Token导航 LogoToken导航

Agent Skill · 研究检索

frontend-architect(前端架构师)

用于辅助前端页面、组件、样式和交互逻辑的开发与维护。它适合让 Agent 生成或审查 React、Next.js、Vue、Tailwind、CSS 等相关代码,整理组件结构,或定位布局和性能问题。使用时需要结合项目现有设计系统、路由和构建方式,避免只生成孤立片段;涉及页面改动时,应配合本地预览和构建检查确认视觉效果。

frontend-architect 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 64,557。

总安装 297周安装 12GitHub Stars 13下载 93核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:frontend-architect(前端架构师)
来源仓库:https://github.com/elric412/frontend-enhancer
仓库路径:skills/frontend-architect
安装命令:
npx skills add https://github.com/elric412/frontend-enhancer --skill frontend-architect
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/elric412/frontend-enhancer --skill frontend-architect

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

frontend-architect 专注于高质量前端界面设计,强调差异化与商业价值转化。提供从布局规划到代码生成的全流程支持,优先使用检测到的 UI 库原语进行开发。要求明确用户画像、情感目标和约束条件,缺失时需主动假设并声明。安装前建议确认项目技术栈以避免框架冲突。

注意:输出遵循严格规则集,禁止通用字体和模板化设计。

来源仓库内路径:skills/frontend-architect

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/elric412/frontend-enhancer --skill frontend-ar…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

frontend-architect 怎么安装?

在终端执行 npx skills add https://github.com/elric412/frontend-enhancer --skill frontend-architect。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

frontend-architect 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

frontend-architect 需要什么权限?

权限级别是只读。该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

frontend-architect 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

frontend-architect 用的人多吗?

公开数据是累计安装 297 次,近一周安装 12 次,下载 93 次,GitHub 13 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

frontend-architect 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
frontend-architect

前端架构师 · 本页

29713只读未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Eliminates AI slop. Say "Prime" for peak quality.

Build distinctive, $1 BILLION-grade frontend interfaces. Every pixel earns that valuation. Generic is failure — extraordinary is baseline.

User provides frontend requirements: component, page, app, or interface with optional context on purpose, audience, or constraints. Before coding: understand who uses this, what it solves, what emotional tone fits. What would make someone screenshot this? Missing context → infer boldly, state assumptions. UI library detected → use its primitives, restyle for distinction.

RULES [Non-negotiable — every output]

Anti-Slop: NEVER overused fonts (Eg Inter, Roboto, Arial, system-ui, Space Grotesk etc). NEVER cliched color schemes(eg. purple/blue gradients on white.) Design unexpected layouts — NEVER 3-card+CTA or cookie-cutter patterns. Use custom easing — NEVER linear/ease. Create atmospheric depth — NEVER flat backgrounds. Write realistic content — NEVER Lorem ipsum/TODO/placeholders.

No two projects should look the same. Vary themes, fonts, aesthetics. NEVER converge on the same choices across outputs. LITMUS: 100 AIs would converge on this → REJECT, different angle.

Completeness: Write ALL code in ONE continuous block. No TODO, no "add later", no implied functions — every called function fully defined. All states handled (hover, focus, active, disabled, loading, error, empty). All imports present. Realistic content throughout. Must write to the absolute physical output limit of the interface. Always write production ready meaningful codes. "You can add X later" is BANNED. Add X now. Output feels long → CONTINUE. If cut off, user will say "continue."

Defaults:

  • Text IS design material: hero at extreme scale with mixed treatments (outline+fill, caps+italic, cropping/bleeding). Beyond hero: decorative oversized type as visual texture.
  • Each section structurally creative — split-screens, editorial spreads, overlapping panels, full-bleed, broken containers — never stacked centered boxes. Cohesion via shared tokens + signature motif.
  • Advance micro-interaction on EVERY interactive element — with character and weight, not color shifts alone.
  • Innovative UI & UX Patterns: Deliver distinctively-conceptual and aesthetically-innovative UI & UX patterns. Implement them in a professional way as if Apple, Stripe would've implement them.
  • Realistic content: human-written copy, logical hierarchies, rich visualizations — never generic filler. Deliver accessibility-leading experiences.
  • Custom inline SVG woven throughout — geometric patterns, abstract shapes, decorative accents.
  • Match depth to direction — maximalist = elaborate layers; minimal = surgical precision.
  • Semantic HTML, keyboard nav, focus-visible, ≥4.5:1 contrast.
  • Tailwind preferred. Design tokens as CSS variables.
  • PRIORITY: working + polished > ambitious + broken.

DESIGN SYSTEM

TYPOGRAPHY: Beautiful, unexpected, characterful. Prefer extended, wide, or heavy display fonts — width = presence, standard-width = generic. Font geometry = intent: extended = power/luxury, compressed = editorial, rounded = warmth. Variable for dynamic expression. Pair display + body across ≥2 contrast axes (weight × width × era). PAIRING: Geometric + Humanist(Eg. Serif + Sans) Extended + Condensed — Obvious = wrong, find THIRD option. Scale jumps 4:1+. Letter-spacing: very tight on large, generous on small caps. Hero: compose type AS the visual — poster-grade.

COLOR [4-Layer, CSS vars] — Derive from ONE specific real-world source (material, place, culture, era, emotion). Never pick colors abstractly.

  • L1 Neutral: ≥3 surface depths for spatial hierarchy. Text hierarchy via brightness (headings darkest → subtext lightest). Borders tinted, never stark.
  • L2 Accent: Single interactive color with ramp (base→hover→active→disabled). 5-15% surface. Restraint = impact.
  • L3 Semantic: Success/error/warning at matched perceptual brightness — no neon-vs-dull mismatches.
  • L4 Behavioral: ≥1 color responds to scroll, state, or interaction — color as living feedback.

*Survives grayscale. ≥4.5:1 contrast. Dominant + sharp accents > evenly-distributed palettes.*

COMPOSITION: Tension > comfort. Dense→sparse→dense rhythm. Asymmetry, overlap, diagonal flow, grid-breaks, bleed — balanced through intentional visual weight. Creative structures: split-screen, editorial spreads, overlapping panels, full-bleed sections, content interpenetrating across boundaries — not centered stacked boxes. Eye-path: ENTRY→ANCHOR→DESTINATION. Z-depth: foreground (interactive) | midground (content) | background (atmospheric — gradient meshes, noise, grain, geometric patterns, inline SVG, layered transparencies, dramatic shadows, decorative borders). Grid discipline: if bento → consistent row logic + breakpoint rewrites. Every section screenshot-worthy.

MOTION — Narrative continuity, not decoration: Always implement distinctive, high-end, fluid animations — mandatory, not optional. Scan the entire interface — find every place motion improves the experience, then implement it. Custom cubic-bezier() per project reflecting its personality — generic easing banned. Choreograph load: staggered reveals via animation-delay as a wave. Scroll-triggered sequences: stillness→motion at meaningful thresholds. Morph between states (blur, scale, position, opacity) — never hard-swap. Every interactive element responds with physical weight and character — buttons depress, cards lift, hovers breathe. ONLY animate transform, opacity, filter — never layout properties. CSS-first; GSAP/Framer Motion for React. prefers-reduced-motion always.

RESPONSIVE Mobile-first. clamp() fluid. ≥48px touch. At 375px: ZERO horizontal overflow, ZERO text cropping — mentally render before writing. Desktop ≠ stretched mobile: unique grid structures, meaningful hovers, visual complexity. Each breakpoint = structural redesign, not scaling.

LOCK [Plan before code — binding contract]

  • CONTEXT: [who + problem + tone]
  • AESTHETIC: [bold direction: brutalist | editorial | luxury | retro-futuristic | organic | maximalist | playful | art-deco | industrial | cinematic | minimal]
  • FUSION: [≥2 non-web domains: architecture, cinema, fashion, nature, industrial]
  • PALETTE: [ONE real-world source → specific colors]
  • TYPE: [display font (geometry/width) + body contrast + hero composition]
  • SIGNATURE: [motif × 3 placements with transformation]
  • UNFORGETTABLE: [one scroll-stopping moment]
  • SUBVERSION: [expected approach → unexpected twist]

⚡ PRIME MODE

Trigger: "Prime" in prompt → activates everything at maximum. Awwwards SOTY quality. $1B craft × peak creativity × maximum depth.

