Communitygithub.com

design-showcase-websites

Design and build distinctive, high-impact websites and clickable, presentation-ready prototypes for any industry, language, or site type from business briefs, public links, local media, existing sites, or product requirements.

What is design-showcase-websites?

This skill guides AI agents (such as Claude Code, Cursor, Codex) through a comprehensive workflow to design and build distinctive, high-impact websites and clickable prototypes. It covers the entire process from concept to a fully verified, runnable prototype, integrating brand identity, conversion-focused design, localized typography (Chinese, English, Japanese, Korean, and multilingual), and rigorous QA. The skill emphasizes deriving a unique visual direction from the subject and audience, maintaining a single source of truth for facts, and creating genuinely interactive prototypes with semantic structure and accessibility. It supports multiple engagement modes (concept, prototype, redesign, polish, web product) and ensures final handoff includes a production build, product and design documents, and a verified quality gate covering specificity, authenticity, conversion, hierarchy, responsiveness, restraint, inclusion, and reliability.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/dingchenhui0618-arch/agent-skills/tree/HEAD/skills/design-showcase-websites

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Design Showcase Websites

Turn sparse briefs and scattered source material into a site that feels specific to its subject, audience, language, and purpose. Support focused landing pages, multi-page websites, and interactive web products. Use this skill as the workflow spine; combine it with any design, browser, image, product-interface, or framework skills the user explicitly names.

Core rules

  • Design a conversion experience, not a collection of attractive sections.
  • Derive the identity from the business, audience, place, product, and real imagery. Do not reuse the visual styling of an earlier project unless the user requests it.
  • Separate verified facts from creative interpretation. Never invent addresses, prices, opening hours, reviews, clients, awards, features, or performance claims.
  • Preserve original user assets. Optimize copies inside the project and document generated or temporary imagery.
  • Choose language from the audience and brief. Never assume Korean, Chinese, English, Japanese, or bilingual content without evidence.
  • Treat localized typography, line breaking, punctuation, reading order, content density, and culturally familiar actions as part of the design.
  • Prefer one strong art direction over a collage of fashionable effects.
  • When the user grants creative freedom, infer and commit to a justified direction instead of asking them to design the site for you.
  • Make the prototype genuinely clickable and demonstrable when implementation is requested. Do not stop at a written concept unless the user asks only for strategy.
  • Keep all essential information visible without animation and honor reduced-motion preferences.

Choose the engagement mode

  1. Concept only: Produce positioning, page flow, visual direction, and key-screen descriptions. Do not mutate files unless asked.
  2. Prototype: Build a responsive, runnable single-page or multi-page site with working primary interactions and presentation-quality content.
  3. Redesign: Audit the existing page and repository first, preserve functional requirements and the current stack, then improve the interface.
  4. Polish: Inspect the live or local page, identify the highest-impact gaps, implement focused changes, and re-verify.
  5. Web product: Design task flows, states, navigation, and data presentation; pair this skill with a product-interface skill when the work is application-heavy.

When no repository exists, choose the simplest maintainable stack that fits the interaction needs. When a repository exists, inspect its conventions and reuse them.

Workflow

1. Frame the outcome

Resolve only decisions that materially change the result:

  • audience and locale;
  • site type, content scale, and whether the experience is single-page, multi-page, or application-like;
  • primary and secondary conversion actions;
  • delivery form and required interactions;
  • source material and usage rights;
  • desired first impression and unacceptable directions;
  • launch constraints such as stack, route, viewport, or port.

Ask at most three compact questions at a time. If the user grants creative freedom and the evidence is sufficient, state reasonable assumptions and proceed.

For sparse briefs, map links, or mixed local assets, read discovery-and-research.md before gathering content.

Read site-types-and-localization.md when choosing information architecture, working in Chinese, English, Japanese, Korean, or building a multilingual experience.

2. Build a source-of-truth ledger

Inspect provided links, map listings, existing sites, files, images, and videos. Record:

  • the verified value;
  • its source and verification date;
  • whether it can be published directly;
  • anything requiring owner confirmation.

Keep operational facts centralized in data or content files instead of scattering them through components. Label placeholders explicitly. If public information conflicts, prefer the business-owned or most recently updated source and flag the conflict.

3. Define the conversion story

Write these before choosing visual effects:

  • the one sentence visitors should remember after ten seconds;
  • the primary CTA and one optional secondary CTA;
  • a belief ladder from first impression to action;
  • three to five proof points already supported by evidence;
  • the minimum information needed to remove hesitation.

Create or update PRODUCT.md in implementation projects with audience, purpose, positioning, conversion, proof, voice, anti-references, and accessibility expectations.

4. Establish a specific art direction

Name one concept that combines category truth with an emotional promise and a deliberate contrast. Examples of useful contrasts include energetic but disciplined, local but editorial, technical but human, or premium but welcoming.

Set design dials before implementation:

  • design variance, 1–10;
  • motion intensity, 1–10;
  • information density, 1–10;
  • color strategy;
  • photography strategy;
  • typography strategy.

Create or update DESIGN.md with concept, tokens, type, layout, image treatment, motion, and responsive behavior. Read direction-and-structure.md when selecting the concept, page architecture, or media treatment.

Avoid automatic visual defaults: generic black-neon fitness styling, generic beige restaurant styling, decorative glass panels, random bento grids, glow without meaning, excessive pills, uniform cards, and repeated fade-up sections.

5. Compose the page around decisions

Choose the information architecture from the site's purpose. For multi-page work, define the smallest complete sitemap, global navigation, page-specific goals, and cross-page conversion path before styling individual sections. For web products, prioritize task completion and state coverage over marketing rhythm.

