tokensagents .COM home
ADA Conformance Pack

Fix accessibility in the code, not the widget.

Two installable agent skills that audit and fix WCAG issues in your own repo, plus sourced research that tracks the rules as they move. First-class in Claude, portable to any agent that reads Markdown. Nothing leaves your codebase.

Annual license. Web, iOS, Android, desktop, enterprise internal tools. If it renders a UI, the rules apply.

Why AI-built code fails WCAG

3,117

Federal website accessibility lawsuits filed in 2025, up 27 percent from 2024. About one in five of the top 500 e-commerce retailers was sued in that year alone. (Seyfarth, UsableNet.)

72%

Engineering leaders at 30 million dollar-plus orgs who say AI-assisted development raised their compliance risk. Roughly 95 percent of AI models emit critical accessibility issues in their default output. (Deque, July 2026.)

2

Times the federal accessibility rules moved in 2026 alone: DOJ in April, HHS in May. The ground under the deadline shifts, so a one-time audit is stale the day it lands.

Inside the two accessibility skills

Two skills, because accessibility has two jobs: how a thing looks and how a thing is built. Plus one research system both skills check their findings against.

ada-visual

The design layer

Contrast minimums and the AAA maximums, color independence, typography and reflow, target size, focus visibility, reduced motion, dark mode, and forced colors. It verifies real contrast ratios and can enforce them through your design tokens, so a failing color pair cannot be assembled by hand.

ada-aria

The web code layer

Semantic HTML first, ARIA only where HTML cannot express the need. Keyboard interaction, focus management, live regions, forms, and screen reader support. It favors native elements and warns you where automated tools stop, because they catch only 30 to 40 percent of real issues.

The research system

Shared, sourced, dated

Conformance levels A, AA, and AAA, plus a regulatory profile mapped to the DOJ Title II baseline. Testing methodology, mobile and enterprise platform layers, VPAT and ACR conformance statements, and one graded severity scale: Blocker, Critical, Major, Minor. Every source is dated 2025 to 2026.

How the WCAG audit runs in your repo

Install the skills, point one at a repo or a pull request, and get findings rated by severity with the fix in place. It works downstream of any scanner: your scanner finds the issue, this ships the fix.

Sample review output
> run ada-aria on src/components/Dialog.tsx

Dialog.tsx, web, AA. Blocker 1, Critical 0, Major 1, Minor 0.
Tested: keyboard and DOM review, no screen reader smoke test.

1. Close control is a div with an onclick, no keyboard support
   Criterion: WCAG 2.1.1 Keyboard, A (APG Dialog)
   Severity: Blocker
   Fix: use a native button, close on Escape, return focus to
        the element that opened the dialog

2. Icon-only close button has no accessible name
   Criterion: WCAG 4.1.2 Name, Role, Value, A
   Severity: Major
   Fix: add aria-label="Close" to the button

Severity-rated per criterion, so findings translate into the Supports, Partially Supports, and Does Not Support rows a VPAT asks for when procurement wants one.

What a report looks like

This site runs behind an accessibility gate that fails on any violation, across 8 pages and 2 viewports. Here is its last run, published in full, numbers generated by the run itself rather than typed in by us.

tokenstoagents.com, WCAG 2.2 AA 2026-07-31
0 violations
344 checks passed
16 flagged for human review
34 rules exercised

That third number is the one that matters. A tool that reports zero problems and zero uncertainties is lying to you. Automated rules catch 30 to 40 percent of real accessibility issues. The other 60 gets flagged for a human, because deciding whether a heading actually describes the section under it is not something a rules engine can do.

Read the full report. Every finding names the rule, the level, a severity from Blocker to Minor, the exact element, and the fix, mapped per criterion so it translates into the Supports and Partially Supports rows a VPAT asks for.

ADA, Section 504, and EAA deadlines

Current as of July 29, 2026

Current US and EU accessibility deadlines and the standard each names, as of July 29, 2026.
Rule and who it covers Standard named Date
DOJ ADA Title II, public entities with population 50,000 or more WCAG 2.1 AA April 26, 2027
DOJ ADA Title II, public entities under 50,000 and all special districts WCAG 2.1 AA April 26, 2028
HHS Section 504, health programs with 15 or more employees WCAG 2.1 AA May 11, 2027
HHS Section 504, health programs under 15 employees WCAG 2.1 AA May 10, 2028
EU Accessibility Act, wherever your customers are EN 301 549 (WCAG 2.1 AA) In force since June 28, 2025

The original April 24, 2026 Title II deadline never took effect. DOJ extended both dates by one year on April 20, 2026 (Interim Final Rule, 91 FR 20902), and HHS extended its Section 504 dates in 2026 as well. On May 21, 2026 the National Federation of the Blind sued DOJ and HHS to vacate those extensions. That case is pending, so these dates are under active litigation and could move again in either direction. A new EN 301 549 version that imports WCAG 2.2 is expected.

The pack re-verifies annually because the ground does.

