Skip to content

solution 15 agents

github-actions[bot] edited this page May 5, 2026 · 1 revision

Solution Reference: Challenge 15 -- Discover Accessibility Agents

This shows example agent discovery notes.

Example: Three agents examined

Agent 1: accessibility-lead

Location: .github/agents/accessibility-lead.agent.md

Purpose: Coordinates accessibility reviews across the team. Acts as an orchestrator that delegates to specialist agents.

Key observation: This agent does not do the accessibility checking itself -- it routes requests to the right specialist (ARIA, keyboard, contrast, etc.). This is a design pattern called orchestration.

Agent 2: aria-specialist

Location: .github/agents/aria-specialist.agent.md

Purpose: Reviews ARIA (Accessible Rich Internet Applications) attributes in HTML and web components. Checks that interactive elements have correct roles, states, and properties.

Key observation: The agent instructions reference specific ARIA patterns from the W3C specification. Custom agents can encode domain expertise.

Agent 3: keyboard-navigator

Location: .github/agents/keyboard-navigator.agent.md

Purpose: Ensures that all interactive elements are operable by keyboard. Checks tab order, focus management, and keyboard shortcuts.

Key observation: The guardrails section says it cannot test actual keyboard behavior -- it can only review code. Testing still requires a human with a keyboard.

Example: Running an agent

What I asked: "@accessibility-lead review the heading structure of this file"

What it did: The lead agent delegated to the alt-text-headings specialist, which analyzed the heading levels and reported that the file skipped from H2 to H4.

What I learned: Agents can chain together. The lead interpreted my request and chose the right specialist. I did not need to know which agent handles headings.

Example: Reading agent instructions

Looking at the .agent.md file structure:

  • YAML frontmatter: Contains the agent name, description, and tool restrictions
  • Responsibilities section: Lists what the agent is designed to do
  • Guardrails section: Lists what the agent should NOT do or cannot do reliably

What matters

The learning objective is understanding that AI agents are configurable tools with defined boundaries. If you examined at least one agent's instructions and can explain what it does and what it will not do, you completed this challenge.

Clone this wiki locally