Token导航 LogoToken导航

Agent Skill · 研究检索

accessibility(无障碍)

用于辅助无障碍访问检查、页面可用性审计和前端可访问性改进。它适合让 Agent 检查语义标签、键盘操作、颜色对比、ARIA 属性和自动化检测结果。使用时需要结合真实页面和浏览器验证,不应只依赖静态文本判断;涉及修复建议时,应兼顾设计系统、组件复用和 WCAG 等通用无障碍规范。

accessibility 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 43,453。

总安装 984周安装 41GitHub Stars 14下载 328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:accessibility(无障碍)
来源仓库:https://github.com/connorads/dotfiles
仓库路径:skills/accessibility
安装命令:
npx skills add https://github.com/connorads/dotfiles --skill accessibility
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/connorads/dotfiles --skill accessibility

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。

用于辅助无障碍访问检查与前端可访问性改进。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。可审查语义标签、键盘导航、颜色对比与 ARIA 属性。适合结合真实页面与浏览器工具验证检测结果。涉及修复建议时应遵循 WCAG 规范并保持组件复用性。

accessibility 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/accessibility

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/connorads/dotfiles --skill accessibility;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

accessibility 怎么安装?

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

accessibility 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。

accessibility 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。

accessibility 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。

accessibility 用的人多吗?

公开数据是累计安装 984 次,近一周安装 41 次,下载 328 次,GitHub 14 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。

accessibility 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
accessibility

无障碍 · 本页

98414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

Accessibility

Web accessibility done right means your UI is navigable, understandable, and operable by people who cannot use a mouse — primarily those using screen readers (NVDA, JAWS, VoiceOver), keyboard-only users, and those with motor, cognitive, or visual impairments. The 2024 WebAIM Million report found 95.9% of home pages failing basic accessibility checks. Most failures are preventable with the right mental model.

The Core Mental Model

Screen readers linearise a 2D page into a 1D audio stream. A blind user never sees the whole page at once — they navigate sequentially by headings, landmarks, form fields, links, and interactive controls using keyboard shortcuts. Every decision you make should answer: *"What will a screen reader announce, and does it make sense in isolation?"*

The three rules that flow from this:

  1. Semantics over style — use native HTML elements (<button>, <nav>, <h2>) before reaching for ARIA. Native elements come with free keyboard support, accessible names, and correct roles.
  2. Context must travel with the element — a screen reader user navigating by tab or by links list sees elements stripped of their visual neighbours. Labels, descriptions, and states must be programmatically attached, not implied by proximity.
  3. Dynamic changes must be announced — screen readers only notice changes if focus moves to new content or a live region announces it. Silent DOM mutations are invisible to AT.

When Auditing Existing UI

Review in this priority order — fix critical issues before polishing low-impact ones:

PriorityCategoryWCAG LevelSee
1Accessible names (buttons, inputs, links)Areferences/aria-patterns.md
2Keyboard operability (all interactive elements)Areferences/focus-management.md
3Focus management (dialogs, SPAs, live regions)A/AAreferences/focus-management.md
4Semantic structure (headings, landmarks, lists)Areferences/wcag-checklist.md
5Form errors and validationA/AAreferences/common-fixes.md
6Colour contrast and visual statesAAreferences/wcag-checklist.md
7Dynamic content announcementsAAreferences/aria-patterns.md
8Images and mediaAreferences/wcag-checklist.md

Quote the exact failing snippet, name the WCAG criterion, and propose the smallest viable fix. Do not refactor unrelated code.


When Building New UI

The quick decision tree

Need an interactive control?
  ↓
Does a native HTML element do this?  → YES → Use it. Done.
  ↓ NO
Use the correct ARIA role + required attributes + keyboard handler.

Adding dynamic content?
  ↓
Does focus move to the new content? → YES → No live region needed.
  ↓ NO
Is it a transient status (toast, cart count, form error)?
  → Use aria-live="polite" (or role="alert" for errors)

Opening a dialog/modal?
  → Trap focus inside. Restore focus to trigger on close.
  → See references/focus-management.md

Mandatory checks before shipping any interactive component

  • Every input, select, textarea has an associated <label> (not just placeholder)
  • Every button has an accessible name (text content, aria-label, or aria-labelledby)
  • Every icon-only control has aria-label; the icon has aria-hidden="true"
  • Focus is visible on all interactive elements (never outline: none without a replacement)
  • Tab order is logical and matches visual order
  • All pointer interactions have a keyboard equivalent
  • No tabindex greater than 0

The Five Most Common Failures (and their fixes)

