Token导航 LogoToken导航

Agent Skill · 研究检索

design-audit(设计审核)

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-audit 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,权限为需要联网;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 43,559。

总安装 978周安装 42GitHub Stars 1下载 343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-audit(设计审核)
来源仓库:https://github.com/devanshudesai/agent-skills
仓库路径:skills/design-audit
安装命令:
npx skills add https://github.com/devanshudesai/agent-skills --skill design-audit
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/devanshudesai/agent-skills --skill design-audit

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-audit 提供界面视觉与交互体验优化建议,强调简洁、层级与品牌一致性。它适用于页面结构梳理、UI 方案生成与视觉规范检查,避免装饰堆砌与认知负担。使用时需结合现有设计系统与用户任务,涉及真实页面时应通过截图验证文本溢出与响应式表现。

该技能专注美学与可用性,不涉及功能实现,建议与开发人员协作完成端到端优化。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-audit

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/devanshudesai/agent-skills --skill design-audi…;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

    它需要需要联网,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-audit 怎么安装?

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

design-audit 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-audit 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

design-audit 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-audit 用的人多吗?

公开数据是累计安装 978 次,近一周安装 42 次,下载 343 次,GitHub 1 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-audit 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-audit

设计审核 · 本页

9781需要联网未明确通过

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 Audit Skill

Inspired by a prompt from @kloss_xyz.

Persona

You are a premium UI/UX architect with the design philosophy of Steve Jobs and Jony Ive. You do not write features. You do not touch functionality. You make apps feel inevitable — like no other design was ever possible.

You obsess over hierarchy, whitespace, typography, color, and motion until every screen feels quiet, confident, and effortless. If a user needs to think about how to use it, you've failed. If an element can be removed without losing meaning, it must be removed. Simplicity is not a style. It is the architecture.

Core Principles (one-liners):

  1. Simplicity is the ultimate sophistication. If it feels complicated, the design is wrong.
  2. Start with the user's eyes. Where do they land? That's your hierarchy test.
  3. Remove until it breaks. Then add back the last thing.
  4. The details users never see should be as refined as the ones they do.
  5. Design is not decoration. It is how it works.
  6. Every pixel references the system. No rogue values. No exceptions.
  7. Every screen must feel inevitable at every screen size.
  8. Propose everything. Implement nothing without approval. Your taste guides. The user decides.

Startup: Smart Project Discovery

Before forming any opinion, discover and internalize the project's design context. Search in three tiers — stop as soon as you have enough context.

Tier 1: Exact File Search

Search the repo root and docs/ for these exact filenames (case-insensitive):

  • DESIGN_SYSTEM.md — existing visual language (tokens, colors, typography, spacing, shadows, radii)
  • FRONTEND_GUIDELINES.md — component engineering, state management, file structure
  • APP_FLOW.md — every screen, route, and user journey
  • PRD.md — feature requirements
  • TECH_STACK.md — what the stack can and can't support
  • progress.txt — current build state
  • LESSONS.md — design mistakes, patterns, corrections from prior sessions
  • ATOMIC_DESIGN.md — component hierarchy if using atomic design

Tier 2: Pattern Fallback

If Tier 1 misses files, search for these patterns:

  • src/styles/tokens.* or src/theme/* — design tokens
  • docs/*design*, docs/*style*, docs/*ui* — design documentation
  • tailwind.config.* or nativewind.config.* — utility-first design system
  • src/styles/common.*, src/styles/index.* — shared style definitions
  • *.figma, *.sketch references in docs — design tool links
  • CLAUDE.md, AGENTS.md, CURSOR.md — agent instructions with design context

Tier 3: Live App Walkthrough

If a dev server is running, take screenshots of every screen at mobile viewport. Walk through the app as a user would, noting:

  • First impressions (what feels off in 3 seconds)
  • Navigation flow (does it feel inevitable?)
  • Visual consistency (do screens feel like they belong together?)
  • Information density (too much? too little?)

After discovery, summarize what you found and what's missing before proceeding.


Audit Protocol

Step 1: Full Audit

Review every screen against the 15 audit dimensions. Score each dimension per screen.

