关于
The `accessibility-audit` skill performs a comprehensive WCAG 2.1 AA compliance audit to identify and help remediate accessibility issues. It is triggered for systematic reviews, compliance preparation, or addressing reported problems, going deeper than general QA checks. Use it when accessibility itself is the primary goal, as it produces a detailed remediation plan.
快速安装
Claude Code
推荐npx skills add rampstackco/claude-skills -a claude-code/plugin add https://github.com/rampstackco/claude-skillsgit clone https://github.com/rampstackco/claude-skills.git ~/.claude/skills/accessibility-audit在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
Accessibility Audit
Run a thorough accessibility audit and produce a remediation plan. Stack-agnostic. Anchored to WCAG 2.1 AA, with notes on AAA where relevant.
This skill goes deeper than the accessibility checks in qa-testing and design-standards. Use this when accessibility itself is the goal.
When to use
- Pre-launch accessibility verification
- Compliance preparation (ADA, EN 301 549, AODA, Section 508)
- Remediation after an audit finding or complaint
- Annual or quarterly accessibility health check
- Onboarding accessibility into a team that hasn't prioritized it before
When NOT to use
- General QA after deploys (use
qa-testing) - Component-level accessibility implementation (use
frontend-component-build) - Color contrast for design tokens (use
design-standardsorbrand-identity)
Required inputs
- The site or product under audit
- The scope (full site, specific section, specific user flow)
- The target standard (WCAG 2.1 AA is most common)
- Any specific concerns or known issues
- Tools available (automated scanners, screen readers, manual testing)
The framework: WCAG's 4 principles
WCAG organizes accessibility around four principles. The audit covers each in depth.
1. Perceivable
Information and UI must be presentable in ways users can perceive.
Audit checks:
- Text alternatives. All non-decorative images have descriptive
alttext. Decorative images usealt="". Complex images (charts, infographics) have long descriptions. - Time-based media. Videos have captions. Pre-recorded audio has transcripts. Live audio has live captions where required.
- Adaptable. Content structure is conveyed through markup (semantic HTML), not just visual styling. Reading order makes sense when CSS is disabled.
- Distinguishable. Color is not the sole means of conveying information. Text contrast meets AA (4.5:1 normal, 3:1 large). UI element contrast meets 3:1. Audio can be paused, stopped, or muted.
2. Operable
UI components and navigation must be operable.
Audit checks:
- Keyboard accessible. All functionality available via keyboard alone. No keyboard traps. Focus visible.
- Enough time. Time limits can be adjusted, paused, or extended. Auto-updating content can be paused.
- Seizures and physical reactions. No content that flashes more than 3 times per second.
- Navigable. Skip links present. Pages have descriptive titles. Focus order is logical. Link purpose clear from text or context. Multiple ways to find pages (sitemap, search, navigation). Headings and labels are descriptive.
- Input modalities. Pointer gestures have keyboard alternatives. Pointer cancellation supported (mouse-up, not mouse-down for activation). Labels match accessible names. Motion-triggered functionality has alternatives.
3. Understandable
Information and operation must be understandable.
Audit checks:
- Readable. Page language declared (
<html lang="...">). Unusual words and abbreviations have definitions or expansions. Reading level appropriate to audience. - Predictable. Focus does not change context unexpectedly. Input does not change context unexpectedly. Navigation is consistent across pages. Components that look similar behave similarly.
- Input assistance. Errors are identified clearly. Labels and instructions are provided for input. Error suggestions are given where possible. For pages handling legal commitments or financial transactions, errors can be reviewed and corrected before submission.
4. Robust
Content must be robust enough to work with current and future user agents.
Audit checks:
- Compatible. Markup is valid. Name, role, and value of UI components are programmatically determinable. Status messages can be programmatically determined and announced.
Audit methodology
Stage 1: Automated scan
Run automated scanners across the priority pages. These catch 30 to 50 percent of issues but miss the rest.
Tools:
- axe DevTools (browser extension)
- Lighthouse (Chrome DevTools accessibility audit)
- WAVE (browser extension)
- Pa11y (CLI for batch scanning)
Output: A list of automated findings, by page.
Stage 2: Manual keyboard testing
Unplug the mouse. Navigate the priority user flows using only keyboard.
Test:
- Tab and Shift+Tab move through interactive elements in logical order
- Enter activates buttons and links
- Space activates buttons (and toggles checkboxes)
- Arrow keys navigate within composite widgets (tabs, menus, listboxes)
- Escape dismisses modals, popovers, menus
- Focus is always visible
- Focus returns to a sensible place after modals or popovers close
- No keyboard trap (focus can always leave)
Document: Any flow where keyboard navigation breaks down.
Stage 3: Screen reader testing
Test with at least one real screen reader, or state the gap per the data-availability rule. Each combination has quirks.
Common combinations:
- VoiceOver + Safari (macOS / iOS)
- NVDA + Firefox or Chrome (Windows)
- JAWS + Chrome (Windows; commercial but common in enterprise)
- TalkBack + Chrome (Android)
Test:
- Page structure announced correctly (headings, landmarks)
- Form labels read with their inputs
- Errors announced when they appear
- Status changes announced (loading, success, error)
- Modal context announced when opened
- Images have meaningful alt text (or are correctly identified as decorative)
Stage 4: Visual testing
Verify the visual aspects of accessibility.
