Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    VisionLoop Replicator — Multi-Agent Website Replication System

    1

    Multi-agent system that replicates websites page by page in Next.js: 5 parallel analysts, per-section builders, pixel-diff self-correction to ≥99% desktop / ≥97% mobile.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    VisionLoop Replicator — Multi-Agent Website Replication System

    VisionLoop Replicator — Multi-Agent Website Replication System

    Example session with this skill installed

    • Read your context and instructions
    • Compiled the visionloop replicator
    • Generated the design tokens stylesheet

    ✅ Page completed in 12 iterations

    📊 Final scores (current page: home)
    Desktop: 99.2% · Tablet: 97.8% · Mobile: 97.4% · Lighthouse: 94 / 96 / 92 / 100

    📂 Project in: replica/ — app/page.tsx, components/sections/pages/home/, shared Navbar & Footer, frozen design tokens in lib/design-tokens.ts

    ⚠️ IMPORTANT — before you publish this replica: it is a structural/technical starting point only. It CANNOT be published or deployed publicly as-is. Replace the site in its entirety — all content, texts, images, logos and branding — with your own material first.

    🔗 Next page? Detected in the navbar: /about (8–15 min), /pricing (12–20 min), /blog (10–18 min).

    page.tsx

    CSS · design tokens

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Replicate the structure of any website — page by page, with measured visual fidelity. VisionLoop is a multi-agent system for Claude Code that takes a URL, a full-page screenshot, or raw HTML and rebuilds the page in Next.js 15 + Tailwind + TypeScript, comparing its own output against the reference and correcting itself until the match is proven, not eyeballed.

    ⚠️ IMPORTANT — what this is for (read before buying). A replica produced by this system is a structural/technical starting point: for site migrations, redesigns, rebuilds, and learning how a layout is put together. You may NOT publish or deploy a replica as-is. The original site's content, texts, images, logos and branding belong to its owner; before any public use you must replace the site in its entirety with your own material. This is a condition of the license (see LICENSE.txt in the package), and the skill itself repeats this warning every time it delivers a finished page.

    How it works

    1. Smart crawling. It opens the page with Playwright, detects the navbar and every available page, captures full-page screenshots at desktop / tablet / mobile, downloads reusable assets, and extracts the DOM and computed styles.
    2. Interactivity inspection (the step everyone else skips). Before analyzing anything, it greps the DOM and probes the live page for carousels, accordions, modals, hover states and scroll-triggered animations — then captures every carousel slide and hover delta. A static screenshot never tricks it into building static grids.
    3. 5 analyst agents in parallel. Layout, design, typography, component classification, and motion each dissect the reference into their own analysis file. Design tokens (palette, typography, spacing) are then frozen — no page can drift from them.
    4. Parallel builders. One builder agent per section generates the components; shared Navbar and Footer are built once and reused across every page.
    5. Autonomous comparison loop. 3 comparator agents (desktop / tablet / mobile) run Pixelmatch + SSIM per section crop, and fixer agents apply corrections component by component. The gate is threefold: ≥99% desktop / ≥97% tablet and mobile per-section similarity and a structural-fidelity score ≥ 0.95 (sections, carousels, animations, links) — so a misleading pixel score can't fake success. After 3 stalled iterations, the component is rebuilt from scratch.
    6. 4-validator final check. Visual, functional, interactivity (carousels advance, keyboard nav works, prefers-reduced-motion respected, no orphan scaffold files) and Lighthouse > 90.
    7. Next page, zero drift. It proposes the next page from the navbar and reuses the frozen tokens and shared components, so a 5-page site stays consistent.

    What's included

    • SKILL.md — the complete 8-step multi-agent pipeline (907-line production system).
    • scripts/ — the real production scripts: crawl.ts (per-page crawler), capture.ts, compare.ts (Pixelmatch/SSIM diff reports), validate-interactivity.ts, plus their package.json. Generic defaults — no site-specific data.
    • reference/setup-guide.md — full setup, workflow, troubleshooting and (gated) deploy guide.
    • reference/startup-behavior.md — save as CLAUDE.md for the guided startup greeting.
    • LICENSE.txt — commercial license, including the intellectual-property terms for replicas.

