- Home
- Skills
- Frontend & Web Apps
- Agent-Readiness Audit
More screenshots
Works with the AI Agents you already use
Agent-Readiness Audit
Your coding agent audits a web page for AI-agent readiness (42 checks, 0–100 score), fixes the issues in your source code, and re-audits until it passes. Improves WCAG accessibility at the same time.
Free
Agent-Readiness Audit
Example session with this skill installed
Our product page is at http://localhost:3000/products/trail-tent. Make it agent-ready so AI shopping agents can read it and add it to the cart. Fix the issues in the source and keep everything accessible.
- Read your context and instructions
- Compiled the agent-readiness audit
Summary
Audited the product page, fixed all 17 issues in the source, and re-audited. The score went from
48 to 100, "Hard for agents" to "Agent-ready". 11 of the fixed issues were also WCAG 2.2 failures.
| Category | Before | After |
|---|---|---|
| Overall | 48 | 100 |
| Real controls | 44 | 100 |
| Labels and names | 50 | 100 |
| Forms | 14 | 100 |
| Page structure | 40 | 100 |
| Shopping agents | 4 | 100 |
| Machine-readable signals | 53 | 96 |
What was wrong
"Add to cart" was a styled <div>. Agents look for buttons in the accessibility tree, so they couldn't find it (WCAG 4.1.2, 2.1.1).
The wishlist button was an icon with no name. An agent told to "add to wishlist" had nothing to match (WCAG 4.1.2).
- The email field had only placeholder text, no label or input type (WCAG 3.3.2, 1.3.5).
- No Product structured data, so shopping agents couldn't read the price or stock.
- No
<main>,<h1>orlang, and a generic "Home" title.
The fix
<!-- Before -->
<div class="buy" style="cursor:pointer" onclick="addToCart()">Add to cart</div>
<button><svg>…</svg></button>
<input placeholder="Your email">
<!-- After -->
<button type="button" onclick="addToCart()">Add to cart</button>
<button type="button" aria-label="Add to wishlist"><svg aria-hidden="true">…</svg></button>
<label for="email">Email for restock alerts</label>
<input id="email" type="email" autocomplete="email">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Product",
"name": "Trail Tent",
"brand": { "@type": "Brand", "name": "Northfold" },
"sku": "TT-2",
"offers": {
"@type": "Offer",
"price": "199.00",
"priceCurrency": "USD",
"availability": "https://schema.org/InStock"
}
}
</script>
Also added lang="en", a descriptive title ("Trail Tent | Northfold Outfitters"), <header>, <nav>, <main> and an <h1>, alt text on the product photo, a meta description, Open Graph tags and a canonical link. The structured-data price matches the price shown on the page.
Next steps
- One optional item is left: publish an
/llms.txtat the site root (a warning, not a failure). - Add a CI check so the score can't drop:
npx -y agent-readiness-audit http://localhost:3000 --min-score 90 - Run
scan_siteon the home page to find problems shared across the other product pages.
Connects securely to your tools. The creator never sees your data.
About this skill
Make your web pages work for AI agents
AI agents like Claude in Chrome, Gemini in Chrome and Perplexity Comet don't see your design. They read the page's accessibility tree, the same structure screen readers use. A clickable <div>, an icon button with no name or a form field with only placeholder text stops them, and blocks keyboard and screen-reader users too.
This skill gives your coding agent a tested loop to find and fix those problems in your actual source code.
How it works
- Audit: 42 checks give a 0–100 score, and every problem element comes with its CSS selector and suggested code.
- Plan: failing checks are ordered by impact, with the issues that block agents completely first.
- Fix in source: in your components and templates, not the rendered HTML.
- Re-audit: until the score reaches your target (90+ by default).
- Report: the score before and after, the checks fixed and the files changed.
For whole sites, it scans up to 25 pages and finds problems shared across pages, so the agent fixes the shared template once.
What it checks
- Real controls: clickable divs, dead links, keyboard reachability, visible focus
- Labels and names: icon buttons, vague names like "Click here", alt text, misleading ARIA
- Forms: labels, input types, autocomplete, required fields
- Interaction traps: overlays, CAPTCHAs, cross-origin frames, tiny targets, layout shifts
- Stable targeting: generated class names and auto-generated IDs
- Page structure: title, language, landmarks, headings, content that needs JavaScript
- Agent safety: hidden prompt-injection text
- Shopping agents: Product data, price, stock, shipping, returns, the cart button
- Machine-readable signals: metadata, structured data, llms.txt, robots.txt rules for AI
19 checks are also WCAG 2.2 criteria, so you improve accessibility at the same time.
Built-in fix rules
The agent prefers native HTML, writes names that say what each control does, keeps fixes keyboard- and screen-reader-friendly, and never hides problems (no aria-hidden on focusable elements, no removed focus outlines, no weakened tests).
Works with
Claude Code, Cursor, GitHub Copilot, Claude Desktop, Windsurf, Gemini CLI, Codex CLI and any MCP client. It uses the free agent-readiness-audit MCP server (npx -y agent-readiness-audit mcp), or its CLI in agents without MCP. Setup instructions for each agent are included.
Private
Everything runs on your machine. The audit reads the page and a few files on the same site, changes nothing, and sends nothing anywhere.
Also available
The free Agent-Readiness Auditor Chrome extension audits any page in your own browser, including signed-in pages, and runs a live agent test.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Frontend & Web Apps.
Cannot audit pages behind a login. Requires Playwright/Chromium installed locally. Scores are heuristics and should be verified with real agent tests for critical flows.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 12 days ago
- Free with an Agensi account
- Software
- agent-readiness-audit MCP server (npm)
Node.js 18+Playwright Chromium
Playwright
Permissions
Allowed Hosts
File Scopes
Terminal: runs the free agent-readiness-audit CLI/MCP server via npx, installs Playwright's Chromium once, and runs your project's tests after fixing. Read/Write Files: finds the components and templates behind each problem element and fixes them in your source; it never edits generated, vendored or test files to make a check pass. Browser: the audit loads your page in a local headless Chromium. Network: only the page you ask it to audit, plus /robots.txt, /llms.txt, /sitemap.xml and /.well-known/ files on that same site; npx downloads the package from npm. No API keys, accounts or environment variables. Nothing is sent to any other service.