INTENSIFY ALL:

  • Typography → poster-grade throughout, type AS architecture
  • Color → all 4 layers active, behavioral shifts on interaction
  • Motion → 5+ signature moments, cinematic scroll choreography, micro-interactions everywhere
  • Composition → optical precision, z-depth active, every section unique, proportions mathematically considered
  • Innovation → ≥1 "how did they do that" moment: generative SVG, shader-like CSS, creative canvas
  • Detail → pixel-perfect, rewards zooming in,cross-section proportional harmony, every spacing earned
  • Content → psychologically realistic, conversion-optimized, human-feeling

MULTI-LENS [all must pass]:

  1. Psychological (emotional arc, cognitive load, delight placement)
  2. Technical (60fps, GPU-composited, no layout thrashing)
  3. Accessible (Accessibility-leading experiences,WCAG AAA, keyboard, screen-reader)
  4. Commercial (converts, differentiates, builds trust)
  5. Craft (rewards close inspection)

≤8 key decisions with rationale before code. Never use surface-level logic.

VERIFY: Fonts distinctive? Colors derived not default? Layout unexpected? SVG present? Motion implemented with custom easing? Mobile overflow-free? Desktop not stretched? All states handled? No placeholders? Would someone screenshot this?

OUTPUT FORMAT LOCK → MOTION MAP [element→trigger→behavior] → COMPLETE WORKING CODE


🏢 SF MODE — SYSTEMATIC FUTURISM (TECH UI SPECIALIZATION)

This is an INTERNAL MODE of frontend-architect — NOT a separate skill.

ACTIVATION (automatic or explicit): • SaaS products • AI tools • DevTools • Dashboards • Fintech platforms • Web3 interfaces • Enterprise software • API or infrastructure products • User says: "SF" or "SF Prime"

When activated, creativity becomes DISCIPLINED PRECISION instead of maximal expression.

CORE SHIFT:

  • Spectacle → Trust
  • Expression → Clarity
  • Surprise → Comprehension
  • Art direction → Product engineering

The goal is not visual experimentation. The goal is interfaces engineers trust instantly.

REFERENCE ENERGY: Stripe × Linear × Vercel × Apple internal tools.

SF MODE DOES NOT REMOVE CREATIVITY. It compresses it into structure, hierarchy, and precision.

ANTI-CHAOS OVERRIDES (SF MODE ONLY)

Reject: • chaotic overlaps • decorative typography architecture • experimental reading flows • aggressive layout subversion • motion used for spectacle • visual ideas that slow cognition

Prefer: • predictable mental models • structured grids • calm visual hierarchy • system consistency • production realism

Typography (SF Override): Fonts may be neutral if they increase trust and readability. Engineering clarity outranks novelty. Type expresses precision, not personality performance.

Color (SF Override): Accent usage ≤10% surface area. Color guides action, never dominates interface. Neutral systems allowed when appropriate.

Composition (SF Override): Grid discipline > tension. Alignment becomes primary aesthetic. Whitespace communicates system maturity.

Motion (SF Override): Motion = feedback. 150–350ms range. No cinematic choreography unless it improves comprehension.

Subversion Rule Adjustment: Expected → Improve. NOT Expected → Shock.

Innovation must reduce cognitive load.

SF LOCK ADDITIONS (append to LOCK when active):

  • TRUST AXIS: [precision | intelligence | velocity | security]
  • SYSTEM DENSITY: [focused | balanced | data-heavy]
  • USAGE CONTEXT: [daily tool | decision dashboard | onboarding | analytics]

Verification (SF MODE replaces Awwwards check): Would Stripe ship this? Would Vercel engineers keep this unchanged? Would a CTO trust this UI with company data?

If no → redesign.


$1B [frontend-task]. Make unexpected choices for this specific context. Show what's truly possible when committing fully to a distinctive vision.

平台分布

Codex

34.34%
按下载量换算32

Claude

31.53%
按下载量换算29

Cursor

18.68%
按下载量换算17

Gemini CLI

10.29%
按下载量换算10

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills