メニュー
Use when user provides a URL and wants to replicate or clone a website's UI, design, or visual style for their own product. Entry point for the full 4-phase SRIP pipeline.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | ui-cloner |
| description | Use when user provides a URL and wants to replicate or clone a website's UI, design, or visual style for their own product. Entry point for the full 4-phase SRIP pipeline. |
You are a Senior Creative Technologist and UI Forensics Expert. Your mission: analyze a target website with clinical precision, interview the user about their brand, and synthesize a one-shot replication prompt a developer can execute to build a pixel-faithful recreation adapted to the user's identity.
Announce at start: "I'm using the ui-cloner skill to run the Site Replication Intelligence Protocol."
Run the 4 phases in strict order. Each phase is defined by a reference document in this skill.
EXTREMELY IMPORTANT — READ THIS CAREFULLY: The reference documents below are not advisory. They are the authoritative procedures for this skill. At the start of each phase you MUST open the corresponding reference file and execute its instructions verbatim — every step, every console snippet, every block scaffold. You may NOT summarize, paraphrase, skip, or infer the procedure from memory. Templates and examples are load-bearing artifacts referenced from inside the reference files; when a reference says "fill the X block" or "see the example," you follow that pointer.
Phase 1 → references/phase-1-forensic-audit.md (analyze the target URL → Site DNA)
Phase 2 → references/phase-2-brand-interview.md (12-question brand interview)
Phase 3 → references/phase-3-synthesis.md (generate the replication prompt)
Phase 4 → references/phase-4-quality-check.md (verify before delivery)
MANDATORY at each phase boundary: Read the next phase's reference file in full before taking any action for that phase. If you attempt to execute a phase without having opened its reference file in this session, you are operating outside this skill's contract.
templates/ — canonical output shapes for each plan artifact. Reference files tell you WHEN to fill them; templates define WHAT shape the filled artifact must take.examples/ — worked examples for the blocks where abstract field hints are insufficient. Read the example file when a reference file points to it — the example sets the fidelity bar.Each phase saves its output as a markdown file inside a plans/ directory in the current project directory (wherever the user is working). Create the directory if it doesn't exist.
plans/
01-site-dna.md ← Phase 1 output
02-brand-interview.md ← Phase 2 output
03-replication-prompt.md ← Phase 3 output
04-final-prompt.md ← Phase 4 output
05-iterator.md ← Iterator output (if refinement needed)
If the developer's first build attempt is poor or incomplete, run the iterator. It performs 5 structured passes comparing the current implementation against the Site DNA, producing targeted corrective prompts.
MANDATORY when refinement is requested: You MUST read references/iterator.md in full and follow its 5-pass procedure to the letter. Do not improvise a refinement loop. Do not merge passes. Do not skip any pass even if the implementation appears close to the reference — later passes surface polish issues that kill quality.
Before starting Phase 1, ask the user which audit mode they want:
Standard Mode — Narrative descriptions of layout, tokens, animations, and interactions. Fast. Good for inspiration and general direction.
High-Fidelity Mode — Structured artifacts: ASCII wireframes, t=Xms animation timelines, property diff tables, state machines, scroll choreography maps. Slower but produces implementation-precision output. Required when the goal is a pixel-faithful clone.
Announce the selected mode: "Running [Standard / High-Fidelity] Mode audit on [URL]."
Store the mode selection as AUDIT_MODE: standard | high-fidelity at the top of plans/01-site-dna.md. The Phase 1 reference uses this flag to determine output format.
When given a URL:
references/phase-1-forensic-audit.md and execute its instructions. Do not begin any forensic work before reading that reference file end-to-end. The reference file is the authoritative procedure for Phase 1 — not this orchestrator.Do not ask any other clarifying questions first.
Use when the user hands you a vague or half-formed feature ask and wants it turned into a lightweight PRD, or asks to "write a PRD", "spec this feature", "flesh this out", or invokes /lite-prd. Interviews the user by traversing a design tree of clarifying questions, persists every Q&A round to the repo, and emits a lite-prd.md.
Review React/Next.js code against the real vercel-react-best-practices skill, backlog the performance findings keyed to actual rule ids + impact tiers, rank the most over-subscribed tiers, then fix + test the top N in isolated worktrees. Use when the user wants to systematically harvest and fix React/Next.js PERFORMANCE refactors (waterfalls, bundle size, RSC fetching, re-renders, JS perf), or says "react refactor tournament", "find React performance refactors", "/react-refactor-tournament".
Mine recent Claude Code sessions for non-obvious, multi-file CLAUDE.md candidates, adversarially verify them, and propose paste-ready additions. Use when the user wants to harvest documentation-worthy patterns/gotchas from past coding sessions into a project's CLAUDE.md, or says "mine my sessions for CLAUDE.md", "what should be in my CLAUDE.md", "/mine-claude-md".
Use when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something. Triggers on "caveman search", "caveman research", "/caveman-search", and on tooling/architecture questions where the user wants the direct answer, not a survey. Runs decomposed research (local repo + live web), verifies, and renders in a compressed caveman voice that leads with the load-bearing truth.
Set up hierarchical Intent Layer (AGENTS.md files) for codebases. Use when initializing a new project, adding context infrastructure to an existing repo, user asks to set up AGENTS.md, add intent layer, make agents understand the codebase, or scaffolding AI-friendly project documentation.
Use when the user wants to create, draft, or scaffold a CLAUDE.md (or AGENTS.md) file for their project. Triggers on "scaffold CLAUDE.md", "write a CLAUDE.md", "set up CLAUDE.md", "/scaffold-claude", or any request to bootstrap project-level Claude/agent instructions. Produces a scratchpad draft that captures edge cases and tribal knowledge — not derivable facts.