building-accessible-ui

作者: microsoft

必须用于任何 UI 工作。在生成、修改或审查任何渲染、样式化或连接用户界面的代码之前,请调用此技能——…

npx skills add https://github.com/microsoft/a11y-llm-eval --skill building-accessible-ui

building-accessible-ui

Checklist for producing and reviewing accessible UIs. Each rule leads with the platform-agnostic principle and, where relevant, the Web (HTML + ARIA + CSS) implementation. Apply the web guidance only when the output is web.

Detailed rationale lives in references/; widget-specific guidance in components/. Open a file only when it's relevant to the current task. Do not preload. Every file opened and every line a tool prints stays in context — don't re-read.

Accessibility constitution

Ground rules. Use them to resolve conflicts and decide how much custom work is justified. The checklist below is their mechanical application.

1. Accessibility is a core outcome

A UI inaccessible to realistic users is not "done". Treat accessibility as a first-class criterion alongside correctness, performance, and security — not a finishing step. When scope must be cut, record the gap explicitly. Never claim output is "fully accessible"; state what was addressed and known limitations.

2. Build for real people

Evaluate designs against these personas; if a decision breaks one, justify it and offer an alternative:

  • Screen reader — landmarks, headings, accessible name/role/state, reading order.
  • Keyboard-only — Tab/arrows/Enter/Space/Escape, visible focus, no traps.
  • Low-vision — zoom, reflow, contrast, Forced Colors.
  • Cognitive — plain language, clear labels, actionable errors, forgiving interactions.
  • Deaf / hard of hearing — captions/transcripts; no sound-only cues.
  • Motor / voice / switch — large hit targets, named controls, no precise/timed gestures.
  • Situational — sunlight, one-handed, noisy, flaky network.

3. Implementation priority

Use the highest option that fits:

  1. Existing accessible component in this codebase / design system.
  2. A component library.
  3. Native platform semantics (<button>, <a href>, <input>, <label>, <fieldset>/<legend>, <dialog>, <details>, <nav>, <main>, headings).
  4. Native element + minimum necessary ARIA (aria-describedby, aria-expanded, aria-current, etc.).
  5. Fully custom ARIA widget — only when nothing above fits, and only if you implement the APG keyboard, focus, and state behavior end-to-end.

No ARIA is better than bad ARIA. Don't duplicate native semantics (no role="button" on <button>). Don't use role="menu" for site navigation. Don't invent new patterns when a standard one exists.

4. Balance, don't trade away

Accessibility, performance, security/privacy, and visual design are joint constraints — not dials to trade off. If an optimization removes a label, breaks focus, or hides content from AT, redesign the optimization. Accessible names must not leak secrets, but security is not a reason to ship an unlabeled control — find a labeling approach that doesn't leak data. Visual polish doesn't justify removing focus indicators or semantic structure. Under schedule pressure, prefer cutting scope over shipping an inaccessible feature. When constraints genuinely conflict, surface it explicitly.

5. Respect existing code

Don't rewrite an existing component or shared utility just because it could be more accessible — other code depends on it. When you see issues outside the current task's scope: note them (issue, affected persona, suggested fix) and ask before changing. Fix in place only when the change is required by the task, localized, and low-risk. Inside scope, fix real issues; never silently remove existing affordances (labels, landmarks, focus management, live regions) without an equal-or-better replacement.

How to use this checklist

Identify which components the request involves (form, checkbox group, radio group, disclosure, modal, full view, etc.) and open the matching components/<name>.md once. Then work the checklist below. Open a references/*.md only when an item is unclear or you need the concrete fix pattern.

Do not claim the output is "fully accessible". State what was addressed and known limitations.

Do NOT use this skill for: backend-only changes, data migrations, build/CI configuration, non-UI tests, or tasks that do not touch the UI layer.

