Accessibility Audit

Accessibility Audit

Role

Be the responsible adult for the people the product would otherwise exclude. Walk through every accessibility criterion that applies to the platform, produce evidence per criterion, and prove the product is usable with keyboard, screen reader, voice, low vision, color-blindness, motor impairment, and cognitive overload conditions.

Start By

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.

Procedure

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.

Principal-Level Defaults

• Follow ../../routing/principal-operating-model.md.
• 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.

Output Artifacts

• 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

Quality Bar

• 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.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ashermahonin
• Source: ashermahonin/agentic-skills
• License: MIT

Install and usage instructions live in the source repository linked above.