AI agent guidance for this repository. Generated for AgentRC compatibility.
This repository contains 80 accessibility-focused AI agents across multiple platforms (Copilot, Claude Code, Codex, Gemini). This file provides high-level guidance for AI coding assistants working in this codebase.
Repository-wide hard rule: do not add emoji characters in generated, edited, or reviewed content.
- Applies to release notes, changelog entries, docs, prompts, instructions, agent files, issue/PR text, and chat-generated copy intended for repository publication.
- Replace emoji with plain text labels.
- If existing content contains emoji and you are touching that content, remove emoji as part of the update.
A11y Agent Team is a collection of accessibility-focused AI agents that enforce WCAG 2.2 AA standards across web, document, and mobile development. The agents work together as specialized teams to catch accessibility issues that LLMs typically miss during code generation.
| Path | Purpose |
|---|---|
.github/agents/ |
Copilot agent definitions (80 agents) |
.github/skills/ |
Copilot reusable skills (25 skills) |
.github/prompts/ |
One-click workflow prompts |
.github/instructions/ |
Always-on instruction files |
.claude/agents/ |
Claude Code agent definitions |
.codex/ |
Codex CLI configuration (11 roles) |
.gemini/ |
Gemini CLI extension |
docs/ |
Documentation site |
mcp-server/ |
HTTP-based MCP server (Streamable HTTP + stdio) |
vscode-extension/ |
VS Code extension (planned) |
scripts/ |
Build and validation scripts |
templates/ |
Scan configuration templates |
Led by accessibility-lead, coordinates specialists for comprehensive web audits:
aria-specialist- ARIA roles, states, propertieskeyboard-navigator- Tab order, focus managementcontrast-master- Color contrast, visual accessibilityforms-specialist- Form labeling, validation, errorsmodal-specialist- Dialog focus trapping, escape behaviorlive-region-controller- Dynamic content announcementsalt-text-headings- Images, SVGs, heading hierarchytables-data-specialist- Data table accessibilitylink-checker- Ambiguous link text detectiontext-quality-reviewer- Non-visual text quality reviewi18n-accessibility- Internationalization, RTL, and multilingual accessibility
Led by document-accessibility-wizard, handles Office and PDF audits:
word-accessibility- Microsoft Word (.docx)excel-accessibility- Microsoft Excel (.xlsx)powerpoint-accessibility- Microsoft PowerPoint (.pptx)pdf-accessibility- PDF/UA conformanceepub-accessibility- ePub accessibilitypdf-remediator- Programmatic and manual PDF remediationoffice-remediator- Programmatic Office document (Word/Excel/PowerPoint) remediation
Led by github-hub / nexus, manages repository operations:
daily-briefing- Morning overview of issues, PRs, CI statuspr-review- Code review with accessibility focusissue-tracker- Issue triage and priority scoringanalytics- Repository health metricsrepo-admin- Collaborator and branch protection managementteam-manager- Organization team membershipcontributions-hub- Contributor activity tracking and recognitioninsiders-a11y-tracker- VS Code Insiders accessibility regression trackingtemplate-builder- Issue and PR template generationrepo-manager- Repository settings, labels, and workflow managementprojects-manager- GitHub Projects v2 boards, views, custom fields, and iterationsactions-manager- GitHub Actions workflow runs, logs, re-runs, and CI debuggingsecurity-dashboard- Dependabot, code scanning, and secret scanning alert triagerelease-manager- Releases, tags, assets, and release note generationnotifications-manager- Notification inbox management, filtering, and subscriptionswiki-manager- Wiki page creation, editing, search, and organization
Led by developer-hub, handles desktop and Python development:
python-specialist- Python debugging, packaging, testingwxpython-specialist- wxPython GUI developmentnvda-addon-specialist- NVDA screen reader addon developmentdesktop-a11y-specialist- Desktop accessibility APIsa11y-tool-builder- Building accessibility scanning tools
Specialist agents for CI pipelines, standards education, and screen reader simulation:
ci-accessibility- CI/CD accessibility pipeline setup and managementscreen-reader-lab- Interactive screen reader simulation for educationwcag3-preview- WCAG 3.0 draft education and transition planningwcag-aaa- WCAG AAA conformance auditing
Copilot agents (.github/agents/*.agent.md):
---
name: Agent Name
description: What this agent does (required)
tools: ['read', 'edit', 'search', 'runInTerminal', 'askQuestions']
model: ['Claude Sonnet 4.5 (copilot)', 'GPT-5 (copilot)']
---
Agent instructions in markdown...Claude Code agents (.claude/agents/*.md):
---
name: agent-name
description: What this agent does
tools:
- Read
- Edit
- Grep
- Task
---
Agent instructions in markdown...| Platform | Read | Edit | Search | Shell | Sub-agent |
|---|---|---|---|---|---|
| Copilot CLI | read |
edit |
search |
runInTerminal |
agent |
| Claude Code | Read |
Edit |
Grep/Glob |
Bash |
Task |
Skills must have SKILL.md with YAML frontmatter:
---
name: skill-name
description: What this skill provides
---
Skill content in markdown...No build step required - agents are markdown files.
Validation:
# Check agent YAML frontmatter
node scripts/validate-agents.js
# Run AgentRC readiness check
npx github:microsoft/agentrc readinessLocal testing:
# Install to local Copilot CLI
./install.sh --global --cli
# Verify agents load
copilot /agent| Workflow | Purpose |
|---|---|
a11y-check.yml |
Lint HTML/JSX/CSS for accessibility |
verify-sources.yml |
Validate URLs in documentation |
update-manifest.yml |
Generate installation manifest |
sync-docs-site.yml |
Deploy documentation |
- Agent changes go in both
.github/agents/AND.claude/agents/ - Use platform-specific tool names (see conventions above)
- Every agent needs a
descriptionfield - Run validation before committing
- Update CHANGELOG.md for user-facing changes
- Parallel agent definitions: Each platform has its own agent files because tool names and capabilities differ
- Skills for reusable knowledge: Common patterns (WCAG rules, scoring formulas) are in skills, not duplicated in agents
- Wizard orchestrators: Complex workflows use wizard agents that delegate to specialists
- Read-only scanners: Scanner agents never modify files - they only report findings
- Fixer agents require confirmation: Agents that modify code always ask before applying changes
- axe-core: Web accessibility rule engine
- Playwright: Browser automation for behavioral testing
- markdownlint: Markdown accessibility linting
- veraPDF (planned): PDF/UA validation