Token导航 LogoToken导航

Agent Skill · 研究检索

design-exploration(设计探索)

用于辅助界面设计、视觉规范、排版、配色、布局和交互体验优化。它适合让 Agent 根据产品场景整理页面结构、生成 UI 方案、检查视觉一致性或改进组件层级。使用时需要结合现有品牌、设计系统和用户任务,不应只堆装饰元素;涉及真实页面改动时,应通过截图或浏览器预览检查文本溢出、对齐和响应式表现。

design-exploration 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,权限为操作浏览器;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 44,563。

总安装 906周安装 37GitHub Stars 106下载 293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:design-exploration(设计探索)
来源仓库:https://github.com/carson2222/skills
仓库路径:skills/design-exploration
安装命令:
npx skills add https://github.com/carson2222/skills --skill design-exploration
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/carson2222/skills --skill design-exploration

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。

design-exploration 生成同一组件的 N 种差异化设计方案,拓展视觉可能性与创意空间。适用于新项目启动或既有系统升级时的多方向探索与对比选择场景。每种变体保持功能一致但视觉语言迥异,便于快速收敛至最优解。使用前应提供现有设计系统与品牌指南,防止风格漂移与认知负荷增加。

适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/design-exploration

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/carson2222/skills --skill design-exploration;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

    它需要操作浏览器,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

design-exploration 怎么安装?

在终端执行 npx skills add https://github.com/carson2222/skills --skill design-exploration。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

design-exploration 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。

design-exploration 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

design-exploration 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。

design-exploration 用的人多吗?

公开数据是累计安装 906 次,近一周安装 37 次,下载 293 次,GitHub 106 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。

design-exploration 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。

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。

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
design-exploration

设计探索 · 本页

906106操作浏览器未明确通过

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

Design Exploration

Generate N distinct design variants for a given component, page, or section. Each variant takes a fundamentally different visual direction. Present all for comparison, then fully implement the chosen one.

When to Trigger

1. New project

The user is bootstrapping a frontend from scratch. No existing design system -- the variants define it.

2. Existing app

The user has an established design language and wants multiple approaches for a new or existing component, section, or page. Every variant must stay consistent with the app's existing patterns and tokens.

Bypass: Detailed design spec provided

If the user provides exact fonts, colors, hex values, border radii, and specific styling instructions -- skip Phase 2 entirely. This is an implementation request, not an exploration request. Do not suggest alternatives or inject creative direction. Read each target file, preserve all logic, apply exactly the specified changes. Go directly to Phase 3.

Phase 1: Research

Before designing anything:

  1. Read the existing code -- understand every component, import, animation, and logic in the target files.
  2. Study the design language -- globals.css, tailwind config, component library, existing patterns (spacing, colors, typography, card styles).
  3. Identify constraints -- what must NOT change: data, logic, API calls, routing, state management.
  4. Understand the content model -- what data is displayed, edge cases (long text, many items, empty states).
  5. Check assets -- verify images, icons, and assets exist with correct formats (transparency, resolution, aspect ratio). If assets need preprocessing (background removal, format conversion), handle that before building variants on top of them.
  6. Look beyond the brief -- find data values available in the codebase but not currently displayed, existing assets that could be repurposed, values that could be visualized differently (charts, progress bars, indicators instead of plain text), opportunities for grouping or layering information.

Phase 2: Generate Variants

Create exactly the number requested (default: 5). Determine mode automatically based on project state -- do not ask.

Mode A: New Project

No existing design system to follow.

  1. Research broadly -- search for design approaches across disciplines: product design, editorial, print, architecture, fashion branding, game UI, packaging. Do not rely on memorized style lists.
  2. Analyze context -- product purpose, audience, intended mood. A trading app has different needs than a portfolio.
  3. If context is insufficient, ask -- what is the product, who is it for, what feeling should it evoke. Exception: if the user explicitly wants broad exploration, skip this.
  4. Select styles optimal for this project -- each must be a plausible direction for this exact product, not a showcase of range. Respect explicit style exclusions as hard constraints. Do NOT fall into deterministic patterns of always picking the same style families.
  5. Maximize diversity -- every variant must differ in spatial thinking, typographic voice, and emotional register. Not "same layout, different colors."

Mode B: Existing App

Established styles, tokens, components, and patterns exist.

  1. Study the design system first -- globals.css, tailwind config, theme tokens, color palette, typography, spacing, animation conventions, border radii, card styles.
  2. The app's style is the baseline -- all variants must be consistent with the existing system. You are exploring different layouts and compositions, not different design systems.
  3. Vary structure, not identity -- differ in layout, information hierarchy, component composition, animation approach, and density. Do NOT differ in font families, color palette, or border radius unless the user explicitly asks for a style refresh.
  4. Align with surrounding UI -- match sizing, spacing, and alignment of adjacent elements. If next to a button, match its height. If inside a card grid, follow the same padding and gap.
  5. Style change escape hatch -- if the user explicitly asks for a style change, break from the existing system. Treat it closer to Mode A but use the current app as context.