Checklist

  • Prefer existing components. If available, reuse existing UI components rather than creating new ones from scratch or custom implementations.
  • Platform-native semantics. Prefer native platform controls and structures over custom constructs; add accessibility overrides only when a native control genuinely can't be used. → references/structure.md.
    • Web: Prefer semantic HTML (<button>, <a>, <input>, <label>, <fieldset>/<legend>, <nav>, <main>, <header>, <footer>, <h1><h6>) over <div>/<span> with ARIA. Use ARIA only when no native element fits.
  • Regions / landmarks. View structure is exposed via semantic regions/landmarks; duplicated landmarks have unique accessible names.
    • Web: Exactly one <main>; <header>, <nav>, <footer> used when applicable.
  • Headings. Logical outline labels sections without skipping levels; one top-level heading per view. → references/structure.md.
    • Web: One <h1>, typically the first heading in <main>. Set a descriptive <title>.
  • Bypass blocks on web pages. Provide a mechanism to skip repeated navigation when delivering traditional web pages. (Not required for Electron or non-web surfaces.) → references/keyboard-focus.md.
    • Web: A "Skip to main content" link as the first focusable element
  • Name / role / value. Every interactive element exposes an accurate accessible name; role matches purpose; dynamic states (pressed, expanded, selected, checked, disabled, invalid) stay in sync with visuals.
    • Web: Prefer native attributes over ARIA. If necessary, use the minimum ARIA needed and update state attributes alongside DOM/visual changes.
  • Name-label match. The accessible name of each interactive element contains the visible label text.
    • Web: If aria-label is used, include the visible label text. For multiple controls that share a label (e.g., "Remove"), add context ("Remove item: Socks").
  • Labels and help text. Every form control has a programmatic label describing its purpose; help/error text is programmatically associated with its control. → components/forms.md.
    • Web: <label for> or wrapping <label>; never placeholder alone. Associate help/error via aria-describedby / aria-errormessage.
  • Grouping. Related options (checkboxes, radios) are grouped so their shared name is part of the accessible name of each option. Group-level help/error text is associated with the group itself — not with each option and not with an intermediate wrapper.
    • Web: <fieldset> with a <legend>. Put aria-describedby on the <fieldset> (not on a child <div>, and never on an extra <div role="group"> inside the fieldset — <fieldset> already is the group).
  • Required fields. Marked both visually and programmatically; not indicated by color alone.
    • Web: Use an asterisk to indicate required fields. Native required on the control or aria-required="true".
  • Keyboard operability. Every interactive element is keyboard operable; tab order matches reading/visual order; expected keys work (activation, arrow keys inside composite widgets, Escape closes overlays); no keyboard traps; static content is not sequentially focusable.
    • Web: Do not remove focus outlines without equal-or-better replacement. Use tabindex="-1" only for elements that need programmatic (not sequential) focus. → references/keyboard-focus.md.
  • Focus management. Focus is always visible. Overlays/dialogs/disclosures move focus appropriately and restore it on close; no focus traps outside modals.
  • Hidden content. Content hidden from assistive technology is not focusable and is hidden consistently across visual, semantic, and focus layers.
    • Web: hidden / display: none / aria-hidden="true" used consistently.
  • Graphics. Informative graphics have meaningful text alternatives; decorative graphics are hidden from AT. → references/images-graphics.md.
    • Web: <img> informative → alt; decorative → alt="". Informative <svg>role="img" + accessible name. Other decorative → aria-hidden="true".
  • Contrast. Text ≥ 4.5:1 (3:1 large); focus indicators and key boundaries ≥ 3:1. Never color-only cues. → references/contrast-forced-colors.md.
  • Respect OS accessibility settings. Never override OS high contrast, reduced-motion, or color-scheme preferences; adapt to forced-colors / high-contrast. → references/contrast-forced-colors.md.
  • Reflow. Content adapts to narrow viewports (target 320 CSS px) without two-dimensional scrolling for multi-line text; controls remain operable. → references/reflow.md.
  • Navigation. Uses semantic navigation grouping with state-exposing toggles for expandable menus. → references/navigation.md.
    • Web: <nav>, not role="menu"; aria-expanded on triggers.
  • Tables / grids. Static tabular data uses table semantics with header/cell associations; interactive grids only when truly warranted. → references/tables-grids.md.
  • Status messages. Provide status messages for dynamic content updates that are relevant to the user (loading indicators, form submission results, etc.). → references/status-messages.md
    • Web: Use aria-live="polite" or aria-live="assertive".
  • Testing. Add and run automated accessibility tests unless the project explicitly opts out. Writing or configuring a test is not enough — execution, fixes, and a result report are part of the deliverable. The final automated test run must be on the exact artifact you submit: any edit after a passing test invalidates that test, so re-run before submitting. Open references/testing.md before writing any test code for the opt-out signals, strategy precedence, runtime probe order, and reporting rules.
    • Web: Prefer @axe-core/* bindings that match the existing test runner; render the component/page fully so interactive state, focus, and live regions are evaluated.
    • Other platforms: Use the platform's native audit (Android AccessibilityChecks, iOS XCUIAccessibilityAudit, .NET AccessibilityInsights) under the same precedence.
  • Specs/Documentation. Follow the project's documentation pattern and document accessibility considerations for each view, component, and interaction. → references/specs-documentation.md.

