Jakeschincariol/replica-skill

What it claims

replica-skill is a chain of eleven Claude skills that clone an app's features and flows, then rebrand and ship the result. It does not copy code, logos, copy or content. The skills run in a fixed order:

recon, architect, design, build, backend, test, diff, entrepreneur, brand, launch, deploy.

Each skill reads what the previous one wrote into a replica/ folder in the project. Any skill can also run alone. The author suggests running /replica-entrepreneur alone on an app you are only considering cloning, to judge whether the clone is worth building.

How it works

The README walks through a worked example: cloning a scheduling app.

• Recon reads public pages and the user's own account. It produced recon.md (18 screens, 7 flows, an inferred data model) and a features.csv feature matrix.
• Architect chose Next.js, Postgres, Stripe and Resend. It added a schema constraint that prevents double-booking a time slot.
• Test wrote Playwright specs and logged bugs until no S1 or S2 bug was open.
• Diff scored features 86 and the booking-page layout 91. It listed round-robin scheduling and the website embed as missing.
• Entrepreneur ranked complaints from 140 reviews across the App Store, G2, Capterra and Reddit. It turned them into a fix plan and a positioning angle.

The six Python tools define the gates:

• imgdiff.py compares edge maps of two PNGs, so a new colour scheme does not count against layout parity.
• parity.py weights features as must, should or could, and counts a partial feature as half. It does not score features left out on purpose or features added. Any missing must-have marks the clone "not shippable".
• reviews.py drops reviews that have no link. It marks a theme as thin when it has fewer than three reviews or only one source.
• sweep.py finds the original app's name, domain and colours, including inside identifiers such as OriginalAppEmbed.
• listing.py checks App Store and Google Play character limits, ranking claims in the title, and any mention of the original's name.

flowchart LR
  T[test: no S1/S2] --> P{deploy preflight}
  D[parity score] --> P
  S[sweep clean] --> P
  L[listing lint] --> P
  P -->|all pass + user says go| Ship[ship to own domain]
  P -->|any fail| Block[deploy blocked]

Deploy runs this preflight and ships only after the user confirms.

How to use it

Install as a Claude Code plugin. Skills then appear namespaced, for example /replica-skill:replica-recon.

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

Or copy the folders to get plain names such as /replica-recon:

git clone https://github.com/Jakeschincariol/replica-skill.git
cp -r replica-skill/replica-* ~/.claude/skills/

To install for one project only, copy the folders into the repo's .claude/skills/ instead. Without Claude Code, any single SKILL.md can be pasted into a chat and run as a mode, but the Python tools are then unavailable.

The tools need Python 3.8 or newer. They use only the standard library and never touch the network.

python3 replica-diff/parity.py replica/features.csv
python3 replica-brand/sweep.py . --avoid "Original App"
python3 -m unittest discover -s tests -v

Limits

• Recon reads only public pages and the user's own account. It does not scrape behind logins or bypass paywalls.
• If the user's account terms forbid building a competitor, recon uses public sources only.
• An app's content, network and licences cannot be cloned. A music app's player can be rebuilt; its catalogue cannot.
• The README gives no guarantee of a perfect clone. It estimates a booking tool at weeks of work and says a spreadsheet engine is out of reach.
• Brand skill lists trademark and domain checks but leaves them to the user to run.
• The author states this is not legal advice and recommends consulting a lawyer before selling.