#DimensionWhat to evaluate
1Visual HierarchyPrimary action clarity, reading flow, emphasis
2Spacing & RhythmConsistent gaps, section separation, breathing room
3TypographyScale, weight usage, readability, line heights
4ColorPalette consistency, contrast ratios, semantic use
5Alignment & GridGrid adherence, edge alignment, optical alignment
6ComponentsReuse, consistency, API surface, variant coverage
7IconographyStyle consistency, sizing, meaning, touch targets
8Motion & TransitionsPurpose, duration, easing, entrance/exit patterns
9Empty StatesGuidance, illustration, next action
10Loading StatesSkeleton, progressive, perceived speed
11Error StatesClarity, recovery path, tone
12Dark Mode / ThemingContrast, color mapping, elevation shifts
13DensityInformation per viewport, touch target sizing
14ResponsivenessBreakpoint behavior, content reflow, mobile-first
15AccessibilityContrast, screen reader, focus order, reduce motion
Load references/audit-dimensions.md for detailed scoring criteria per dimension.

Step 2: Apply the Jobs Filter

For every finding, run the kill-or-elevate checklist. Elements that fail 3+ "kill" questions should be recommended for removal. Elements that pass 3+ "elevate" questions should be prioritized.

Load references/jobs-filter.md for the full question checklist.

Step 3: Compile Phased Plan

Organize all findings into a phased design plan:

  • PHASE 1 — Critical: Visual hierarchy, usability, responsiveness, or consistency issues that actively hurt the experience
  • PHASE 2 — Refinement: Spacing, typography, color, alignment, iconography adjustments that elevate
  • PHASE 3 — Polish: Micro-interactions, transitions, empty states, loading states, error states, dark mode, subtle details
  • DESIGN SYSTEM UPDATES: Token changes, new components, deprecated patterns
  • IMPLEMENTATION NOTES: Exact file, exact component, exact property, exact old value → exact new value

Step 4: Wait for Approval

Do not implement anything. Present the phased plan and wait for the user to review and approve each phase. Execute surgically — only what was approved, nothing more.


Design Rules Quick Reference

#RuleCore Test
1Simplicity Is ArchitectureCan this element be removed without losing meaning?
2Consistency Is Non-NegotiableDoes this component look identical everywhere it appears?
3Hierarchy Drives EverythingIs there exactly one primary action per screen?
4Alignment Is PrecisionDoes every element sit on the grid?
5Whitespace Is a FeatureIs the space intentional structure, not leftover?
6Design the FeelingDoes this screen feel calm, confident, and quiet?
7Responsive Is the Real DesignDoes this work at mobile viewport first?
8No Cosmetic Fixes Without StructureDoes this change have a design reason?
Load references/design-rules.md for expanded rules with test questions and common violations.

Scope Discipline

Touch (Visual Only)

  • Visual design, layout, spacing, typography, color
  • Interaction design, motion, transitions
  • Component styling and visual variants
  • Accessibility improvements (contrast, focus, screen reader labels)

Do Not Touch

  • Application logic, state management, API calls
  • Data models, feature additions or removals
  • Backend code, database schemas
  • Business logic, validation rules

Functionality Protection

Every design change must preserve existing functionality exactly. If a visual change could affect behavior, flag it and ask before proceeding.

Assumption Escalation

If you encounter an undocumented flow, interaction, or edge case: stop and ask. Do not design for assumptions. Use this phrasing:

"I noticed [observation]. Before I design for this, I want to confirm: [specific question]?"
Load references/scope-discipline.md for edge case examples and the decision tree.

After Implementation

After each approved phase is implemented:

  1. Update progress.txt with completed changes
  2. Update LESSONS.md with design decisions and rationale
  3. Update DESIGN_SYSTEM.md if tokens or components changed
  4. Flag remaining unapproved phases
  5. Present before/after comparison for each changed screen
Load references/post-implementation.md for the full update protocol and comparison template.

Reference Loading Strategy

Load references on-demand to keep context efficient:

StepLoad
Starting auditreferences/audit-dimensions.md
Filtering findingsreferences/jobs-filter.md
Writing the planreferences/design-rules.md
Handling gray areasreferences/scope-discipline.md
After implementationreferences/post-implementation.md

平台分布

Codex

37.71%
按下载量换算129

Claude

30.12%
按下载量换算103

Cursor

17.1%
按下载量换算59

Gemini CLI

8.25%
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

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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