> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dallaspeters.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Design reviews and the AI UX audit

> Create a client-facing UX review, capture or upload screens, run the Claude UX audit, capture apps behind a login with the audit agent, and share it.

A design review is a client-facing UX critique of a website or app. This page covers creating a review, adding screens and notes, running the AI UX audit, capturing pages behind a login with the audit agent, and sharing the finished review with a client.

## Create a review

**1. Start a draft**

In the admin, open **Reviews** and select **New review**. Enter a **Title** and, optionally, a **Client (as it should read on the page)**. The address preview below the title shows the public path, `/review/<slug>`. Select **Create draft**.

**2. Fill in the settings**

In the editor's side panel, the **Settings** group holds **Title**, **Client label**, **Client email**, **Brand kit** and **Address**. The draft saves itself 1.2 seconds after each change. A new **Address** is kept only when you select **Save**.

**3. Add screens and sections**

Select **Add screen** and **Add section** to build the body. Every block folds away, and you drag sections and screens into a new order by their handles.

## Add screens

Each screen has a **Screen title**, a **Viewport** (**Desktop (1440)** or **Mobile (390)**) and a **Page address**.

* Select **Capture** to screenshot the page address. Once a screen has a picture, the button reads **Recapture**; pins and markup stay where they are on the page.
* Select **Upload screenshot** to use your own PNG, JPEG or WebP image instead.

An address typed without `https://` has it added before the capture.

## Notes and markup

Notes are numbered pins placed on a screenshot. Each note is one of four kinds: **Issue**, **Recommendation**, **Suggestion** or **Works well**.

Markup draws over the screenshot with four kinds of mark: box, highlight, arrow and pen. With the Edit tool, click a mark to select it, drag it or its ends to move or reshape it, and press Delete to remove it.

## Run the UX audit

The **Audit** group in the side panel has two buttons.

> [card] Run UX audit
>
> Audits every screen that has a picture. If no screen has one yet, it asks which page to capture, screenshots it at desktop and phone width, then audits those.

> [card] Capture behind login
>
> Sends the audit agent to sign in to an app with a stored test account, capture its key pages, then runs the same audit on them.

The audit scales each screen to a readable width and cuts it into tiles for Claude, together with the page's HTML when the source address is reachable. Claude judges the screens against the vendored references: the heuristic audit, all 21 Laws of UX, the design references and the severity model.

An audit usually takes one to three minutes. It adds to the review without changing anything already there:

* The executive summary is appended to the **Findings** text.
* New sections are added: **Strengths**, **Laws of UX** (grouped as **Needs attention**, **Opportunity** and **Working**), **Recommendations**, and **Further findings** for points no screenshot shows.
* Each screen gets 10 to 18 pinned findings, spread over the whole screen, including at least two **Works well** pins where they are earned.

## Capture behind login

**Capture behind login** is for apps that need an account. The audit agent drives its own browser.

**1. Name the app**

Enter the app's address in **Which app should the agent sign in to?** and select **Capture and audit**.

**2. Name the sign-in page**

In **Where is the sign-in page?**, leave the app's address if the app shows its sign-in page there, or enter the sign-in page. Select **Start**.

**3. Wait for the agent**

The button label shows the agent's current step. It captures the start page and up to three more key pages, each at desktop and phone width. This takes a few minutes, and the editor stops waiting after 30 minutes.

When the agent finishes, its screens join the review. It also returns a check against every reference, added as **Validation: Heuristic audit**, **Validation: Laws of UX** and **Validation: Design references** sections. The UX audit then runs on the new screens and adds its pins.

> [!WARNING]
>
> The agent looks but never changes anything. It only navigates, scrolls, opens menus and read-only views. It never submits a form, saves, deletes, sends, pays, changes settings or logs out.

The test account's credentials are stored on the agent and are never shown to the model or typed into the editor. Each session has a \$5 cost cap.

## Share with the client

> [accordion] Publish and preview
>
> **Save and share** holds save, **Publish** (or **Unpublish**), preview, **PDF** and **Send to board**. A draft is not visible to the client.

> [accordion] Client access
>
> Under **Client access**, a review is open to **Anyone with the link** until you select **Set password**. Passwords need at least 4 characters. Screenshots live in a private store and are only served through signed links that expire.

> [accordion] Send to client
>
> Under **Send to client**, select **Copy link**, or add an optional note and send the link to the **Client email**. Sending requires a published review and a saved client email.

> [accordion] Client comments
>
> On the public page the client reads the review, adds comments, pins and drawings, and downloads a PDF. Their comments appear under **Client access**, where you can mark each resolved, reopen it or remove it.

> [!TIP]
>
> **Send to board** turns each screen into a Browser node on the [canvas](https://docs.dallaspeters.com/guide/canvas.md), with its markup and notes. Work on it there and publish it back; the pictures stay private in the review.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.