Accessibility Audit Co. | WCAG 2.2 AA Audit
DISCLAIMER — READ FIRST. This Agent produces a technical accessibility assessment, not legal advice. Its output is not a compliance certificate, does not establish ADA, WCAG 2.2, EN 301 549 or EAA conformance, and is not a substitute for advice from a qualified attorney or a certified accessibility auditor. For any specific claim, demand letter, procurement requirement or legal question, consult a licensed attorney. Use only on websites you own or are explicitly authorized to test. No guarantee of compliance, of legal outcomes, or of freedom from litigation is made or implied.
Give it a URL. Get back an evidence-backed accessibility audit and three client-ready files — no manual checklist and no copy-pasting findings into a document.
What it does

1. Samples up to 20 same-domain pages from the URL you provide, using breadth-first link traversal.
2. Checks every page with a static HTML engine (image alt text, form labels, html lang, page title, heading order, generic link text) and, when a browser runtime is available, with axe-core in headless Chromium (colour contrast, ARIA / name-role-value, link and button names, and the WCAG 2.2 target-size rule).
3. Delivers deduplicated, severity-ranked findings mapped to WCAG Success Criteria with copy-paste fixes, plus a Markdown report, a branded PDF and a fixed-schema JSON.
   Example input
   python scripts/audit.py --url https://your-client-site.com
   Or simply ask in chat: "Audit https://your-client-site.com for WCAG 2.2 AA."
   No flags are required. Crawling, browser checks and file generation are all on by default.
   Example output — excerpt from a real scan

WCAG 2.2 AA Crawl Audit

This report is based on WCAG 2.2 AA, which is backward-compatible with
and satisfies WCAG 2.1 AA (the U.S. ADA Title II statutory baseline).

Site Summary

• Pages scanned: 20 (errors: 0)
• Total issues: 390 (Critical 80, Serious 157, Moderate 153)
• Unique issue types: 8
  These figures come from one real run against a public e-commerce storefront and are illustrative only; every site returns different results. The full report continues with scan-quality notes, a Remediation Priority table and per-page findings with suggested fixes.
  Deliverables and output format
  File
  What it contains
  Typical use
  report.md
  Site summary, scan-quality notes, remediation priority table, per-page findings with fixes
  Engineering hand-off, client review
  report.pdf
  Branded cover, executive summary, findings grouped by severity, WCAG 2.2 basis, disclaimer
  Sending to a non-technical client
  report.json
  Fixed-schema data: summary, findings[], warnings[], axe diagnostics, wcag_2_2
  Jira import, CI checks, internal dashboards
  Core conclusions are also printed as plain text in the chat, so they are readable without opening any file. All deliverables are written in English with ASCII filenames.
  Capabilities
  Area
  Detail
  Standards baseline
  WCAG 2.2 Level AA, backward-compatible with 2.1 AA
  Automated static checks
  SC 1.1.1, 1.3.1 / 3.3.2, 3.1.1, 2.4.2, heading hierarchy, 2.4.4
  Browser-level checks
  axe-core: SC 1.4.3 contrast, 4.1.2 ARIA, link/button accessible names, 2.5.8 target size (24 px)
  False-positive control
  Links hidden by stylesheet rules are suppressed; an axe link-name violation restores them
  Deduplication
  Identical issue types found on multiple pages are aggregated with a page count
  Severity ranking
  Critical / Serious / Moderate, with a remediation priority table
  Scan-quality evidence
  axe.status, result_quality, content_drift, suspect_empty_pages, warnings[], requires_manual_review
  Honest conformance basis
  WCAG 2.2 SC 2.4.11, 2.5.7, 3.2.6, 3.3.7 and 3.3.8 are flagged as manual review required, never reported as automated passes or failures
  Offline mode
  --html-file audits a local HTML file with no network access
  Honest limits — please read before buying
• A bounded sample, not a full-site crawl. The default is up to 20 pages discovered by BFS from the seed URL. There is no per-page-type quota. Pages not linked from the seed, pages beyond the cap, and JS-only routes are not covered. Raise the cap with --max-pages N.
• Some things cannot be decided automatically: keyboard interaction, focus order in dynamic widgets, hover/focus states, external CSS beyond inline stylesheets, @media and !important edge cases, and content that appears only after user interaction. These need a human audit.
• axe incomplete results are never reported as violations.
• If the browser runtime is unavailable, browser-level checks are skipped — not silently treated as clean. The report records axe.status = skipped, result_quality = not_run, explicit warnings, and requires_manual_review = true.
• A PASS or FAIL label is a technical result of the checks that actually ran. It is not a legal conformance determination.
  Use cases
• Agencies running pre-launch accessibility QA and needing a client-facing deliverable.
• Small businesses gathering technical evidence to hand to counsel before responding to an accessibility demand letter.
• Development teams checking a redesign or release for regressions.
• Procurement and vendor reviews that ask for accessibility evidence.
• Periodic rechecks after remediation, to see which issues were resolved and which remain.
  Network and data flow (declared)
• At runtime the Agent sends HTTP GET requests only to the URL you provide and to same-domain pages discovered from it. Headless Chromium loads those same pages to evaluate rendered CSS and element geometry. It does not submit forms, log in, or send page content anywhere else.
• The code contains no analytics, no telemetry, no advertising SDK, no third-party API calls and no hardcoded credentials. Reports are written to the local output directory (or the Agent workspace output directory when hosted) and returned to you; the publisher does not receive a copy.
• Dependencies are fetched at install/build time from official registries only: reportlab from PyPI, axe-core and puppeteer from the npm registry. These are package sources, not runtime destinations for your data.
• In --html-file offline mode, no network requests are made at all.
• You are responsible for having authorization to scan the target site and for any personal data that may appear on its public pages.
  Requirements
• Python 3.12+ with reportlab (PDF generation). The static engine uses only the Python standard library.
• Node.js with puppeteer and axe-core for browser-level checks. Optional: without them the static engine still runs and reports the reduced coverage.
• When hosted on this platform, dependencies are installed during the build. For the downloadable package, see the bundled README.
• Useful flags: --max-pages N, --no-crawl, --no-axe, --grouped, --output-dir, --no-files, --html-file.
  FAQ
  Is this a compliance certificate? No. It is a technical assessment with evidence and fixes. Conformance is a legal and human determination.
  Does it scan my entire site? No. It samples up to 20 same-domain pages by default. Unlinked, gated and JS-only pages are out of scope.
  Does a PASS mean I am ADA compliant? No. It means the checks that ran found no violations in the sampled pages. Manual review is still recommended.
  What if the site blocks the scanner? Rate limits (HTTP 429), WAF challenges and redirects are retried and recorded. Affected pages are listed in the scan-quality notes and are never counted as clean.
  Can I use it on a client site? Only with the owner's authorization. Unauthorized scanning is out of scope.
  Which language are the reports in? English. The Agent Card may be localized by the platform, but generated deliverables are English only.
  What if axe cannot start? You still get a complete static report, with the missing browser-level checks explicitly disclosed.
  Support
  For questions, bug reports or help interpreting a report, contact the Support Email shown on this Agent Card. The platform does not provide technical support on the publisher's behalf.