Agent remediation workflow
a11y-skill
A WCAG 2.2 AA accessibility audit and remediation skill for AI coding agents. It turns a11y-engine scans into a controlled, evidence-based fix workflow with approval gates and verification re-audits.
a11y-skill is the agent-facing workflow layer of the a11y-engine ecosystem. It defines when to run audits, how to discover pages, how to normalize URLs, how to present findings, and how to move from scan results to safe remediation without improvising from generic WCAG knowledge.
The skill is intentionally strict: remediation.md is treated as the source of truth, reports are generated only when requested, new files require explicit approval, style changes require exact diffs before applying, and every fix cycle must be followed by a verification re-audit.
It supports multiple coding agents, including Codex, Claude Code, Cursor, Gemini CLI, Antigravity, and Windsurf. The repo also includes evaluation scenarios, quality gates, troubleshooting references, report standards, and integration tests around the CLI and engine behavior.
What it does
- Guides AI agents through page discovery, audit execution, findings presentation, remediation, verification, and final delivery.
- Uses a11y-engine for multi-engine WCAG scans through axe-core, Chrome DevTools Protocol checks, and pa11y.
- Requires evidence-first fixes based on the generated remediation guide instead of generic accessibility assumptions.
- Adds hard approval gates for creating files and for style changes that affect visual design.
- Defines quality gates for pre-audit setup, post-audit integrity, fix safety, and re-audit deltas.
- Ships references for CLI usage, report standards, troubleshooting, and evaluations for common audit scenarios.