    No example projects and no third-party site assets are included — deliberately. The package contains the system and its tools, nothing copied from anyone's website.

    Who it's for

    • Developers migrating a site to Next.js who need the new build to match the reference exactly.
    • Agencies rebuilding client sites (with the client's own content and rights) that can't afford visual drift between pages.
    • Designers and founders who want a faithful structural replica of a reference layout as the starting point for their own, original site.

    Requirements

    • An agent with SKILL.md support — the full multi-agent pipeline is designed for Claude Code.
    • Node.js 20+ and ~600 MB of free disk space (the skill auto-installs Playwright, Sharp, Pixelmatch and the rest on first run).
    • A target: a public URL, a full-page screenshot, or raw HTML.

    Compatibility note

    Works with any agent that supports the SKILL.md standard (Claude Code, Codex, Cursor and others); the parallel sub-agent pipeline delivers its full fidelity in Claude Code. The generated project is a standard Next.js 15 app — no lock-in to this skill after generation.

    Known limitations

    • Page by page, by design. It replicates one page to ≥99% before moving on; a full site is built page after page (a 5-page site takes ~50–120 minutes of agent time).
    • Output stack is Next.js 15 + Tailwind + TypeScript only. It does not generate WordPress, plain HTML, or other frameworks.
    • It needs to see the page. Login-gated or paywalled pages require you to provide a screenshot or the HTML instead of a URL.
    • Replicas are starting points, not finished sites — see the IP notice above. Replacing the original content and branding with your own is your responsibility and a license condition.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Frontend & Web Apps.

    Page by page, by design: it replicates one page to ≥99% before moving on (a 5-page site takes ~50–120 minutes of agent time). Output stack is Next.js 15 + Tailwind + TypeScript only — no WordPress or plain HTML. Login-gated or paywalled pages require a screenshot or the HTML instead of a URL. Replicas are starting points, not finished sites: replacing the original content and branding with your own is your responsibility and a license condition.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Accounts
    Vercel
    Software
    Node.js 20+Claude Code600 MB free disk space
    Target URL, screenshot,
    orHTML
    Playwright

    The full multi-agent pipeline is designed for Claude Code.

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Browser
    Network Access

    Allowed Hosts

    example.com
    Listed2 days ago

    What's inside

    Frequently Asked Questions

    A complete Next.js 15 project (replica/) with the page's components, the shared Navbar/Footer, frozen design tokens, downloaded reference assets, and a final report with per-breakpoint similarity scores and a visual diff.

    Measured, not claimed: the loop only stops at ≥99% desktop and ≥97% tablet/mobile per-section Pixelmatch similarity plus a structural-fidelity score ≥ 0.95. If a section can't reach the threshold, that's stated in the delivery notes.

    No. That is expressly prohibited by the license and repeated by the skill at every delivery. Replicas are structural starting points for migrations, redesigns and rebuilds — before publishing, you must replace all original content, texts, images and branding with your own.

    It inspects interactivity before analyzing (DOM signatures + live Playwright probing + multi-frame carousel capture), and a dedicated motion analyst maps every detected behavior to its implementation. Reducing a slider to a static grid is an explicit never-break rule.

    Roughly 8–20 minutes of agent time for a homepage, 8–18 for a typical page, 12–25 for a complex multi-section page. A 5-page site: ~50–120 minutes, resumable across sessions.

    Next.js 15 (App Router), React 19, strict TypeScript, Tailwind CSS with the extracted design tokens, Framer Motion, shadcn/ui and Lucide icons. Standard project — deployable to Vercel once the content is yours.

    Yes — give it a full-page screenshot or the page's HTML instead of a URL. Crawling is only one of three input modes.

    Yes. Say "Edit first", change anything in your editor, then continue — later pages reuse your edited shared components and the frozen tokens.

    No, deliberately: page-by-page is what makes ≥99% per page achievable. It detects all pages from the navbar and proposes them in order.

    After 3 stalled iterations the component is rebuilt from scratch; after repeated replans the skill documents exactly which section fell short and why, instead of reporting false success.

    New in Frontend & Web Apps