Pricing and annual license

Individual

$149 per year

  • Both skills, ada-visual and ada-aria
  • The full sourced research system
  • Every update while your license is active
  • License key, one seat
Get the individual license

Renewal ships updated skills whenever WCAG, EN 301 549, or the DOJ and HHS dates change, plus refreshed date-stamped research and a changelog.

14-day money-back, no argument. Founding price, locked while your license stays active.

Prices in USD. Tax and VAT are handled at checkout by Polar, our merchant of record, who issues your receipt and invoice. You get two .skill files plus the research folder, downloaded the moment you pay.

One manual audit of a small site runs 3,000 to 7,000 dollars and goes stale the day it lands. The pack re-audits every review you run.

Enterprise

A custom rule profile for your design system, VPAT and ACR workflows your procurement team expects, named support with a 2-business-day SLA, and net-30 terms. Rolled out and governed at your scale.

For teams that need IT approval

  • It runs entirely in your environment. Nothing leaves your repo.
  • No telemetry, no vendor access to your code, no network calls of its own.
  • Plain-text skill files your IT team can read line by line.

The full security architecture, the enterprise pinning story, and the requirements are on the install and IT-approval page.

WCAG and ADA questions

What exactly do I get when I buy?

Two files and a research library, downloaded from Polar the moment you pay: ada-visual.skill and ada-aria.skill, each a folder of plain Markdown, plus the dated research the skills check their findings against. You also get a license key and every update while the license is active. No app to run, no account on our servers, nothing installed but files you can read.

Why is the license annual?

Because the ground moves. DOJ extended the Title II dates in April 2026, HHS extended its Section 504 dates weeks later, and the National Federation of the Blind is suing to undo both. Your renewal is a re-verification: updated skills and refreshed, date-stamped research whenever WCAG, EN 301 549, or the DOJ and HHS dates change. Current as of July 29, 2026, and under active litigation.

Does this make us compliant?

No, and anyone who tells you a tool can is selling you a lawsuit. The pack finds and fixes conformance issues in your code and rates them by severity. Compliance is a legal determination about your whole product and process, not a checkmark a skill hands you. This is not legal advice. Confirm your obligations with counsel.

What about accessibility overlay widgets?

An overlay is a script you paste in that promises instant compliance and does not deliver it. Sites running overlays drew hundreds of accessibility lawsuits in 2025, and the FTC's one million dollar order against an overlay vendor for deceptive compliance claims became final in April 2025. The pack does the opposite: it changes the code in your repo, where the problems live, and never touches your production site.

Do I need Claude to run this?

No. The skills are open-standard Markdown that any agent can read. Claude is the first-class path: it fires them on matching work without being asked, and it is what we test against every release. In another agent you point its rules or context at the skill folder and include the rules with the work. The install page walks both paths.

Does this work with my framework?

Yes. The rules are about accessibility, not your stack: contrast ratios, keyboard contracts, semantic structure, target sizes. Contrast math does not care which framework computed the color. React, Swift, Kotlin, Flutter, or plain HTML, the rules are the same.

How is this different from axe or Lighthouse?

Those are scanners. They tell you a contrast ratio failed on line 40. The pack works downstream of that: it reads the finding, decides what the fix should be in your system, and writes it. Automated scanners also catch only 30 to 40 percent of real issues, because most accessibility failures are judgment calls a rules engine cannot make: is this the right heading level, does this button name describe what it actually does, does this focus order match the reading order. Keep your scanner. This handles the part it cannot.

Which platforms does it cover?

Web, iOS, Android, desktop, and enterprise internal tools. The design layer translates every size and contrast rule into the right units per platform, CSS pixels, iOS points, Android dp, and the code layer routes native work to platform-specific modules instead of applying web ARIA rules where they do not belong.

Can I use it on client work?

Yes, on both tiers. An individual license covers one named person on any project they work on, including client projects. A team license covers up to 10 named people in one organization across unlimited client projects. What neither license allows is reselling or redistributing the pack itself as a product.

How does the team license count seats and client sites?

The team license covers up to 10 named people inside one organization, and unlimited client projects. If you are an agency, one team license covers every client site your seats work on. Resale or redistribution of the pack itself is not included.

What happens when a license lapses?

Updates and downloads stop. The skill files you already installed keep working, because they run locally in your own environment. You just stop getting the refreshed research and the rule-change updates that are the whole reason it is annual.

Will my security team approve this?

It is the easiest review they will do all quarter. The pack is static plain-text files that run inside the AI environment you already approved. It has no server, makes no network calls of its own, sends no telemetry, and gives us no access to your code. Your team can read every rule before installing it. The install page has the full architecture statement to forward them.

What is your refund policy?

14 days, money back. If the pack is not what you needed, email us within 14 days of purchase for a full refund. You keep anything you already fixed with it. Telling us why is optional, but it helps us improve the pack.

How do I buy for a team that needs a PO or an invoice?

Checkout issues a receipt and invoice automatically through Polar, our merchant of record, which covers most team purchases. If you need a purchase order, net-30 terms, a W-9, or a quote on letterhead for a p-card file, that is the enterprise tier. Tell us what your procurement process needs and we will work to it.

The rules moved twice this year. Ship fixes, not promises.

Get the pack