More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    UI Polish Reviewer: Pro Design Critique and Fixes

    1

    Give your agent a screenshot, code or a page and get a senior-designer critique instead of vague taste comments.

    $4.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    UI Polish Reviewer: Pro Design Critique and Fixes

    UI Polish Reviewer: Pro Design Critique and Fixes

    Example session with this skill installed

    Make this landing hero look professional (desktop screenshot attached).

    • Read your context and instructions
    • Compiled the ui polish reviewer

    Purpose: Not provided. Assumption (unconfirmed): the page exists to make small-team leads start a free trial.

    Primary action: assumed "Get started"; please confirm.
    First seen → second → third (Inferred from the screenshot): gradient and headline → the card row → the buttons are barely noticed.

    Scores (0-5, with evidence):

    • hierarchy 2: two same-size buttons, the gradient outranks the buttons
    • spacing 3: gaps look consistent between the cards, the hero is tight above the subtitle
    • type 2: subtitle looks about 12px, headline is large but generic
    • color 2: gradient behind text, grey subtitle on purple; contrast estimated from the image only
    • alignment 4: everything centered on one axis, cards equal size
    • components/states 2: scored on the visible components only (two identical buttons); states not visible, not counted against
    • responsive n/a (not visible: one desktop screenshot)
    • a11y 2: scored on the small grey text only; focus state not visible, not counted against
      Total 17/35 scored (responsive not scored)

    What is already working: consistent card sizes; the headline is large and readable; the layout is simple to scan.

    AI-look tells: purple gradient hero → flat brand color; three identical cards with emoji icons → one lead block plus two smaller ones, one consistent icon set; "Unlock … seamless" headline → a concrete outcome (taste unless it breaks a defect rule).

    Findings

    Every before line below is assumed from the screenshot; no code was provided. Tokens such as --primary, --text-on-hero and --border-on-hero are proposed and must be defined first (starter scales are in reference/scales.md).

    1. [Major · S] Hero subtitle — grey text, estimated about 12px, on the purple gradient.
      Why: it is the one line that explains the product, and at the 12px floor with an unmeasured grey on purple it risks being hard to read (contrast not measured, no hex values given).

    Fix

    /* before */ .hero p { font-size: 12px; color: grey; }
    /* after  */ .hero p { font-size: var(--text-lg); line-height: var(--lh-body); color: var(--text-on-hero); max-width: 60ch; }
    

    Then measure the real colors and keep at least 4.5:1.
    2. [Major · S] Two buttons identical in size and color: the primary action has no priority.
    Why: users cannot tell which step you want them to take.

    Fix

    /* before */ .btn { height: 40px; background: var(--brand); color: #fff; }
    /* after  */ .btn { height: 48px; padding: 0 var(--space-5); border-radius: var(--radius-md); }
    .btn-primary { background: var(--primary); color: var(--primary-contrast); }
    .btn-secondary { background: transparent; color: var(--text-on-hero); border: 1px solid var(--border-on-hero); }
    

    [Minor · S] Emoji icons are usually inconsistent in size and style (Inferred: only "emoji icons" is visible, not their rendering).
    Why: mixed icon styles make the cards look unfinished.

    Fix

    /* before */ .card .icon { font-size: 1.5rem; }  /* emoji */
    /* after  */ .card .icon { width: 24px; height: 24px; color: var(--primary); }  /* one inline SVG set, same stroke width */
    
    1. [Minor · S, Unverified] Focus and hover states cannot be judged from a screenshot.
      Why: keyboard users need a visible focus ring on every button and link; whether one exists is not visible here.
      Fix (a default to add if absent; no existing code was provided):
      .btn:focus-visible { outline: 3px solid var(--primary); outline-offset: 2px; }
      
    2. [Polish · M, taste] Headline is a generic claim ("Unlock the power of seamless productivity").
      Why: it says nothing a visitor can check, so it does not help them decide.
      Fix: replace it with the real outcome in the owner's words, for example "[what the team can do] in [real time or number the owner can prove]". Do not use a number you cannot back up.
    3. [Polish · M] Three identical cards read as template rhythm.
      Why: equal weight hides which benefit matters most.

