WebFV
AI coding workflow skills / Field guide

Replica Skill: install the 11-skill pack and understand clean-room rebuilding

Follow the documented plugin or manual setup, map the 11 skills from recon to deployment, and understand clean-room, licensing and evidence boundaries.

By WebFV Editorial · Checked October 28, 2026 · Plugin v1.0.0 · source checked October 28, 2026

What the evidence coversOfficial repository README, plugin manifests, skill tree, recon, branding, deployment definitions and MIT license reviewed. No skill installation, tool execution, application cloning or third-party app inspection performed.

Replica Skill is Jake Schincariol's MIT-licensed pack of 11 Claude skills for planning and independently rebuilding an application's functionality. It is a workflow pack, rather than a hosted app builder or a new coding model. The official source is Jakeschincariol/replica-skill.

We checked source revision 77c9436 on October 28, 2026. The plugin manifest reports version 1.0.0, and the tree contains 11 skill definitions. We reviewed the repository and documentation; we did not install the pack, execute its tools, inspect a target application or build a replica.

What “clone an app” means in this project

The author's README uses “clean-room” to describe studying observable functionality and UX flows, then writing a fresh implementation. It explicitly excludes copying the target's source code, logos, proprietary assets, wording, private APIs or licensed content.

A scheduling flow can be a reference for requirements; its original implementation and branding do not become yours. Likewise, reconstructing a music player's controls does not supply a music catalog or distribution rights. The pack's own instructions are boundaries for its workflow, not proof that every possible project is permitted or that generated work has passed a rights review.

Start with a bounded task: for example, a booking link and rescheduling flow, rather than an entire platform. The recon stage records what was observed, what is inferred and what cannot be included. Treat those distinctions as inputs to your implementation rather than filling unknowns with invented features.

Install as a Claude Code plugin

The marketplace definition registers a plugin named replica-skill. The README documents these commands inside Claude Code:

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

Marketplace registration and plugin installation are separate actions. After installation, use the namespaced skill command, for example:

/replica-skill:replica-recon

Give that skill the target URL and the slice of functionality you want to study. The documentation describes namespacing; we have not verified discovery in a particular Claude Code release. If the command is absent, check installation status and the installed skill list before assuming an application was analyzed.

Manual folders and chat-only use

The README also documents copying the replica-* folders into ~/.claude/skills/ for global discovery, or a project's .claude/skills/ directory for local use. This installation form uses plain commands such as /replica-recon instead of the plugin namespace. Preserve the folders' supporting templates and scripts rather than copying only their names.

For chat-only use, the author suggests pasting one SKILL.md as instructions. That does not supply the repository's Python utilities or turn an arbitrary chat client into a tool-enabled coding environment. Choose one installation route appropriate to your host; do not assume the same invocation works everywhere.

The pack is free to obtain. Your coding host, model usage, hosting, payment services and other infrastructure can still have their own requirements and costs. “No API key” in the pack's introduction does not provide free inference or credentials for external integrations.

The 11 skills and their responsibilities

The names below identify the individual skills. Use the plugin prefix when installed through the marketplace.

Skill Main responsibility
replica-recon Record screens, user flows, components, inferred data and a feature matrix from permitted sources
replica-architect Plan the stack, database schema, API and implementation order
replica-design Define design tokens and component specifications using your own assets
replica-build Implement the mapped screens, states and flows
replica-backend Add authentication, database behavior, payments and integrations
replica-test Derive flow tests and track bugs
replica-diff Compare feature coverage and layout, identifying missing requirements
replica-entrepreneur Turn linked user-review evidence into proposed fixes and positioning
replica-brand Create a distinct name, palette, logo brief and wording; check for source-brand remnants
replica-launch Prepare landing-page, pricing and store-listing material
replica-deploy Review release prerequisites and prepare production deployment

This is the documented order, from recon through deployment, although the author also allows individual skills to be used alone. A row describes an intended responsibility, not an action performed or a result guaranteed by installation.

How the clean-room handoff works

The recon definition writes replica/recon.md, replica/features.csv and reference screenshots under replica/screens/. Later stages read the shared replica/ artifacts.

The recon rules permit public material and the user's own account only where its terms allow the intended research. They prohibit bypassing access controls, bulk scraping, taking source bundles, decompiling binaries or capturing private API calls to reproduce endpoints. Screenshots are study references and are not shipped in the new product. Inferred data structures must be labeled as guesses; licensed catalogs, partner networks and other unavailable assets belong in the exclusions.

The branding stage writes replica/brand.md and replica/brand.json, calls for fresh identity and copy, and checks the codebase for original names, domains and colors. Its name checks are explicitly screening checks, not legal clearance. A clean text sweep cannot establish every permission needed for launch.

The deployment stage requires completed release checks, a clean rebrand and the user's approval before the generated application goes live. Reading or installing this pack is not authorization to purchase services, access another person's account or publish an application.

Supporting tools and evidence limits

The README documents six standard-library Python utilities and Python 3.8 or newer:

  • imgdiff.py compares screenshot edge layouts.
  • parity.py scores the supplied feature matrix.
  • reviews.py summarizes supplied review rows with source links.
  • contrast.py checks design-token contrast.
  • sweep.py finds configured original-brand remnants.
  • listing.py checks store-listing constraints and branding references.

These local helpers do not obtain missing review evidence or prove that every application flow works. A score depends on its inputs: an incomplete feature matrix produces an incomplete comparison. The scheduling-app counts, review sample and scores in the README are an illustration, not a replica we built or measured.

MIT license and third-party material

The repository's MIT license allows reuse, modification and distribution of the skill software and associated documentation, with its copyright and permission notice retained in copies or substantial portions. It also supplies the software without warranty.

That license does not extend to the application you study or to third-party code, names, images, fonts, catalogs and other materials. Keep independent implementation work separate from reference material, use assets you are entitled to use and retain their applicable notices. The useful outcome of this pack is a documented route from permitted observations to your own implementation, not permission to copy everything visible on a screen.