Checks

Each check is a library method and a journey command. In a journey it lists what it found, and with --strict fails the step instead.

var violations = await page.AuditAsync();            // or AuditAsync("main") for one element
Console.Write(AccessibilityAudit.Format(violations));

violations: 12
[serious] aria-hidden-focus: Content hidden with aria-hidden must not be focusable (1 node)
  - div[aria-hidden="true"]
[critical] button-name: Buttons need an accessible name (1 node)
  - button
[serious] color-contrast: Text contrast meets WCAG AA (4.5:1, or 3:1 for large text) (1 node)
  - p (1.92:1)
…

Our own rules, with no third-party engine. Rule names and the layout follow agent-browser's a11y (which runs axe-core), so results read alike.

Names come from Chrome's accessibility tree, which computes them as assistive technology does. Contrast composes the text colour over the nearest solid background through semi-transparent layers; text over a background image or gradient is skipped as undecidable, and disabled controls are exempt, as WCAG says. It finds the common problems; it is not a full WCAG audit.

a11y
a11y -s main
a11y --strict

1. list
2. one element's subtree
3. fail the step on any violation

var links = await page.LinksAsync();                  // or LinksAsync("main")
var results = await page.CheckLinksAsync(links);
foreach (var link in results.Where(l => l.Broken))
    Console.WriteLine($"{link.Outcome} {link.Url}");

1. 404 http://localhost:5000/old-page

LinksAsync collects every a[href] and area[href] with an http, https or file URL, once each, without fragments, in document order. CheckLinksAsync fetches them through the browser, four at a time: a GET outside the page, so the page's CORS rules don't apply and page script never sees it, with the body closed unread. Cookies go only to the page's own origin. File links are checked on disk.

In a journey, links checks same-site links only. --external adds links to other sites, but only hosts in the allow list are fetched, so a page cannot make a run send requests anywhere.

links: 2 checked, 1 broken, 2 on other sites not checked (--external)
404 /missing-page

--all lists working links too; --strict fails the step on a broken one. Same-site links print without their origin, so a transcript does not depend on the port. A link behind HTTP authentication reports 401: set credentials applies to the page's own requests only.

network requests --failed

GET http://localhost:5000/missing.png (Image) 404

Requests answered with 400 or more, and requests that got no answer at all (net::ERR_NAME_NOT_RESOLVED, a refused connection). Requests the allow list refused are left out, since the run reports those already, and so are requests a navigation cancelled. --strict fails the step when anything is listed. In code, filter page.RecordNetworkAsync()'s requests on Status and Error.

errors lists uncaught errors, one per line with their location. To fail a journey on them, snapshot the errors checkpoint: an empty list that becomes non-empty is a difference.