Skip to main content
You don’t call the tools yourself. Name the site and say what you need, and the assistant picks the tools and chains them. The prompts below are starting points.

Check a site’s health

  • “How many critical and serious issues are open on example.com?”
  • “Compare open issue counts across all our sites, worst first.”

Triage issues

  • “List the critical issues on example.com with the pages they affect.”
  • “Which open issues on example.com fail WCAG Level A?”
  • “Show every color contrast issue on example.com.”
  • “Explain issue #11 on example.com in plain language for a client.”

Fix issues in code

Use this in Claude Code, Cursor, or another coding assistant, with the site’s codebase open.
1

Ask for the fix

For example: “Get the Ility fix prompt for the ‘Form elements must have labels’ issue on example.com and fix it in this repo.”
2

Review the change

The assistant pulls the failing markup and the fix guidance, finds the template that renders it, and proposes a change. Review the diff before you accept it.
3

Link the issue

Ask the assistant to link the Ility issue in the pull request, for example: “Open a pull request and link the Ility issue in its description.” The link opens the issue in Ility. See Linking to an issue.
4

Confirm the fix

Deploy the change, then ask the assistant to recheck the issue. It rescans every page where the issue is open, and the issue resolves once a scan no longer finds it.

Run and monitor scans

  • “Start a WCAG 2.2 AA scan of staging.example.com.”
  • “Scan only the unscanned pages on example.com.”
  • “Rescan the pricing page on example.com.”
  • “Is that scan done yet?”
  • “Show any failed scans this week and why they failed.”
Starting a scan and changing an issue are the only actions that change anything, and most assistants ask you to confirm before they run them.

Report to a client or stakeholder

  • “Summarize the open issues for example.com as a short client update: the open issue counts, the top five issues, and what we’re fixing first.”

Tips

  • Separate WCAG failures from best practices. Best-practice findings can outnumber WCAG failures, and they don’t count against your WCAG conformance. Ask for “WCAG failures only” when compliance is the question. See WCAG and best practices.
  • Ask for other statuses when you need them. Issue lists show open issues unless you ask for resolved, won’t fix, or false positive issues too.
  • Fix shared templates first. One issue can cover many pages. If an issue affects 140 pages, fixing the shared template clears it everywhere.
  • Run fix prompts where the code is. A fix prompt contains the failing markup, not the source file, so the assistant needs your codebase open to find the right template.

Limits

  • Scans test against the surface’s WCAG target, or WCAG 2.1 AA when it uses the workspace default, unless the request names another level. Automated testing doesn’t catch every barrier, so you still need manual review. See What a scan checks.
  • Each tool call returns at most 100 items. The assistant pages through longer lists, so large sites take several calls.
  • The assistant can comment on issues, assign them, label them, and change their status. For anything else, use Ility.

Troubleshooting