Best for
- Designers who own a live site and need a review they can act on, not a score.
- Product managers who need to know what regressed since launch, page by page.
Connectors
Neither connector is needed for the review itself. Connect them for the routine
version in Make it work for you.
Before you start
Have the public URL, and name the pages that matter most. Decide what you want out of this pass: a list of findings to hand to whoever owns the code, or fixes applied directly. Only the second needs the site open as a Replit project.Prompt the agent
What to expect
Agent works through the angles in the order you gave them, and reports what it covered: the pages it walked and the links it resolved. Each angle comes back as its own section. Every finding names the page, the issue, and the change to make. The review closes with the fix Agent would do first.Make it work for you
Troubleshooting
Links behind a checkout or a modal were never checked
Links behind a checkout or a modal were never checked
A crawl reaches the links present in the rendered page. Anything that only
appears after a form, a modal, or a payment step needs its own pass.
The layout and accessibility findings feel thin
The layout and accessibility findings feel thin
Name the pages and the viewport width you want checked, instead of leaving
the sample to Agent. Start with the pages that carry your main task.
The design feedback is generic
The design feedback is generic
Say what the page is for and who it’s for. Without the goal, a critique
defaults to layout advice that fits any site.
Related
When the review says the problem is the design itself.
Put everything on one system when the review finds drift.
For a published Replit app, audit crawlability, metadata, and markup with
one-click fixes.