Required Differentiation

Each variant MUST differ across ALL of these axes:

AxisMode A (New)Mode B (Existing)
TypographyDifferent font pairing per variantUse app fonts; vary weight, size, hierarchy
ColorDifferent accent, background, contrast approachUse app palette; vary application and emphasis
LayoutGrid vs. list vs. asymmetric vs. card-based vs. dense vs. airyDifferent arrangement within app conventions
MoodDistinct personality per variant, discovered through researchDistinct compositional approach, shared personality
Border/radiusSharp vs. rounded vs. pill vs. mixedUse app's existing radius
MotionDifferent animation philosophyDifferent choices within app motion conventions
BackgroundSolid vs. gradient vs. texture vs. pattern vs. atmosphericUse app patterns; vary section treatments

Variant Rules

  • No AI slop -- no generic purple-on-white, no Inter/Roboto/Arial defaults, no cookie-cutter card grids, no Space Grotesk convergence across sessions.
  • No gimmick styles by default -- never select neo-brutalism, terminal/hacker, retro-CRT, or vaporwave unless the user explicitly requests it. Default to styles that could ship in production.
  • Simplicity is valid -- if the user asks for simple/clean/minimal, deliver exactly that. Clean typography, generous whitespace, conventional layout. No decorative elements, no effects.
  • Responsive required -- every variant must work on mobile. Use prefers-reduced-motion and touch-friendly interactions.
  • Cross-browser safe -- avoid bleeding-edge CSS that breaks in Safari or Firefox.
  • Mock data only -- use realistic mock data covering edge cases. Do NOT wire up real API calls or state management during exploration. That happens in Phase 3.

Variant Presentation

For each variant, provide:

  1. Name -- short and evocative (e.g., "Ember", "Signal", "Nocturne")
  2. Description -- 2-3 sentences on the aesthetic direction and why it fits
  3. Key choices -- fonts, accent color, layout approach, motion style
  4. What changed -- if the variant introduces new data, repurposes assets, restructures information, or deprioritizes existing data, state it explicitly
  5. Full implementation -- working code with mock data

Implement a simple switching mechanism so the user can cycle through variants in the browser (e.g., a small floating selector, query param, or keyboard shortcut). The user must be able to compare all variants side by side or toggle between them without touching code.

If All Variants Are Rejected

  1. Ask what went wrong -- style direction, layout, quality, or everything.
  2. Ask for a reference -- website, app, Dribbble shot, screenshot.
  3. Offer 1-2 targeted variants based on feedback, not another blind batch.
  4. If "just make it simple" -- drop exploration. One clean, conservative design.

Phase 3: Implementation

Once the user picks a variant:

  1. Delete all other variants -- clean up completely:

- Search for remaining variant references, conditionals, and selection logic - Check for orphaned files, unused imports, unused CSS classes - Run the build to verify no broken imports - The codebase should look like the selected variant was the only implementation

  1. Wire up real data -- replace all mock data with real API calls, state management, and logic.
  2. Preserve all existing functionality -- every import, handler, and logic piece from the original must survive. Only visual styling changes.
  3. Read before writing -- always read current file content before overwriting. Never work from memory.
  4. Verify animations -- use GPU-accelerated properties (transform, opacity) instead of layout-triggering ones (top, left, width, height). Use ease-out for entrances, ease-in-out for state changes. If an animation looks janky, simplify it.
  5. Test responsive -- verify at mobile, tablet, and desktop breakpoints.
  6. Verify no regressions -- run build/lint if available. Check all imports resolve.

Aesthetic Standards

These apply to every variant in every mode.

Typography -- pair a distinctive display font with a refined body font. Never default to system fonts or overused families. Each variant uses a different pairing (Mode A) or different weight/hierarchy treatment (Mode B).

Color -- commit to a dominant color with sharp accents. Timid, evenly-distributed palettes look undesigned. Use CSS variables for consistency.

Motion -- one well-orchestrated entrance sequence (staggered reveals, animation-delay) creates more impact than scattered micro-interactions. Prioritize page load and scroll-triggered moments over hover effects.

Spatial composition -- vary layouts meaningfully: asymmetry, overlap, diagonal flow, grid-breaking elements, controlled density, or generous negative space. Avoid defaulting to centered card grids.

Background and depth -- create atmosphere through gradient meshes, noise textures, geometric patterns, layered transparencies, or dramatic shadows. Solid white/gray backgrounds are a last resort, not a default.

Anti-convergence -- never converge on the same font, color scheme, or layout pattern across variants or across sessions. Each generation should feel curated by a different creative director.

Critical Constraints

  • All data must be accounted for -- every value and metric from the original must appear in every variant. Display format may change (text to chart, number to progress bar), but nothing is silently dropped. If a variant deliberately omits or deprioritizes a data point, it MUST be called out in the variant description.
  • No placeholder content -- mock data during variants, real data in final implementation.
  • File completeness -- write the ENTIRE file. No "rest remains the same" comments.
  • Font imports -- update layout/entry file imports when changing fonts (next/font/google, @font-face, etc.).
  • Design tokens -- update CSS variables/theme tokens to match the chosen variant. Do not hardcode colors inline when the app uses a token system.

平台分布

Codex

35.93%
按下载量换算105

Claude

32.94%
按下载量换算97

Cursor

17.83%
按下载量换算52

Gemini CLI

9.69%
按下载量换算28

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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