1. Icon-only button with no accessible name

<!-- ❌ Screen reader announces: "button" -->
<button><svg>...</svg></button>

<!-- ✅ Screen reader announces: "Close, button" -->
<button aria-label="Close"><svg aria-hidden="true">...</svg></button>

2. Input with no label

<!-- ❌ Screen reader announces: "edit text" -->
<input type="email" placeholder="Email" />

<!-- ✅ Screen reader announces: "Email address, edit text" -->
<label for="email">Email address</label>
<input id="email" type="email" />

3. div or span used as a button

<!-- ❌ Not keyboard accessible, no role announced -->
<div onclick="save()">Save</div>

<!-- ✅ Free keyboard support, correct role -->
<button onclick="save()">Save</button>

4. Form error not linked to field

<!-- ❌ Error visible but not associated with the field -->
<input id="email" type="email" />
<span>Please enter a valid email</span>

<!-- ✅ Screen reader announces error when field is focused -->
<input id="email" type="email"
       aria-describedby="email-err"
       aria-invalid="true" />
<span id="email-err" role="alert">Please enter a valid email</span>

5. Dynamic content updated silently

<!-- ❌ Cart count updates, screen reader users never know -->
<span id="cart-count">3</span>

<!-- ✅ Announces "4 items in cart" when count changes -->
<span id="cart-count" aria-live="polite" aria-atomic="true">4 items in cart</span>

Screen Reader Testing

Automated tools catch ~30–40% of accessibility issues. The rest require AT testing.

Minimum viable test matrix:

  • NVDA + Chrome or Firefox (Windows) — covers ~66% of screen reader users
  • VoiceOver + Safari (macOS/iOS) — covers Apple ecosystem
  • Add JAWS + Chrome for enterprise contexts

Core navigation patterns to test manually:

  1. Tab through all interactive elements — are names and roles announced correctly?
  2. Press H to navigate by headings — is the page structure logical?
  3. Press D to navigate by landmarks — are regions clearly labelled?
  4. Open and close any dialogs — does focus trap, then restore?
  5. Submit a form with errors — are error messages announced?
  6. Trigger any dynamic content update — is the change announced?

See references/screen-readers.md for NVDA/JAWS/VoiceOver commands, browse vs. forms mode, and testing scripts.


ARIA: The Rules

Rule 0: Don't use ARIA if native HTML solves it. Bad ARIA is worse than no ARIA.

Rule 1: aria-label and aria-labelledby provide the accessible name (what the element *is*). Rule 2: aria-describedby provides supplementary description (what it *does* or *needs*). Rule 3: aria-live="polite" for non-urgent updates; role="alert" (implicit assertive) for errors. Rule 4: Live regions must exist in the DOM on page load — inject text into them, don't inject the region itself. Rule 5: aria-hidden="true" removes from the AT tree completely. Never apply to focusable elements.

Full ARIA pattern library → references/aria-patterns.md


Visually Hidden Content

To show content to screen readers but hide it visually:

.visually-hidden:not(:focus):not(:active) {
  clip-path: inset(50%);
  height: 1px;
  overflow: hidden;
  position: absolute;
  white-space: nowrap;
  width: 1px;
}

Use for: skip links, supplementary link context ("Read more about caching"), icon button labels when aria-label is impractical for translation reasons.

Do not use for: content that sighted users need. Hiding meaningful content from one group creates disparity, not accessibility.


Colour and Contrast (WCAG AA)

Content typeMinimum ratio
Body text (<18pt / <14pt bold)4.5:1
Large text (≥18pt or ≥14pt bold)3:1
UI components (borders, icons, focus rings)3:1
Placeholder text4.5:1
Disabled elementsExempt

Never convey information by colour alone — always pair with a shape, pattern, or text label.


Reference Files

FileContents
references/screen-readers.mdNVDA/JAWS/VoiceOver commands, browse vs. forms mode, testing scripts per component type
references/aria-patterns.mdARIA roles, labelling hierarchy, live region patterns, complex widget ARIA (combobox, tabs, tree)
references/focus-management.mdModal focus trap, SPA route change focus, skip links, focus restoration patterns
references/wcag-checklist.mdWCAG 2.2 AA criterion-by-criterion checklist with pass/fail examples
references/common-fixes.mdCode-level fix templates for the 20 most common audit findings

平台分布

Codex

37.46%
按下载量换算123

Claude

29.5%
按下载量换算97

Cursor

16.96%
按下载量换算56

Gemini CLI

8.29%
按下载量换算27

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