One commit, 7,759 stars: replica-skill didn't take off because of its code. It took off because of one line: "Comment CLONE and I'll DM you the link"

replica-skill is a Claude skill chained from 11 steps, pitched as "clone any app, then fix the things its users complain about, and turn it into a product you can sell." It was created in Oct 3, has only one commit, and hasn't been touched since. A month later it had 7,759 stars and 1,384 forks (GitHub, 11-06). This piece covers what it is, where the stars came from, why copying the approach probably won't work for you, and how to build something similar if you still want to.

A Reel with nothing but "🙌" under it

In Nov 2, an Indian AI creator named Ritik Gupta posted a Reel on Instagram. The caption's first line: "Save this for your next side project 🔖". Second line: "Comment \"CLONE\" and I'll DM you the link 👇". Below that: "Claude can now rebuild almost any app for free, then make it better 🤯". The comments are row after row of "🙌" and "Clone", each one someone who wants the link.

He isn't the author. He's just another repost account reading the README aloud. In Oct 30, a Facebook page under the name Thomas Smith posted nearly the same video: "Claude can now rebuild the apps you pay for. For free". By early November it had 5.7K views, and the comments were the same row of "Clone". Going further back, similar accounts did the same thing in Oct 9, Oct 11, and Oct 29.

This is how the repo's stars were collected, one by one.

What it is

A Claude Skill that runs 11 commands in order. First it reads the public pages, help center, store listing, and screenshots of the app you want to clone, then maps the pages and flows (recon). Next it picks a tech stack and database, extracts the design system, rebuilds one screen at a time, adds login and payments, clicks through every flow looking for bugs, and scores fidelity against the original. Then comes the step that gets reposted most, "Entrepreneur": it reads real negative reviews of the original app from the App Store, G2, and Reddit, tells you what to fix and where, and finally names the new product, writes the landing page and store copy, and deploys to your own domain.

On top of that, 6 small scripts that use only the Python standard library and make no network calls: screenshot comparison, fidelity scoring, review classification (reviews without links get dropped), and a scan for leftover original-product names in the code. The README spends a lot of space on boundaries: rebuild only features and flows, don't touch the original's code, logo, or copy, don't log into anyone else's accounts, and rename before launch.

Installation is one sentence: paste the repo link into Claude and write "install skill".

How the stars grew

(Star counts come from daily records starting at 10-17; data before 10-09 is coarser.)

This curve is not a one-off spike. It's a steady +150 to +280 every day for a month. Spikes usually come from a Hacker News post or a big account's tweet. A steady platform means a machine keeps supplying traffic: one account posts today, another posts tomorrow, and every video has hundreds of people in the comments asking for the link.

Forks are 18% of stars, higher than the 11% typical for new repos over the same period. (Inference: the "paste the link into Claude to install" method and people who want to modify the code would generate forks.) The curve is smooth, the fork ratio is high, and it matches the visible reposts. I saw no signs of star inflation, but I don't have per-star detail, so I can't rule it out completely.

Whose machine is it

The author, Jake Schincariol, goes by @opusjake on Instagram, with 138k followers, and his bio reads "I build Skills & Agents to 10x your life". He doesn't run one repo, he runs a pipeline: the same template, "N free Claude skills, MIT, no signup, no key", released every few days on a new theme. Nearly twenty repos under the same account look almost identical:

Combined, the account has 32k stars. The few that took off passed 4,000; the newer ones are still waiting for traffic. Every Reel runs the same funnel: keyword in comments → DM with link → link to GitHub. The README and every article link back to his site, opusjake.ai.

The money isn't in the repos. The site sells a $20/month Skool community (267 members, per the Skool pages in search results) and custom agent development starting at "$10–15k". The repo is a lead magnet, not the product.

Why would repost accounts help distribute it? Because the README is already a short-video script you can read straight through: a one-line title, 11 steps that are easy to read aloud, one memorable hook ("it reads user complaints and helps you fix them"), a "stop paying for subscriptions" line, and zero-friction install. Repost accounts grow followers and run comment-for-link, and the author gets stars and backlinks. Both sides profit.

Who is actually using it

The issue tracker shows who is clicking through. Among the issues up to #21, titles include "Building a ride-hailing app in Yemen" (Arabic, 11-03), "games" (Arabic), "Instagram", "Klon", "Crazy fox", and "hello friend". Most are non-coders treating the issue tracker as a wishlist. Only two pieces of technical feedback: the screenshot-comparison script throws an error on a bad PNG, and in Oct 30 someone reported a plugin install failure ("Path escapes plugin directory"). As of 11-06, the author hasn't fixed it, and there's been no second commit.

In other words, no one has publicly shared an app built and sold with it. The stars are buying the thought "I could clone an app and sell it too," not endorsement from people who used it.

Who it's competing with

The "clone an app with AI" topic was already being hyped half a year ago:

replica has no technical moat. The skill is a few SKILL.md files that can be copied in a day. What it occupies is a narrative slot: translating REA's "hacker reverse engineering" into "clone a product you can sell," which ordinary people can follow. Even that slot is already being recycled. A site called AGI Point used replica to make a "12-page free guide" to collect emails, then sold its own products.

If it were you

The verdict: there's a lot to learn, but copying it probably won't get you this many stars. Its stars came from a 138k-follower account plus a network of repost accounts willing to read the script aloud, not from the code. The topic is also already crowded with REA, bama, and a pile of derivative repos.

Who it suits: people who already have a short-video or social audience (tens of thousands of followers or more) and want to use an open-source repo to bring in traffic, or people with real development experience in a specific ecosystem (Flutter, WeChat mini programs, native Android) who can make "clone + read reviews" actually work.

Who it doesn't suit: people with no distribution channel who are hoping a good topic alone will land them on GitHub Trending. replica doesn't appear to have hit GitHub Trending (inference); its stars were mostly pushed in from off-platform. Without that off-platform machine, the same README would most likely get a few dozen stars.

How to do it: don't build the Nth "clone any app". Do the half it didn't do well: actually run one clone end to end, and put the whole process from a real app to a launched clone in the repo along with the outputs (it still has no public finished case study). Or narrow it to one ecosystem or one category of app (say, "clone any budgeting app as an Android version") so programmers also want to star it. Follow its README format: a one-line title, numbered steps, one memorable hook, a one-line install, and clear boundaries.

How to get users: copy its funnel, but borrow other people's accounts. Prepare copy you can read straight through and a 20-second screen recording, then send it to the accounts on Instagram and YouTube Shorts that do "AI tool recommendations". They need fresh material, and comment-for-link also grows their followers. Then submit to skill directory sites that index automatically (skills.sh, skillsllm, and similar) and communities like r/ClaudeAI.

When to do it: "Claude Skill collection + comment-for-link" has become a standard playbook among AI tool creators. The several repos the author released in November only have a few dozen stars, which suggests audience fatigue is setting in (inference). If you're going to do this, do it in the next month or two.

What would show you're wrong: if the author's next few new repos quickly pass a few thousand stars, the machine is still accelerating, the topic doesn't matter, and distribution is everything, which leaves followers even less room. Conversely, if someone publicly builds a "really cloned an app and launched it" case study and gets a lot of stars, it means the demand on the programmer side is real and worth going deeper on.