- Home
- Skills
- Frontend & Web Apps
- Site Tech Check: website security and performance audit
Site Tech Check: website security and performance audit
Website audit from one prompt: four scores for code, security, performance and accessibility. Every finding comes with proof.
Free
Works with the AI Agents you already use
Site Tech Check: website security and performance audit
Example session with this skill installed
Run a tech check on the site in ./site. The goal is that visitors book a bike repair.
- Read your context and instructions
- Compiled the site tech check
- Generated the UI component
Real run on a demo site (a fictional bike repair shop, 6 pages), with a browser tool connected. These are the top 5 of 9 findings.
Scorecard: Oakline Bike Repair, round 1
| Dimension | Score | Why |
|---|---|---|
| Code and robustness | 3 | Booking form has no backend and no validation, price estimate is wrong |
| Security and privacy | 3 | Admin token in public JavaScript, personal data in the URL |
| Performance | 8 | Whole site is about 12 KB, no fonts or third parties. Load time not measured |
| Accessibility | 3 | No form labels, contrast down to 1.84:1, homepage overflows on phones |
| Total | 4.3 |
Top fixes
1. CRITICAL: price estimate on the booking form is wrong for every option (verified)
- Where:
main.js:12-17,book.html:27-33 - Proof:
const price = PRICES[service.selectedIndex];with a placeholder option at index 0. In the browser: "Standard tune-up => $19", "Full overhaul => $undefined" - Why: the visitor sees a wrong price at the moment of booking
- Fix: give each option a value and map by value. Effort: S
2. CRITICAL: admin token in public JavaScript (verified)
- Where:
main.js:3, loaded on the homepage and the booking page - Proof: the credential search hits
main.js:3,BOOKING_ADMIN_TOKEN = "demo..."(first 4 characters shown only) - Why: anyone can read it. Treat as leaked
- Fix: delete the line, rotate the token, keep tokens server-side. Effort: S
3. CRITICAL: booking form sends personal data in the URL and delivers the booking nowhere (verified from the HTML, not submitted)
- Where:
book.html:22 - Proof:
<form action="thanks.html" method="get" novalidate>, and no request to a backend anywhere inmain.js - Why: the booking is lost, and name, email, phone and address land in browser history and server logs
- Fix: POST to a real endpoint or form service, then redirect. Effort: M
4. CRITICAL: homepage overflows sideways on phones (verified)
- Where:
styles.css:24,.partners { width: 900px; } - Proof: at 375px the page is 900px wide
- Why: the homepage scrolls sideways or shows zoomed out on the main device
- Fix:
flex-wrap: wrap; width: auto;. Effort: S
5. IMPORTANT: text contrast far below 4.5:1 (calculated)
- Where:
styles.css:9andstyles.css:30 - Proof:
#bbbon#fbfaf6is 1.84:1,#aaais 2.22:1 - Why: these styles carry the tax and parts notes, so the price conditions are close to unreadable
- Fix: use
#5f6862or darker for muted text. Effort: S
Keep this
- Tiny footprint: no frameworks, system fonts, no trackers
- All internal links and assets return 200
- Button and link contrast is good (6.29:1)
Not covered by this check
Design, real user testing on phone and desktop, SEO and sales copy, and pricing and offer were not reviewed.
Which fixes first?
Without a browser tool the scores and the top five come out the same. Layout findings such as number 4 are then reported from the CSS and marked as not tested in a browser.
book.html
TSX · React component
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Most site checks give you a list of warnings and no way to tell which ones matter. This skill runs one specialist reviewer over your site, checks its findings before you see them, and hands you a short scorecard ranked by what hurts your site's goal most.
What you get
- Four scores from 1 to 10: code and robustness, security and privacy, performance, accessibility
- A ranked fix list: what is wrong, where, the proof, why it matters, the fix, and the effort (S, M or L)
- "Keep this": what already works, so the next change does not break it
- A review log, so you can see scores move from round to round
What it checks
- Code and robustness: console errors (with a browser tool), broken links and assets, forms, HTML basics
- Security and privacy: secrets in front-end code, HTTPS, security headers, trackers, where form data goes
- Performance: page weight, images, fonts, render-blocking files
- Accessibility: contrast, keyboard use, alt text, tap targets (WCAG 2.2 AA basics)
The rules it works by
- No evidence, no finding. Every item has a location, proof, a fix and an effort estimate
- Verified before you see it. Your agent re-checks every critical finding, number and security claim, and drops what does not hold up or marks it unverified
- No invented numbers. Load times and sizes are measured or marked as not measured
- Read-only. It does not edit your site, submit forms or install anything. It only saves the scorecard and a log in a
.site-reviewfolder in your project. You pick the fixes afterwards
Use it like this
- "Run a tech check on https://example.com, the goal is booking a call"
- "Check the site in ./site before I launch, visitors should sign up for the newsletter"
- "Check the tech fixes again, we fixed items 1 to 3"
Good to know
- Works on live sites, staging sites and local project folders
- Your agent needs a shell with curl and web access. Local sites also need Python 3 or your own dev server
- Commands are written for macOS and Linux. On Windows use Git Bash or WSL
- A browser tool is optional. With one, it also checks the console and the layout on phone and desktop widths
- It fetches the pages of the site you name, and for local folders starts a local preview server that it stops afterwards
- Advice from a basic review, not a penetration test or a compliance audit
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.
Does not review design, SEO, sales copy or pricing. Does not submit forms, so form behaviour is judged from the code. Without a browser tool, console errors, layout, keyboard use, focus styles and zoom are marked as not tested. HTTPS and security headers are skipped for local folders. Load times are only reported when measured.
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 today
- Free with an Agensi account