    Fix

    /* before */ .cards { display: grid; grid-template-columns: repeat(3, 1fr); gap: var(--space-5); }
    /* after  */ .cards { display: grid; grid-template-columns: 2fr 1fr 1fr; gap: var(--space-5); }
    

    Make the first card the main benefit.

    Options (taste, not defects): keep the purple as a flat brand color instead of a gradient.
    Could not verify: other widths, hover/focus, motion, real colors and contrast ratios (estimated from the image).
    Which items should I apply?

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Fixes the gap between "works" and "looks professional". Give your agent a screenshot, code or a page and get a senior-designer critique instead of vague taste comments.

    • Starts from the page's purpose and primary action, then a squint test: what the eye sees first, second and third.
    • Scores hierarchy, spacing, typography, color/contrast, alignment, components & states, responsiveness and accessibility, with evidence for each score. Areas it cannot see are marked n/a and not counted against you.
    • Flags the "made by AI" design tells (gradient blobs, three identical cards, emoji icons, empty headlines) and gives a concrete replacement for each.
    • Every finding is specific: where it is, what is wrong, why it matters, and exact before/after CSS wherever the fix is code. Findings are ranked by severity and effort.
    • Separates real defects from taste, keeps your brand, and says what it could not verify from a screenshot (hover, focus, other widths, real contrast).
    • Applies only the items you approve. Nothing changes until you say so.

    Tested here, one run each: the same agent was given the same flawed fictional landing page, once with this skill and once without. The skill run scored it 14/40 and listed 15 ranked findings, including buttons built from span elements, light text on the gradient below 4.5:1 (hand-computed), a 12px subtitle, two identical buttons and tap targets under 44px. The page improved without the skill also looked good, so this skill does not promise prettier designs; it adds a scored, evidence-based, ranked review with exact before/after CSS, a list of what could not be verified, and an approval step. Results vary between runs.

    What you get: SKILL.md (procedure), a heuristics guide, an AI-look tells list, starter spacing/type scales and a complete worked example.

    Honest limits: from a screenshot it cannot test hover, keyboard, motion or other widths, and contrast values are estimates unless you give colors or code. It will not invent testimonials, logos or statistics. Tested here as instructions given to a Claude agent; other agents are untested.

    Privacy: instructions only. Nothing is installed and the skill itself makes no network calls.

    Cannot verify states or motion from static images. Contrast requires hex values. Only reviews visible elements. Does not perform network calls or downloads.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 12 days ago

    Permissions

    Read Files
    Write Files

    Read Files: your agent reads the screenshot, code or page you point it at. Write Files: only used to apply fixes you approve, one item at a time; nothing is edited without your OK. The skill itself runs no scripts and makes no network calls.

    Listed12 days ago

    What's inside

    Frequently Asked Questions

    A screenshot (desktop and mobile if you have both), code or a file you share, or a URL you name that your own agent can open. It also asks for the page's purpose, main action and audience, because a design cannot be judged without them.

    When you share code, it uses your existing tokens and class names. When you share only a screenshot, the before lines are labelled as assumed and any tokens are marked as proposed, with starter scales included.

    Not from a static screenshot. Hover, focus, motion and other widths are listed under "Could not verify" and are not scored against you. Contrast values are estimates unless you give colors or code.

    It follows a fixed order every time: purpose first, a squint test, a scored review with evidence, the AI-look tells, then ranked fixes with before/after CSS. It separates defects from taste and applies only what you approve. It cannot do anything your agent cannot do; it gives the review a consistent structure.

    Purpose and primary action, an 8-area score with evidence (areas that are not visible are n/a), what already works, the AI-look tells with replacements, ranked findings with severity and effort, and a "Could not verify" list.

    New in Frontend & Web Apps