来自 microsoft 的更多技能

oss-growth
microsoft
OSS增长黑客角色
agent-framework-azure-ai-py
microsoft
使用Microsoft Agent Framework Python SDK(agent-framework-azure-ai)构建Azure AI Foundry代理。在创建使用AzureAIAgentsProvider的持久化代理、使用托管工具(代码解释器、文件搜索、网络搜索)、集成MCP服务器、管理对话线程或实现流式响应时使用。涵盖函数工具、结构化输出和多工具代理。
development
airunway-aks-setup
microsoft
Set up AI Runway on AKS — from bare cluster to running model. Covers cluster verification, controller install, GPU assessment, provider setup, and first deployment. WHEN: "setup AI Runway", "onboard AKS cluster", "install AI Runway", "airunway setup", "deploy model to AKS", "GPU inference on AKS", "KAITO setup on AKS", "run LLM on AKS", "vLLM on AKS", "set up model serving on AKS", "AI Runway controller".
devops
appinsights-instrumentation
microsoft
使用Azure Application Insights对Web应用进行插桩的指南。提供遥测模式、SDK设置和配置参考。适用场景:如何对应用进行插桩、App Insights SDK、遥测模式、什么是App Insights、Application Insights指南、插桩示例、APM最佳实践。
devops
applicationinsights-web-ts
microsoft
使用Application Insights JavaScript SDK(@microsoft/applicationinsights-web)为浏览器/Web应用添加检测。用于真实用户监控(RUM)——页面视图、点击、AJAX/fetch依赖项、异常、自定义事件,以及与后端OpenTelemetry追踪关联的浏览器端GenAI代理追踪。涵盖SDK加载器脚本和npm设置、框架扩展(React、React Native、Angular)、点击分析、遥测初始化器,以及从浏览器发出的代理/工具/模型跨度所遵循的OTel GenAI语义约定。
devops
azure-ai-anomalydetector-java
microsoft
使用适用于 Java 的 Azure AI 异常检测器 SDK 构建异常检测应用程序。在实现单变量/多变量异常检测、时间序列分析或 AI 驱动的监控时使用。
development
azure-ai-language-conversations-py
microsoft
使用azure-ai-language-conversations Python SDK实现对话语言理解(CLU)。当使用ConversationAnalysisClient分析对话意图和实体、构建NLP功能或将语言理解集成到应用程序中时使用。
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 for Python。用于机器学习工作区、作业、模型、数据集、计算资源和管道。 触发词:“azure-ai-ml”、“MLClient”、“工作区”、“模型注册表”、“训练作业”、“数据集”。
development