 stars, growing about +3 to +8 a day. Fairly organic.Covers 31 domestic platforms and has review pages. TokenPlan adds overseas tool subscriptions, cached API prices, and peak/off-peak prices.
anomalyco/models.devOpen source2025-067,597 star / 2,028 forkCovers only API prices and model metadata, and many tools simply pull from it. Does not cover subscription plans.
codingplan.fyiWebAbout 2026-06Not verifiedUses AA benchmarks to compute per-task costs for 22 models. The idea is nearly the same as TokenPlan's "value-for-money ranking."
coding.mcppla.netWeb2026Not verifiedA ranking of "requests per yuan" across 22 domestic platforms.
codingplan.org, BenchLM /coding-plans, RankLLMs, scriptbyai, codepick.dev, the LLM Gateway blogWeb page (English)2026-06~10Not verifiedOverseas plan comparisons and SEO articles.
CreditsPlan(creditsplan.cn / .ai)Web + open source2026-070 GitHub stars, Juejin article with 66 readsAnother "AI plan comparison + buying decisions" site built by an indie developer. It didn't take off either.
A "Coding Plan comparison tool" on coze.site.Web202637 platforms / 99 plansA similar page built on Coze.
Reviews and videos on Zhihu, Juejin, Cnblogs, CSDN, and Bilibili.Itemposted monthly since 2026-03a few thousand to tens of thousands of reads per articleOne-off content that goes stale. TokenPlan's edge is that it updates automatically every day and accepts PRs.
Official Token Plan overview pages from vendors (Tencent Cloud, Bailian, etc.).Big-company features2026—Each vendor only covers its own products.

4. Growth tactics breakdown

  1. Starring/forking to inflate counts (likely, high confidence): evidence is in the previous section. The cost is very low, a star runs a few cents to a few tenths of a yuan. The payoff is GitHub search ranking, topic page ranking (it's already #3 on topic:coding-plan), inclusion in aggregators like trendshift, and backlinks and clicks from the "price comparison site" link on the README's first screen.
  2. Piggybacking on the top repo's name: it's called awesome-coding-plan, the same name as mahonzhan's repo with 2.9k stars, under the same license. Search that term on GitHub or Google and both repos show up together.
  3. SEO/GEO groundwork: 11-09 has a commit "fix(site): render crawlable pricing snapshot" that specifically lets crawlers read prices. Topics add 12 keywords. CI commits automatically every day to keep "recently updated" current. This follows the same playbook as the organization's other products (CLI directory site, Scorecard, beginner tutorials): a toolkit where each piece feeds traffic to the others.
  4. No evidence of real-person promotion. The author's promotional posts aren't on X, linux.do, V2EX, or Juejin.

5. Differentiation

  • Real strengths: (1) The data is current. 11-10 verifies each entry, and it covers the latest tiers like GLM-5.3, Kimi K3, and DeepSeek V4.1 Flash, while the top repo is still on the previous generation. (2) The scope is the broadest: domestic cloud vendor plans, overseas tool subscriptions, API unit prices (with cache and peak/off-peak pricing), and per-task agent costs, all in one. (3) It's engineered like a data product, with a pnpm check data gate, an alert at 7 days stale that blocks at 30 days, a README rendered from the data, and a FACTCHECK process.
  • But these aren't why it got its stars. Real engagement never grew. What grew was manufactured volume. Meanwhile, codingplan.fyi, wmpeng, and mahonzhan already own the "price comparison plus value-for-money" mindshare. The top repo, mahonzhan, itself only gains 0–4 stars a day since October, which suggests this topic's GitHub traffic has passed its peak. The peak was during 2026's 3–May rush, when vendors' Coding Plans sold out.

6. What to borrow

  • Making data credibility the selling point, with official links for each price tier, a verification date, and a staleness gate, can be applied to any high-ticket purchasing category where prices change fast: cloud GPUs, SaaS subscriptions, VPNs, overseas payment channels.
  • Breaking down the combo: piggyback on a top repo's name, inflate stars to climb topic rankings, then use the README to funnel traffic to your own domain. This is common in the Chinese developer tools scene. When you see a new repo with a high fork-to-star ratio, check the fork accounts before drawing any conclusions.
  • For the price directory to make money, it would need affiliate commissions from cloud vendor promotions (Bailian, Volcano Engine, and Tencent Cloud all offer them). It hasn't integrated any yet.

VII. Uncertain

  • The starring accounts can't be looked up directly (stargazers API 404). That the stars were bought is an inference. The evidence is that the forking accounts are all farm accounts, and sibling repos show the same pattern during the same period.
  • The actual traffic to plan.supercli.site is unknown (no Tranco rank, and no CrUX data found).
  • Whether it was promoted in private channels like WeChat groups, QQ groups, or WeChat Official Accounts can't be determined.
  • Whether affiliate links will be added later is unknown.

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