What is Accessibility Agents?
Accessibility Agents is a Community-Access package of 113 Agent Skills for WCAG 2.2 AA work in AI coding tools. Six router skills dispatch specialists for web UI, documents, markdown and GitHub tasks, a hook blocks edits to user-facing files until review, and an MCP server supplies scanners.
Accessibility Agents is a single package of Agent Skills for WCAG 2.2 AA enforcement while an AI coding tool writes code. The README says models forget accessibility while generating, and this package makes that harder instead of relying on memory.
It works as a router. Six entry skills take the request and dispatch 107 further specialist, helper and reference skills by name. The README says those stay invisible to the model until named, so the package costs about 1,100 tokens up front.
What changes in your agent's output
- Edits to user-facing files such as HTML, JSX, TSX, Vue, Svelte, CSS and server templates go through the accessibility lead first.
- A hook refuses a write to those files until that review completes, on clients that support hooks.
- Reviews cover ARIA, keyboard, focus, forms, contrast, modals, live regions, headings, tables, links and cognitive load.
- Audit reports are rendered by a script from JSON findings, with a score and the change since the previous run.
Who it is for
Best for
- Teams that ship UI through an AI coding tool and want a review step that cannot be skipped by the model.
- Projects with documents as well as pages, since the package also scans Word, Excel, PowerPoint, PDF and ePub files.
- People who prefer one package across Claude Code, Copilot, Codex and Gemini CLI to separate copies per client.
Not for
- A light single-skill setup. This is 113 skills plus scanners, so see web-accessibility-skill or fixing-accessibility for one focused file.
- ChatGPT users who expect scanners or the edit gate. The README says that plugin is the skills alone.
- Codex users who expect the edit gate from the plugin, because the README says Codex no longer loads plugin hooks.
Use it with UiChemy
Pages that UiChemy produces are editable WordPress markup, so the web scanners can load the published URL and report issues to fix in the builder. Click Get Started on uichemy.com.
How to install Accessibility Agents: WCAG 2.2 AA Skills for AI Coding Tools
/plugin marketplace add Community-Access/accessibility-agentsOptional scanners need Playwright, axe-core for Playwright, pdf-lib and the veraPDF CLI, which the tools say when missing.Claude Code
Run the marketplace command above, then
/plugin install accessibility-agents@community-access.Codex
Run
codex plugin marketplace add Community-Access/accessibility-agentsand thencodex plugin add accessibility-agents@community-access. The README says this gives skills and the MCP server only, with no edit gate.Copilot CLI, VS Code and Gemini CLI
Copilot CLI has matching marketplace and plugin commands, VS Code installs from the repository URL, and Gemini CLI uses
gemini extensions installwith the repository URL.From a checkout
Run
npm install, thennpm run setup -- --dry-run, thennpm run setup. It copies skills to the personal agents folder and never overwrites a hooks file or skill you edited.
Check that it loaded
Run npm run verify in a checkout. The README describes sixteen gates, and two are optional.
Key features
- Six router skillsaccessibility-lead for UI code, web-accessibility-wizard for full audits, document-accessibility-wizard, markdown-a11y-assistant, developer-hub and github-hub.
- 113 skills in one packageSpecialists, helpers and kb reference skills live in one folder tree, read natively by several clients with no per-client copies.
- Edit gateA guard script blocks writes to user-facing files until the accessibility lead has completed, where the client supports hooks.
- Small always-on costThe AGENTS.md contract stays under 1,200 tokens, and the README puts the whole package at about 1,100 tokens before you speak.
- Web UI review coverageARIA, keyboard, focus, forms, contrast, modals, live regions, headings, tables, links, internationalisation, mobile, email and data visualisation.
- Document scanningOffice, PDF and ePub checks for one file or a whole tree, with generated remediation scripts.
- Markdown auditsLink text, alt text, heading order, tables, emoji and diagrams in documentation files.
- Reports from structured findingsFindings are JSON, and a script writes the report, computes the score and shows the delta against the previous run.
- MCP scannersThirty-nine tools cover axe-core, Office, PDF, ePub, markdown, veraPDF, Playwright behavioural scans, audit history and report rendering.
- Local by designScanning runs on your machine with no telemetry or account. The five GLOW tools are the exception and send files you name to a separate service.
Frequently asked questions
What is Accessibility Agents?
It is an MIT licensed package from Community-Access that adds WCAG 2.2 AA review to AI coding tools. It ships 113 Agent Skills, an edit gate hook, a thirty-nine tool MCP server and a report renderer, and one package is read by several coding clients.
Which AI coding tools does it support?
The README names Claude Code, Codex, GitHub Copilot, Gemini CLI, Antigravity and ChatGPT. Support differs by client: hooks run where supported, Codex gets skills and the MCP server only, and the ChatGPT plugin is skills alone.
How do I install Accessibility Agents?
Use the plugin command for your client, for example the marketplace add and plugin install pair on Claude Code. Without a plugin route, clone the repository, run npm install and then npm run setup, which copies skills to the personal agents folder.
What does the edit gate do?
A hook refuses a write to a user-facing file until the accessibility lead skill has finished reviewing. It works on clients that support hooks. The README notes Codex no longer loads plugin hooks, so Codex users run the setup script to get the gate.
Does it send my code or files anywhere?
Scanning runs locally, with no telemetry, account or key. Outbound traffic happens only for URLs you give the web scanners and for the five GLOW tools, which send a request, and the named file for document tools, to a separate GLOW service.
How many tokens does it use?
The README says the always-on AGENTS.md contract is under 1,200 tokens, and the whole package costs about 1,100 tokens before you say anything, because the 107 specialist skills stay hidden until a router names one.
Does it guarantee WCAG conformance?
The repository does not claim that. It makes accessibility review harder to skip and produces scored reports. Automated scans find only some issues, so keep a manual keyboard and screen reader pass before you call a page accessible.
What is the difference between this and a single accessibility skill?
It is a router with specialists, an edit gate and scanners rather than one SKILL.md. A single file like accessibility-audit is lighter, while this package suits teams that want enforcement and document coverage.


