Four agents review the site: A+ scores and what they found

In the previous post, “The avatars: the face of the team”, I showed how Art created animated avatars in the style of Darth’s family for the agents, giving the team a recognizable face.

What I wanted to solve

By October 3, 2026, my personal website was live at rodrigoesporto.com. The core structure worked, the main pages were approved, and our essential cloud security settings were active.

The trouble with spending days building your own project is that your eyes get used to whatever is on screen. You click the same buttons, know where every page leads, and miss obvious flaws that a first-time visitor would catch right away.

I did not want to call the site finished based only on my own impressions. I needed a genuine audit with clear technical standards and zero attachment to what I had already built.

Around 5:50 PM that day, I asked four agents in the Grok Bot app to run a complete review of the live site. I set one rule upfront: it had to be strictly read-only. No changing files and no editing panel settings while tests were running.

I work during the day, and at night and on Sundays I build with AI agents. I brought in four assistants with different roles to inspect every corner of the site.

Quick summary

• At 5:50 PM on October 3, 2026, four Grok Bot agents (Security, Analyst, Legal, and Art) ran a read-only audit of the live website.
• Security ran external public tests: A+ on Mozilla Observatory, A+ on Security Headers, and A+ on SSL Labs, plus 93% for the website and 100% for email on internet.nl, with no sensitive files exposed and no photos containing location data.
• Analyst measured navigation with Google Lighthouse: speed between 99 and 100, SEO at 100, accessibility between 87 and 95, and all 60 tested URLs loaded without errors.
• Legal confirmed we use no cookies or banners and recommended adding a clear privacy page along with the theme license notice.
• Art recommended replacing the default theme icon with our own, creating a fresh social sharing image, and darkening our accent green to improve text contrast.
• All feedback was compiled into a single prioritized checklist for me to review before touching any code.

First: what is a read-only audit and what is Lighthouse

Before reviewing the results, it helps to explain both concepts in plain English.

A read-only review means inspecting a live system without touching anything. Think of a standard car inspection: the mechanic checks the oil level, tests tire pressure, checks headlights, and takes notes on a clipboard. They do not pull the engine apart or swap parts while inspecting. On a website, this means the agents visited pages and queried external testing platforms without changing any server files. That prevents introducing fresh bugs in the middle of a check.

Google Lighthouse is a free tool maintained by Google to evaluate website quality. It simulates a visit from a mobile phone or computer and scores four areas from 0 to 100:

• Performance: how quickly the page loads and becomes usable.
• Accessibility: whether people with visual impairments or screen readers can navigate easily.
• Best Practices: whether the code follows modern standards for security and stability.
• SEO (Search Engine Optimization): whether content is structured so search engines like Google can understand it.

The car inspection comparison has a clear limit. Lighthouse measures technical metrics: load times in milliseconds or color contrast ratios between text and background. It cannot tell whether your writing is clear or whether an explanation actually helps a reader. It checks the condition of the car, but it cannot tell you if you are heading in the right direction.

What I used

• Security: an agent in the Grok Bot app that checks the site daily from the outside to ensure it stays safe, without touching anything.
• Analyst: an agent in Grok Bot that tracks our X profile and website visits.
• Legal: an agent in Grok Bot that answers legal questions in plain language.
• Art: an agent in Grok Bot responsible for visuals and team artwork.
• Public external audit tools: Mozilla Observatory, Security Headers, Qualys SSL Labs, internet.nl, and Google Lighthouse.

How the four agents reviewed the site

The audit started in the late afternoon of October 3, 2026. Each agent received instructions to examine the live site strictly through the lens of their specialty.

Security focused on infrastructure protection. The agent tested the strength of our encrypted connections and HTTP security headers configured in Cloudflare using independent public tools. The results were clean: an A+ on Mozilla Observatory, an A+ on Security Headers, and an A+ on SSL Labs. On internet.nl, which tracks modern internet standards, the site scored 93% and email domain protection reached 100%. Security also checked for leftover configuration files, exposed directories, or GPS coordinates embedded in images. It found nothing critical or high-risk.

