More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Look Before Done: check the real result before you finish

    1

    Before your agent says a visible change is done, it opens the real page, screen, email or document at phone and desktop sizes and fixes what only shows when you look.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Look Before Done: check the real result before you finish

    Look Before Done: check the real result before you finish

    Example session with this skill installed

    I've finished the "Order your print" page. Before you call it done, look at it the way a customer would, on a phone and on a desktop, and fix anything that's wrong.

    • Read your context and instructions
    • Compiled the look before done

    What I looked at

    The real "Order your print" page, loaded fresh in a browser at phone width (375 pixels) and desktop width (1280 pixels), scrolled to the very bottom. The page has no dark mode, so there was none to check.

    What I found: 5 problems

    Words spill out of their box. "Free returns within 30 days" runs past the edge of the Returns box. Phone width only.
    2. "Buy now" is almost invisible. Near-white words on a white button.
    3. A broken picture. "The print in its frame" shows only its label; the picture's address is wrong.
    4.

    The top runs off the screen. The top section has a fixed width and the tagline is cut off on a phone. Phone width only.
    5.

    Placeholder text at the bottom. "Lorem ipsum dolor sit amet. TODO: shipping details." Only seen by scrolling to the very end.

    What I fixed

    • The top section and the Returns box now fit the screen, and their words wrap instead of running off.
    • "Buy now" is white on dark purple.
    • The picture's address points at the real file, and the framed print shows.
    • The placeholder is replaced with real words: "Ships in 3 to 5 working days, tracked all the way."

    Looked again

    Both widths, top to bottom: clean. A real run on a planted practice page, 23 Oct 2026.

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

    About this skill

    The tests pass, the code reads well, and the page still has white words on a white button. Look Before Done makes looking part of finishing. Your agent opens the real result with whatever browser, preview or screenshot tool it already has, checks it at phone and desktop widths, in light and dark mode and in the states the change touched, and scrolls to the bottom. On every view it runs one checklist: text cut off or spilling out of its box, colours too faint to read, broken or stretched pictures, layouts that run off a phone screen, leftover placeholder text. Then it fixes what it found, looks again, and tells you what it looked at and anything it could not check.

    Tested on a practice page with five planted problems: it caught all five, including one that only shows at phone width and one that only shows at the very bottom of the page.

    Instructions only: no scripts, nothing installed, nothing sent anywhere.

    Created by Greygray, with AI collaboration openly included in the process.

    It needs a way to see the result: a browser, a preview or a screenshot tool your agent already has, or a browser on the computer it can run headless. It does not install one without asking. If nothing can show the result, it says so, lists what you should look at, and does not call the work done. It checks what can be seen; it does not replace your tests, and it does not measure accessibility against a standard.

    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 27 days ago

    Permissions

    Browser
    Write Files
    Terminal / Shell
    Read Files

    Terminal / Shell: to take a screenshot with a browser the computer already has, or start the project's own preview. Read Files: to read the page or document being checked. Write Files: to fix what it finds, and to save screenshots. Browser: to open the real page and look at it.

    Listed27 days ago
    Updated22 days ago

    What's inside

    Frequently Asked Questions

    Tests check that the code does what it should. This checks what a person sees: words cut off, colours too faint to read, broken pictures, layouts that run off a phone, leftover placeholder text. It is the step after the tests pass.

    Any agent that can open a browser or take a screenshot: a built-in browser, a browser extension, a Playwright or Puppeteer tool, a preview server, or a headless browser already on the computer. It does not depend on how the page was built.

    Yes. It checks at least a phone width (about 375 pixels) and a desktop width (about 1280), adds a tablet width if the layout changes there, and checks light and dark mode if the product has both.

    One file, SKILL.md: how to find a way to see the result, which sizes and states to look at, the checklist, and how to report what was seen. No scripts and nothing to install.

    Your agent needs some way to see the result: a browser, preview or screenshot tool. If it has none, the skill says so plainly and does not call the work done.

    Yes. It fixes what it found, then looks again at the views that changed, and stops when a full pass finds nothing. Then it tells you what it looked at, what it fixed, and anything it could not check.

    New in Testing & Debugging