Oct 17, 2026

Clone Any App Clean-Room Style With Eleven Claude Skills

replica-skill is an eleven-skill pipeline that reverse-engineers an app, rebuilds what it does — never what it owns — tests the clone, and ships it on your domain.

#tutorial#ai-agents#engineering

You keep rebuilding the same kind of app from a screenshot and a hunch. replica-skill, a 3k-star MIT collection, turns that into a pipeline: one skill reverse-engineers the app, ten more plan, rebuild, test, score, rebrand, and ship your clean-room version of it.

Why This Skill Matters

Cloning usually means copying artifacts — and that is where the legal risk lives. This collection is explicit about the line it draws: "It rebuilds what an app does, never what it owns." Features and flows, clean-room style; not its code, its logo, its copy, or its content. The README adds that the fine print is at the bottom of its page and that the skills enforce it — one of the six bundled Python tools exists purely to sweep your codebase for leftover traces of the original.

Installation

Paste this into Claude:

https://github.com/Jakeschincariol/replica-skill

install skill

Or install as a Claude Code plugin:

/plugin marketplace add Jakeschincariol/replica-skill
/plugin install replica-skill@replica-skill

Claude Code namespaces plugin skills, so they show up as /replica-skill:replica-recon and so on. To get plain /replica-recon, copy the folders instead: git clone the repo and cp -r replica-skill/replica-* ~/.claude/skills/ (or into your repo's .claude/skills/ for a project-local install). No Claude Code at all? The README says pasting any single SKILL.md at the top of a chat runs it as a mode — you lose the Python tools, but the method works. The tools need Python 3.8 or newer, and nothing to pip install.

Real Workflow: Clone a Scheduling App

The eleven skills run in order, and each reads what the last one wrote in a replica/ folder in your project:

recon -> architect -> design -> build -> backend -> test -> diff -> entrepreneur -> brand -> launch -> deploy

The README's worked example clones a scheduling app — share a link, people book a time:

  1. /replica-recon reads the app's help center, pricing page, and store listing from its URL, walks through your own account, and produces replica/recon.md with the screens, flows, components, and an inferred data model. The partner marketplace gets marked out of scope: that is their network, not a feature.
  2. /replica-architect picks the stack, writes the schema — with a constraint so two guests can never book the same slot — and orders the build.
  3. /replica-design measures screenshots into design tokens; /replica-build rebuilds every screen with its states, ticking off features.csv as it goes.
  4. /replica-test writes Playwright specs from the flow list and logs bugs by severity until no S1 or S2 is open.
  5. /replica-diff scores the clone against the original with parity.py, which weights must/should/could requirements, counts partial as half, and never scores what you left out on purpose or what you added — missing must-haves means not shippable.
  6. /replica-entrepreneur reads real user reviews of the original and turns the top complaints into fixes and a positioning angle for your version.
  7. /replica-brand, /replica-launch, and /replica-deploy name it, write the landing page and store listing, and ship it on your domain — after sweep.py confirms nothing of the original is left in the code.

You can also run any single skill on its own — /replica-entrepreneur on an app you are only thinking about cloning is a cheap way to find out whether you should.

Tips

  • The six Python tools are all standard-library, and none touch the network: imgdiff.py compares screenshots as edge maps so your new colors do not count against you, and reviews.py drops every review quote without a link and flags themes backed by fewer than three reviews or a single source as thin.
  • sweep.py finds the original's name, domain, and colors anywhere in your code — including inside identifiers like OriginalAppEmbed — and blocks the deploy until it is clean.
  • listing.py checks App Store and Google Play limits, ranking claims in the title, and the original's name anywhere in your listing.

When Not to Use This

This is a clean-room workflow, so it will not copy an app's code, logo, copy, or content — if what you actually want is a fork, clone the repo instead. It rebuilds from observable behavior — public pages, screenshots, store listings, and your own account; and the trademark and domain checks the brand step names are still yours to run.


See the leaderboard for more skills.