What is the web quality audit skill?
The web quality audit skill is the orchestrator in addyosmani/web-quality-skills. It has an AI coding agent pick representative URLs, collect a live baseline, keep measured findings apart from code hypotheses, rank issues from critical to low, and re-run the same checks. It routes to Lighthouse, traces and the accessibility tree.
It is the web-quality-audit folder in addyosmani/web-quality-skills, and the README calls it the broad skill that orchestrates the others. When a page can run it collects live evidence first and keeps measured failures separate from source-code hypotheses.
The five steps are: set the audit target, collect a minimal live baseline, use runtime failures to localize source inspection, categorize by user impact and confidence, then re-run equivalent checks and report what still needs field or human validation.
What changes in your agent's output
- The agent states the scope first: representative URLs, key journeys, public or authenticated access, mobile or desktop.
- Findings carry a severity: critical, high, medium or low, each with an action such as fix immediately or fix within the sprint.
- Each need has a preferred tool route and a fallback, such as a performance trace or Lighthouse CLI.
- Agentic browsing is kept apart from SEO, so a Lighthouse pass is never reported as AI visibility.
- A source smoke test script, scripts/analyze.sh, is offered when the page cannot run.
What the audit covers
Five categories: performance with LCP, INP and CLS limits, accessibility, SEO, best practices such as HTTPS, CSP and a clean console, and agentic browsing, which covers the accessibility tree, optional WebMCP and optional llms.txt. A quick checklist lists items to confirm before every deploy.
Who the web quality audit skill is for
Best for
- A first pass over a page or site when you do not yet know which category is weakest.
- Pre-launch reviews where one report must cover speed, access and search.
- Handing off to the deeper skills: performance, accessibility and SEO.
Not for
- Deep fixes inside one category. Use the matching sibling skill after the audit.
- Whole-site crawls of hundreds of pages. For crawler-style audits see audit-website.
Use it with UiChemy
Audit the live WordPress page that UiChemy produced, since the audit judges the rendered result rather than the design file. Fix findings in Elementor or Gutenberg and re-run. Click Get Started on uichemy.com.
How to install Web Quality Audit Skill: Lighthouse-Style Audits for AI Agents
git clone https://github.com/addyosmani/web-quality-skills.gitClone the repository
Run
git clone https://github.com/addyosmani/web-quality-skills.git. Each skill is a folder inside the skills directory, and web-quality-audit is skills/web-quality-audit.Copy the skill folder
Copy skills/web-quality-audit into the folder where your agent loads skills, for one project or globally. The README's manual route copies the whole skills directory, which keeps the cross-links between the sibling skills working. The audit skill links to its siblings, so copy the performance, accessibility and seo folders too, or copy the whole skills directory.
Add live browser tools if you have them
The README says the skills prefer Chrome DevTools MCP for traces and Lighthouse runs when your agent exposes it. Without it they fall back to the Lighthouse CLI, PageSpeed Insights and manual checks.
Ask in plain words
Skills activate when your request matches their description, for example by asking for a quality review of a page.
Check that it loaded
Ask for a quality audit of one URL. The report should separate measured findings from code hypotheses, group issues by severity, and list what still needs field or human validation.
Key features
- Orchestrates the other skillsRoutes to performance, accessibility, SEO and best-practice checks from one entry point.
- Live-evidence firstCollects a minimal baseline before searching the codebase broadly, and keeps measured findings separate from code-only hypotheses.
- Tool routing tableMaps each need to a preferred route and a fallback, for example trace analysis with Lighthouse CLI or PageSpeed Insights as fallback.
- Five audit categoriesPerformance, accessibility, SEO, best practices and agentic browsing.
- Severity levelsCritical, high, medium and low with a matching action for each.
- Defined audit outputThe skill specifies an audit output format so reports are comparable between runs.
- Agentic browsing checksSemantic HTML and accessible controls, valid WebMCP when present, and optional llms.txt, never claimed as search ranking evidence.
- Source smoke testscripts/analyze.sh runs a quick source analysis when a live page is not available.
- Pre-deploy checklistCore Web Vitals, accessibility errors, console errors and HTTPS before every deploy.
- Re-check stepRe-runs equivalent automated checks and the affected manual flows, and reports what is verified versus pending.
Frequently asked questions
What is the web quality audit skill?
It is the orchestrating skill in addyosmani/web-quality-skills. It has an AI coding agent run an evidence-led review across performance, accessibility, SEO, best practices and agentic browsing, rank the findings and verify the fixes.
How do I install it without npx?
Clone the repository with git, then copy skills/web-quality-audit into your agent's skills folder. Copy the performance, accessibility and seo folders too, since the audit links to them, or copy the whole skills directory in one step.
What does it audit?
Five categories: performance and Core Web Vitals, accessibility, SEO, best practices such as HTTPS and security headers, and agentic browsing. Each has a checklist of specific items and a severity scale from critical to low.
What is agentic browsing in this skill?
It is a Lighthouse category about technical signals that help an assistant understand and interact with a rendered page: the accessibility tree, optional WebMCP registrations and optional llms.txt. The skill says it is not evidence of search ranking.
Does it need a running site?
It works best on a runnable page because it collects live evidence first. When the page cannot run, it falls back to source inspection and the analyze.sh smoke test, and labels those findings as hypotheses.
How is it different from audit-website?
This skill uses Lighthouse, traces and manual checks on chosen URLs. audit-website drives the squirrel crawler CLI across many pages with 260 plus rules. They suit different scopes and can be used together.
What license does it use?
The repository is MIT licensed and the SKILL.md frontmatter says MIT. The 29K install count on this page comes from a third-party skills listing and is not stated in the repository.