Use only the sections needed for the visitor's decision. A common conversion-page flow is:

  1. hero with the promise and primary action;
  2. immediate proof;
  3. signature offering or product value;
  4. experience, process, or differentiators;
  5. practical details that remove objections;
  6. location, contact, booking, purchase, or inquiry;
  7. decisive final CTA.

Vary composition through scale, rhythm, crop, whitespace, and editorial alignment—not through arbitrary component variety. On mobile, keep the most valuable action within easy reach when appropriate, such as call, navigation, booking, buy, sign up, search, save, or continue.

6. Write content as interface

  • Write in the audience's primary language first; add other locales only when required.
  • Use short, direct headings that fit the visual rhythm.
  • Replace generic claims with supported specifics.
  • Keep CTA labels literal and culturally familiar.
  • Test target-language line breaks at real widths; do not approve layout using English placeholders.
  • Test every supported locale independently. Do not assume a translation fits the source-language component.
  • Use real content rather than lorem ipsum.

7. Build the prototype

  • Use semantic page structure and reusable tokens.
  • Make links, calls, navigation, forms, menus, filters, galleries, and modals work at the level implied by the prototype.
  • Optimize large images to appropriate web formats and sizes; retain meaningful crop focal points.
  • Use motion to clarify hierarchy, reveal spatial relationships, or reinforce brand character. Keep it sparse and interruptible.
  • Provide keyboard focus, meaningful alt text, sufficient contrast, 44px touch targets, and safe-area-aware mobile controls.
  • Preserve essential content when JavaScript motion does not run.

Read prototype-and-qa.md before implementation or final review.

8. Verify in a real browser

Do not declare completion from a successful build alone. Verify at minimum:

  • wide desktop around 1440px;
  • intermediate/tablet width;
  • mobile around 390px;
  • narrow mobile around 375px;
  • one reduced-motion or keyboard pass.

Inspect the hero and the complete page. Check target-language wrapping, image crops, horizontal overflow, safe areas, sticky controls, console errors, links, forms, menus, focus order, loading behavior, and final CTA reachability. Fix discovered issues and repeat the affected check.

Expected handoff

For a runnable prototype, deliver:

  • working source and a successful production build;
  • a local preview or clear run command;
  • PRODUCT.md and DESIGN.md when the project warrants them;
  • a source/placeholder note for operational facts and media;
  • a concise list of implemented interactions and validated viewports;
  • a short owner-confirmation list for facts that may change before launch.

Lead the final response with the preview or primary artifact. Mention only the most important design decisions and validation results.

Quality gate

Do not hand off until all are true:

  • Specificity: The site could not be relabeled for a random competitor without redesign.
  • Authenticity: Claims and media are real, sourced, or clearly marked as provisional.
  • Conversion: The next action is obvious within seconds and remains easy on mobile.
  • Hierarchy: One idea dominates each viewport; typography and imagery support it.
  • Responsiveness: Compositions adapt intentionally rather than merely stack.
  • Restraint: Motion and effects have a role; the site does not feel like an effects demo.
  • Inclusion: Keyboard, contrast, touch targets, reduced motion, and semantic structure are covered.
  • Reliability: Build, console, assets, and critical interactions are verified.

Individual skills in this repo

This repo contains 4 individual skills — each has its own dedicated page.

kinetic-typography-video

动效 KINETIC — 把口播视频(数字人或真人实拍)做成爆款动效质感的视觉语言与逐帧渲染器。用一条贯穿全片的"光"、按音节落下的大字、剪辑师才认得的圈内物件(框选框、真实声波、关键帧、时间线、口型条)、人物抠像后"字藏身后"、音画同钟的合成音效与节拍对齐的配乐。用户说"用动效做/用动效风格/做得更炫/像那条爆款动效视频一样/Opus 那种动效"、或要给数字人/真人口播加高密度动效包装时使用。不用于:只要简单字幕的口播、纯 MG 无人物讲解(用 knowledge-explainer)、结构化信息单镜头(shotcraft-motion)。

knowledge-explainer

把文字、文章、笔记、PPT 或 HTML 课件提炼为有配音字幕的 16:9 知识讲解视频。 使用 Remotion 语义动画与 Fish Audio / ElevenLabs 配音,先做短样片再扩展。 用于课件转视频、知识视频、内训片、科普 explainer,也用于参考视频的解释型动效拆解与原创复现;不负责真人口播剪辑、数字人生成或未知作者工具归属鉴定。

video-prompt-director

创作者的主力 AI 视频提示词 Skill。先按脚本风格选档位,再像导演一样把脚本、故事、口播稿、分镜表或一句想法,拆成能直接生成的分段视频提示词。短视频档(抖音/小红书节奏:一镜到底、多分镜、街拍、角色小故事、竖屏短剧,标杆是《丹眉》和赛博原子)和电影精品档(慢节奏剧情、表演、对白、对手戏、AU 微表情、八维公式),各有结构和示例。每段含生成模式与参数、参考图绑定、人物道具场景、镜头时间轴、声音和硬约束。适用 Seedance 2.0/2.5、即梦、dreamina、LibTV、TT AI。Use when 用户要写视频提示词、AI 视频提示词、Seedance 提示词、即梦提示词、脚本转视频、拆分镜、分镜提示词、快剪、短视频提示词、storyboard、shot list、多图参考、首尾帧、人物表演提示词、AI 演技、情绪递进、导演式提示词,或给出脚本/故事/想法要变成视频。当前项目若有专用工作流 Skill(Vlog、情感钩子、新闻解说等),优先用专用的。

xhs-card-video

制作无脸说明视频的小红书 3:4 截图卡短视频:把真实 X 截图、动态背景和已核验本地 BGM 组装成一个视频与一份发布稿,并按周归档;不自动发布。

Related Skills