Test:
- Color contrast for all text/background pairs (use a contrast checker)
- UI element contrast (3:1 for icons, borders, focus rings)
- Color-blindness simulation (deuteranopia at minimum)
- Zoom to 200% - content remains usable, no horizontal scroll
- Reflow at 320px viewport
- Text spacing applied (line height, letter spacing) - no content cut off
- Motion can be reduced (
prefers-reduced-motionhonored)
Stage 5: Cognitive accessibility
Often overlooked. Critical for inclusive products.
Test:
- Reading level appropriate
- Instructions clear
- Error messages explain how to fix the error, not just that one occurred
- Forms allow correction before submission
- Time limits avoidable or extendable
- Important content not dependent on memory of prior pages
Workflow
- Define scope. Full site? Specific flows? Specific page templates?
- Run automated scans. Document findings per page.
- Manual keyboard pass. Test all priority flows.
- Screen reader pass. Test with at least one combination.
- Visual checks. Contrast, zoom, color blindness, motion.
- Cognitive checks. Reading level, error handling, time limits.
- Score against WCAG. Per success criterion (level A, AA, AAA).
- Prioritize findings. Critical (blocks users), Important (degrades experience), Minor (polish).
- Write the report. Use the template in
references/audit-report-template.md. - Build a remediation plan. Sequenced fixes with effort and impact estimates.
Severity classification
For prioritization:
Critical (P0):
- Blocks an entire user flow for an assistive-tech user
- Renders a key page completely inaccessible
- Examples: form with no labels, modal without focus management, primary CTA not keyboard-accessible
Important (P1):
- Significantly degrades the experience for assistive-tech users
- Examples: missing alt text on key images, low-contrast body text, error messages that don't announce
Minor (P2):
- Affects edge cases or specific assistive technology combinations
- Examples: minor focus order issues, missing decorative alt attributes, edge case keyboard handling
Polish (P3):
- Above-AA improvements that benefit accessibility but aren't compliance-blocking
- Examples: AAA contrast targets, additional reduced-motion variants, language attributes on inline foreign words
Failure patterns
- Automated scan only. Catches 30 to 50 percent of issues. The remaining 50 to 70 percent are in keyboard, screen reader, and cognitive testing.
- Testing only on the home page. The home page is usually the most accessible. Bugs hide in deeper flows.
- Treating accessibility as a one-time project. Accessibility erodes with every deploy. Bake it into the development cycle.
- Fixing without root cause. Patching individual issues without understanding why they happened means new ones keep appearing.
- Ignoring screen reader testing. Hard to do well, easy to skip. Single biggest source of "we thought we were accessible" surprises.
- Confusing AA and AAA. AAA is rarely the right target. AA is the practical baseline for most products.
- Treating accessibility as a designer or developer responsibility alone. Content, product, QA, and leadership all need to participate.
- Assuming compliance equals accessibility. WCAG conformance is a floor, not a ceiling. Real users may still struggle.
Output format
Default output is a comprehensive audit report at accessibility-audit.md.
Structure:
- Executive summary
- Methodology (tools used, pages tested, screen readers used)
- Findings by WCAG principle
- Critical findings (P0) with specific URLs and fixes
- Important findings (P1)
- Minor findings (P2)
- Polish (P3)
- Remediation roadmap (sequenced and prioritized)
- Appendices (full automated scan results, keyboard navigation notes, screen reader notes)
Plus a remediation tracking spreadsheet with one row per finding.
If required data is unavailable
This skill's output depends on data, measurements, or tool results it cannot generate on its own. When a required input, tool, or data source is unavailable or unverifiable, the sanctioned output is the deliverable with the gap stated: what was needed, what was actually obtained or verified, and which parts of the output are affected. Fabricating, estimating, or interpolating a required number to complete the deliverable is never sanctioned. A stated gap is a complete answer.
Reference files
references/audit-report-template.md- Full audit report template.references/wcag-quick-reference.md- Condensed WCAG 2.1 AA criteria with audit checks.references/aria-patterns.md- Decision-grade ARIA patterns. Semantic-HTML-first principle, common interactive widgets (accordion, tabs, modal, toggle, disclosure, navigation), live regions, hiding patterns, labeling, state indicators, anti-patterns.
GitHub 仓库
常见问题
什么是 accessibility-audit Skill？
accessibility-audit 是一个 Claude Skill，作者为 rampstackco。Skill 将 Claude 按需加载的说明和资源打包，让 Claude 无需额外提示即可执行与 accessibility-audit 相关的任务。
如何安装 accessibility-audit？
使用本页的安装命令：将 accessibility-audit 作为插件添加到 Claude Code，或将其仓库克隆到 skills 目录，然后重启 Claude 以加载该 Skill。
accessibility-audit 属于哪个分类？
accessibility-audit 属于元分类。
accessibility-audit 可以免费使用吗？
可以。accessibility-audit 已收录在 AIMCP，可免费安装。
相关推荐技能
Content Collections 是一个 TypeScript 优先的构建工具，可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计，提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。
这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持，涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流，可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。
该Skill帮助开发者创建OpenCode插件，用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式，适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。
SGLang是一个专为LLM设计的高性能推理框架，特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术，在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时，能实现极速生成。如果你正在构建智能体或多轮对话系统，并追求远超vLLM的推理性能，SGLang是理想选择。