Agent Ecosystem

A11YSmith ships 57 accessibility specialist agents organized into 6 teams. Agents can be distributed to 5 AI coding platforms from a single source of truth.

Overview

The agent ecosystem wraps A11YSmith CLI commands for automated scanning while carrying specialist knowledge for manual guidance that automation cannot cover. Agents are coordinated through a hook enforcement system that ensures accessibility review happens before UI code is merged.

How It Works

  1. Run a11ysmith agent-init to create a .a11ysmith-agents.json configuration file.
  2. Run a11ysmith agent-generate --platform <target> to generate agent files for your AI coding platform.
  3. Agents activate automatically when you work on UI code, running accessibility checks and providing specialist guidance.

Teams

Agents are organized into 6 specialist teams, each with a lead agent that coordinates the team.

Agent teams and their descriptions
Team Lead Agent Description
Web Accessibility Audit accessibility-lead Full web accessibility audit with specialist agents for ARIA, keyboard, forms, contrast, tables, links, modals, live regions, alt text, headings, cognitive, and design system tokens.
Document Accessibility Audit document-accessibility-wizard Document accessibility audit for Word, Excel, PowerPoint, PDF, and ePub files with format-specific sub-agents.
Markdown Accessibility Audit markdown-a11y-assistant Markdown accessibility audit covering links, alt text, headings, tables, emoji, diagrams, dashes, and anchors.
GitHub Workflow github-hub GitHub workflow agents for issue tracking, PR review, daily briefings, analytics, repository management, and community contributions.
Mobile & Desktop Accessibility mobile-accessibility Mobile and desktop accessibility agents for React Native, iOS, Android, wxPython, NVDA addon development, and desktop platform APIs.
Developer Hub & Supporting developer-hub Developer tools hub routing to Python, wxPython, NVDA, desktop accessibility, and tool-building specialists. Cross-team coordination via nexus.

Supported Platforms

Supported AI coding platforms and their output formats
Platform Output
Claude Code.claude/agents/*.md + hooks + slash commands
GitHub Copilot.github/agents/*.agent.md + skills + instructions
Gemini.gemini/extensions/ skills + GEMINI.md
Claude DesktopMCP server with 8 tools + manifest
VS CodeChat participant extension with agent routing

Hook Enforcement

Three hooks enforce accessibility review in your development workflow:

UserPromptSubmit
Detects web project context and injects the accessibility-lead delegation on every prompt.
PreToolUse (Edit Gate)
Blocks edits to UI files until the accessibility-lead has reviewed. Uses permissionDecision: "deny".
PostToolUse (Session Marker)
Creates a session marker and unlocks the edit gate after the accessibility-lead completes its review.

Enforcement is configurable via .a11ysmith-agents.json: strict (block edits), advisory (warn only), or off.

CLI Commands

Initialize agent configuration

a11ysmith agent-init

Generate agent files for a platform

a11ysmith agent-generate --platform claude-code

Supported platforms: all, claude-code, github-copilot, gemini, claude-desktop, vscode

List available agents

a11ysmith agent-list --team web

Filter by --team, --command, or --search. Add --all to include hidden agents.

Browse Agents

Use the dashboard to explore agents interactively: