# Accessibility Audit

> Audit a user-facing web, mobile, desktop, game, voice, or conversational surface against the accessibility requirements of its target platform. Use during design review and before release. Produce criterion-level evidence from automated and manual checks, assistive-technology results, unresolved barriers, remediation owners, and release impact.

- Skill: `ashermahonin/accessibility-audit` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add ashermahonin/accessibility-audit`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ashermahonin/accessibility-audit/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: ashermahonin (https://skillmd.com/u/ashermahonin)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/ashermahonin/accessibility-audit

---


# Accessibility Audit

## Purpose

Evaluate the product against the accessibility requirements of its target platforms. Record evidence per criterion and cover keyboard, screen reader, voice, low-vision, colour-perception, motor, and cognitive use where they apply.

## Product context

1. Read `references/a11y-checklist.md`.
2. Pull the platform matrix from `platform-detector` and UX specs from `ux-design`.
3. Choose the applicable standard per platform (WCAG 2.2 AA for web; platform-native a11y APIs for iOS/Android/desktop; game-readability patterns for games).
4. Use Context7 MCP for current WCAG wording, current Apple/Google/Microsoft a11y guidance, and current accessibility-testing tool docs.

## Design and validation

1. Run automated audits per platform: web (axe-core, Lighthouse, Pa11y), iOS (Accessibility Inspector), Android (Accessibility Scanner), desktop (platform a11y tools).
2. Run manual checks: keyboard-only traversal, screen-reader transcript (VoiceOver, TalkBack, NVDA, JAWS, Narrator), focus order, contrast measurement, reduced-motion behavior, color-blind palette simulation.
3. Per criterion, mark Pass / Concern / Fail / Out-of-scope. Record evidence path, repro steps, suggested fix, owner.
4. Cross-check `ux-design` screen states: empty/loading/error states must also be accessible.
5. Cross-check `i18n-localization`: RTL languages, font glyph coverage, label expansion ratios.
6. Produce a remediation plan ordered by impact × ease.
7. Issue a release-gate verdict: Go / Conditional / Hold.

## Constraints

- Use Context7 MCP for current WCAG, platform-native a11y, and tool documentation.
- Keep a decision trace: standard chosen per platform, tools used, evidence per criterion, accepted-risk register with named approver and expiry.
- Refuse to mark a criterion Pass from an automated tool alone when manual verification is required (e.g., screen-reader behavior).
- Escalate any "Hold" verdict before release.

## Specification and evidence

- Per-criterion status table (Pass / Concern / Fail / Out-of-scope) with evidence
- Screen-reader transcripts per platform
- Contrast measurements per surface
- Keyboard traversal map
- Findings register: severity, repro, fix, owner, verification test
- Remediation plan ordered by impact × ease
- Release-gate verdict

## Acceptance checks

- No criterion marked Pass without evidence.
- No screen-reader Pass without a transcript.
- No "we'll fix in v2" for AA-level findings without a named approver and an expiry.
- No game surface approved without a remappable-controls confirmation.
- No multi-language product approved without an RTL pass when any RTL language is in scope.

## Handoff

Hand off to `service-implementation` per finding and to `qa-eval` to add a11y regression cases.

## References

- `references/a11y-checklist.md`: per-platform criteria, recommended tools, manual-test scripts, screen-reader transcripts template.