Analyst checked the everyday user experience. Running Google Lighthouse across the main routes produced strong scores: speed between 99 and 100, SEO at a flat 100, and accessibility between 87 and 95. Next, the agent tested all 60 addresses that formed the site at the time, across Portuguese pages, English versions, sections, and categories. Every single page opened without loading errors or broken links.

Legal evaluated compliance and usage terms. The opening news was good: because the site uses no tracking cookies or advertising scripts, we do not need screen-blocking consent popups. But Legal pointed out two necessary steps: even without user forms, sound practice requires a public privacy page explaining the technical processing done by our hosting provider. Legal also noted the need to keep the open-source license notice visible for the Hugo theme we selected.

Art evaluated visual polish and reading comfort. That review caught three details I had overlooked: the small browser tab icon was still the generic image included with the theme; the image shown when someone shares a link on X needed an update to match the project; and a green tone used for links was slightly too light, which reduced contrast and made reading difficult.

Once the reports were complete, I did not rush to fix anything immediately. I merged every note into a single list, sorted by priority, so I could decide the next steps calmly.

What I learned from this review

The first lesson was seeing the difference between checking your own work and getting an independent review.

When I browsed the site myself, everything looked fine. My eyes already knew where to click, so I never noticed that the browser tab icon was a template default, or that a light green link required extra effort to read on a dim screen. It took Art to spot the contrast issue and Legal to remind me that skipping cookies does not excuse skipping a clear privacy notice.

The second lesson was the importance of separating diagnosis from repair.

The natural reflex when spotting an issue is to open the file and change the first line of code you see. But editing styles or swapping images while tests are running changes page weight, throws off Analyst’s numbers, and confuses the audit.

Running a strictly read-only check, writing the issues down in a single document, and ranking them by priority turns scattered impressions into an orderly work plan.

Costs and limits

In money, this audit cost zero dollars. Mozilla Observatory, Security Headers, SSL Labs, internet.nl, and Google Lighthouse are free public services. The four assistants ran inside the Grok Bot subscription I was already using daily.

In time, running the tests, reading the reports, and organizing the final checklist took about forty minutes on the afternoon of October 3.

Regarding limits, it helps to be realistic about what these tools actually do:

• High scores do not evaluate writing: scoring 100 in speed and 100 in SEO on Lighthouse simply means the code is fast and well organized. It does not prove the article is good or that readers will find it valuable. Content still has to stand on its own feet.
• Read-only reviews do not fix anything: an agent can spot a flaw, but the fix depends on you. Picking better colors, writing helpful explanations, and updating images still requires human judgment and effort.
• Small adjustments are routine: achieving A+ security scores brings peace of mind on the technical foundation, but any live site keeps evolving. There are minor adjustments in progress that I handle day to day.

What I would do differently

I would have asked the agents for this review a little earlier, right after setting up the initial page templates in Antigravity, instead of waiting until the whole site was built.

Testing the live site worked fine, but catching details like the default theme icon and link contrast during the drafting stage would have saved a few cleanup steps at the end.

Who this is for and who it is not for

Useful if you just put a site, blog, or portfolio online and want to confirm the technical foundation is solid before sharing it.

Also useful if you want to learn how to split work among AI assistants with clear, specific jobs instead of dumping everything onto a single chatbot.

It will not help if you expect automated test scores to make a site successful. The writing still has to be worth reading.

It is also not for anyone in a rush who prefers publishing immediately without checking technical and visual details.

Closing

Publishing a website brings a satisfying sense of progress, but a careful audit is what turns an initial draft into a dependable project.

Having four specialized assistants review security, navigation, legal requirements, and visual design delivered strong marks and gave me a practical list of improvements I could not have caught on my own.

If you have a site online or are getting one ready, try a simple test: run your address through Google Lighthouse or check it on Security Headers. Looking at the numbers without guesswork shows you exactly where your setup is strong and where it still needs work.

To follow the next steps of this journey and discuss practical ways to work with AI agents, you can find me on X: @RodrigoesPorto.

In the next post, “A single name: Rodrigo ES Porto across the entire site”, I will share how I unified my name across every page.

Sources and links

• Google Lighthouse official documentation: https://developer.chrome.com/docs/lighthouse
• Internal review reports from October 3, 2026, by agents Security, Analyst, Legal, and Art.
• Official profile on X: @RodrigoesPorto.