Skip to content
Back to search

looporadata

MarketplaceWorks with Claude CodeGitHub

Claude Code plugins from LooporaData.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 20 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/LooporaData/claude-wcag-audit with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/LooporaData/claude-wcag-audit'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/looporadata--claude-wcag-audit/badge.svg)](/listings/looporadata--claude-wcag-audit)

HTML

<a href="/listings/looporadata--claude-wcag-audit"><img src="/api/listings/looporadata--claude-wcag-audit/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

claude-wcag-audit is a Claude Code plugin, and this repository is also its marketplace. In Claude Code:

/plugin marketplace add LooporaData/claude-wcag-audit
/plugin install claude-wcag-audit@looporadata

From a terminal, the same thing is:

claude plugin marketplace add LooporaData/claude-wcag-audit
claude plugin install claude-wcag-audit@looporadata

To offer it to everyone working in a project, commit this to the project's `.claude/settings.json`. Claude Code then prompts each person to install it when they trust the folder.

{
  "extraKnownMarketplaces": {
    "looporadata": { "source": { "source": "github", "repo": "LooporaData/claude-wcag-audit" } }
  },
  "enabledPlugins": { "claude-wcag-audit@looporadata": true }
}

It runs on macOS and Linux. Windows should work too, but hasn't been tested yet. You need Node.js 18 or later and npm. The first audit installs the pinned axe-core and Playwright into the plugin's data folder, which takes about a minute. The audit also needs Chromium. If Playwright's browser isn't installed, the script falls back to any Chromium already in the Playwright cache or to `CHROME_BIN`, and otherwise prints the command that installs it.

To update, run `/plugin marketplace update looporadata`. To remove it, run `/plugin uninstall claude-wcag-audit@looporadata`.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 15 lines
  1. {
  2. "name": "looporadata",
  3. "description": "Claude Code plugins from LooporaData.",
  4. "owner": {
  5. "name": "LooporaData",
  6. "url": "https://github.com/LooporaData"
  7. },
  8. "plugins": [
  9. {
  10. "name": "claude-wcag-audit",
  11. "source": "./",
  12. "description": "WCAG 2.1 and 2.2 A/AA accessibility audit with axe-core and scripted Playwright checks."
  13. }
  14. ]
  15. }
README.md · README · 186 lines
  1. # claude-wcag-audit
  2. A [Claude Code](https://claude.com/claude-code) plugin that audits a website against WCAG 2.1 and 2.2 Level A and AA. Ask Claude to audit a site and you get an HTML report with a result for every success criterion, the failing elements with fixes, and a checklist of what still needs a person.
  3. ![HTML report showing summary counts and the success criteria table](docs/report.png)
  4. ## Why
  5. Most automated accessibility checks run axe-core and stop there. axe is excellent at what it covers, but it only has rules for about 22 of the 50 WCAG 2.1 A/AA success criteria, and many of those only partially. It doesn't press Tab, zoom the page, open a menu or apply text spacing, so keyboard traps, invisible focus, clipped text at 200% zoom and menus that only open on click all pass silently.
  6. The plugin runs axe and then drives the page with Playwright the way a keyboard or low-vision user would. It's also explicit about coverage. Every criterion in the report is marked as failed, needing review, passed with good or partial coverage, manual only, or not applicable. That way a clean scan can't be mistaken for conformance.
  7. ## Install
  8. claude-wcag-audit is a Claude Code plugin, and this repository is also its marketplace. In Claude Code:
  9. ```
  10. /plugin marketplace add LooporaData/claude-wcag-audit
  11. /plugin install claude-wcag-audit@looporadata
  12. ```
  13. From a terminal, the same thing is:
  14. ```bash
  15. claude plugin marketplace add LooporaData/claude-wcag-audit
  16. claude plugin install claude-wcag-audit@looporadata
  17. ```
  18. To offer it to everyone working in a project, commit this to the project's `.claude/settings.json`. Claude Code then prompts each person to install it when they trust the folder.
  19. ```json
  20. {
  21. "extraKnownMarketplaces": {
  22. "looporadata": { "source": { "source": "github", "repo": "LooporaData/claude-wcag-audit" } }
  23. },
  24. "enabledPlugins": { "claude-wcag-audit@looporadata": true }
  25. }
  26. ```
  27. It runs on macOS and Linux. Windows should work too, but hasn't been tested yet. You need Node.js 18 or later and npm. The first audit installs the pinned axe-core and Playwright into the plugin's data folder, which takes about a minute. The audit also needs Chromium. If Playwright's browser isn't installed, the script falls back to any Chromium already in the Playwright cache or to `CHROME_BIN`, and otherwise prints the command that installs it.
  28. To update, run `/plugin marketplace update looporadata`. To remove it, run `/plugin uninstall claude-wcag-audit@looporadata`.
  29. ## Use
  30. Ask for an audit:
  31. ```
  32. /claude-wcag-audit:wcag-audit https://example.com
  33. ```
  34. or just say "run an accessibility audit on our staging site". Claude finds the pages from the sitemap, runs the audit, checks the heuristic findings against the live page, and reports back with the report path.
  35. ## What it checks
  36. Each page is loaded at four sizes: desktop (1280×800), mobile (390×844), 200% zoom and 400% zoom.
  37. | Area | What the script does | WCAG |
  38. |---|---|---|
  39. | axe-core | Runs axe 4.10.2 with the WCAG 2.0, 2.1 and 2.2 A/AA tags plus best practices, at desktop and mobile. Undecided results (text on gradients, for example) are kept as "needs review". | many |
  40. | Keyboard | Tabs through the whole page. At each stop it compares screenshots with and without focus, checks the element is visible and not hidden under a sticky header or banner, and watches for focus that moves the page or opens a window. Afterwards it reports traps, interactive elements Tab never reached, backwards jumps in focus order and broken skip links. | 2.1.1, 2.1.2, 2.4.1, 2.4.3, 2.4.7, 2.4.11, 3.2.1 |
  41. | Menus and dialogs | Opens each menu, disclosure and dialog trigger with Enter and Space and compares that with a click. Checks `aria-expanded`, reruns axe on the opened state, and checks focus moves into dialogs, Escape closes them and focus returns. | 2.1.1, 2.4.3, 4.1.2 |
  42. | Hover content | Hover-revealed submenus and tooltips must also appear on keyboard focus and close with Escape. | 1.4.13, 2.1.1 |
  43. | Zoom and spacing | Looks for horizontal scrolling and clipped text at 320 CSS px, clipped text at 200%, and clipping after applying the WCAG text-spacing values. | 1.4.4, 1.4.10, 1.4.12 |
  44. | Forms | Field border contrast of 3:1, autocomplete tokens on name, email, phone and similar fields, placeholder-only labels, and `<select>` changes that navigate. | 1.3.5, 1.4.11, 3.2.2, 3.3.2 |
  45. | Motion | Infinite or long animations and autoplaying media, and whether `prefers-reduced-motion` stops them. | 1.4.2, 2.2.2 |
  46. | Content | Generic and ambiguous link text, alt text that looks like a filename. | 1.1.1, 2.4.4 |
  47. | Across pages | Duplicate titles, navigation order and the placement of contact or help links. | 2.4.2, 3.2.3, 3.2.6 |
  48. The full mapping, including the manual steps left for each criterion, is in [`scripts/wcag.mjs`](scripts/wcag.mjs).
  49. ## The report
  50. Each audit writes four things to the output folder:
  51. - `report.html` is a single file with no external requests. It has the summary, a filterable table of all criteria linked to the W3C Understanding pages, findings grouped across pages, a manual checklist that remembers your ticks, evidence for manual review (alt text inventory, forms, headings, media) and a per-page table. It works in light and dark mode, prints cleanly, and passes its own audit.
  52. - `report.md` has the same content for pasting into issues or docs.
  53. - `results.json` holds every raw finding.
  54. - `shots/` holds screenshots for focus-indicator findings.
  55. Findings are graded so you know how much to trust them:
  56. | Grade | Meaning |
  57. |---|---|
  58. | Fail | A definite failure, such as an axe violation or a toggle that ignores the keyboard. |
  59. | Likely fail | Found by a heuristic. Confirm it on the page before filing it. |
  60. | Needs review | The tool can't decide. A person needs to look or make a call. |
  61. | Best practice | Worth fixing, but not a WCAG A/AA failure. |
  62. ### Recording what you verified
  63. Claude checks the likely fails and review items before reporting them, and you can do the same. Write the conclusions to `notes.json` in the output folder, keyed by the check id shown in the report:
  64. ```json
  65. {
  66. "input-boundary-contrast": {
  67. "verdict": "confirmed",
  68. "note": "Border #919ab0 on white measured at 2.82:1."
  69. },
  70. "motion-no-pause": {
  71. "verdict": "decision",
  72. "note": "Hero animation stops under reduced motion. Product to decide on a pause control."
  73. }
  74. }
  75. ```
  76. Claude re-renders the reports after writing notes. To do it yourself from a clone, without scanning again:
  77. ```bash
  78. node scripts/render.mjs a11y-report
  79. ```
  80. `confirmed` turns a likely fail or review item into a fail. `false-positive` and `no-issue` move it to a "Resolved during verification" section and out of the criteria results. `decision` keeps it under review and labels it for the owner. A note without a verdict is shown as-is.
  81. ## Running the script directly
  82. From a clone of this repository, after `npm install`:
  83. ```bash
  84. node scripts/audit.mjs https://example.com --out a11y-report
  85. node scripts/audit.mjs http://localhost:3000 --pages /,/contact,/pricing --out a11y-report
  86. ```
  87. | Option | Default | |
  88. |---|---|---|
  89. | `--sitemap URL` | `<base>/sitemap.xml` | Sitemap indexes are followed. With no sitemap, links on the home page are crawled. |
  90. | `--pages /a,/b` | | Paths or full URLs, instead of the sitemap |
  91. | `--max-pages N` | 40 | |
  92. | `--concurrency N` | 3 | Pages audited in parallel |
  93. | `--max-tabs N` | 200 | Tab presses per page |
  94. | `--skip LIST` | | Any of `keyboard`, `toggles`, `hover`, `mobile`, `zoom`, `spacing` |
  95. | `--out DIR` | `./a11y-report` | |
  96. Expect 20 to 40 seconds per page. The script never submits forms. It does open menus and dialogs and change `<select>` values, so run it against sites you own, ideally staging. It starts each page in a fresh browser with no login, so authenticated areas aren't covered yet.
  97. ## How it works
  98. - axe is injected with Playwright's `page.evaluate` in a context that bypasses CSP, so a strict `script-src` doesn't block it. The script doesn't load anything from a CDN.
  99. - The focus check takes a screenshot of each focused element and its surroundings, blurs it, takes another and counts pixels that changed significantly. Fewer than ten means there's no visible indicator.
  100. - Zoom levels are real viewports at a higher device pixel ratio (640×400 at 2x for 200%, 320×256 at 4x for 400%), which matches how browser zoom changes layout.
  101. - Clipping is found by measuring each text node's rendered boxes against the nearest `overflow: hidden` ancestor. Visually hidden text, `aria-hidden` content and text inside scrollable containers are ignored.
  102. - The keyboard walk understands roving tabindex in tabs and menus, native date inputs and embedded frames, so those don't show up as traps or unreachable elements.
  103. ## Limits
  104. Automated checks can't establish WCAG conformance. The report's manual checklist covers the rest. In particular, a person still needs to:
  105. - test with a screen reader, including status messages and live regions;
  106. - check captions, audio description and flashing content;
  107. - test form error handling on a test environment;
  108. - judge whether alt text, link text, headings and labels are meaningful;
  109. - check pointer gestures, dragging and single-key shortcuts.
  110. ## Development
  111. ```
  112. .claude-plugin/ plugin and marketplace manifests
  113. skills/wcag-audit/SKILL.md instructions Claude follows
  114. scripts/audit.mjs page discovery, browser checks, findings
  115. scripts/deps.mjs finds or installs axe-core and Playwright
  116. scripts/page-helpers.js in-page functions injected into every audited page
  117. scripts/wcag.mjs success criteria, coverage, check definitions
  118. scripts/report.mjs shared report model and Markdown output
  119. scripts/html-report.mjs HTML output
  120. scripts/render.mjs re-renders reports from results.json and notes.json
  121. test/fixture.html page with planted failures
  122. ```
  123. To try local changes in Claude Code, start it with `claude --plugin-dir /path/to/claude-wcag-audit`. Bump `version` in `.claude-plugin/plugin.json` and `package.json` for each release, since installed copies update by version.
  124. After changing a heuristic, audit the fixture and confirm every planted failure is still reported:
  125. ```bash
  126. (cd test && python3 -m http.server 8765) &
  127. node scripts/audit.mjs http://localhost:8765 --pages /fixture.html --out /tmp/fixture-report
  128. ```
  129. Also run it against a real site you know well. Most heuristic bugs show up as false positives on production markup, not on the fixture.
  130. ## License
  131. MIT, see [LICENSE](LICENSE).
  132. axe-core (Deque Systems, [MPL-2.0](https://www.mozilla.org/MPL/2.0/)) and Playwright (Microsoft, Apache-2.0) are installed from npm at first run and aren't included in this repository. Versions are pinned in `package-lock.json`.
  133. The idea of running axe from a Claude Code skill comes from [airowe/claude-a11y-skill](https://github.com/airowe/claude-a11y-skill).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.