Screenshot
Open a local preview and capture the rendered page at desktop (1440px) and mobile (390px). The agent sees what ships, not what should ship.
Automated visual QA agent
LOOPSCOPE screenshots your build at 1440px and 390px, critiques every frame against your brief — hierarchy, spacing, contrast, broken images, mobile overflow — fixes what it finds, and runs the loop again until nothing changes.
The difference between code that should work and code that does work is that something looked at the rendered page. That something is the loop.
Open a local preview and capture the rendered page at desktop (1440px) and mobile (390px). The agent sees what ships, not what should ship.
Grade each frame against your brief: visual hierarchy, spacing rhythm, contrast ratios, broken images, and horizontal overflow on mobile.
Edit the .astro and CSS to resolve every flagged defect — no guessing, each change is tied to a screenshot the agent can re-verify.
Re-screenshot, re-critique. The loop only exits when a full pass surfaces zero visual defects across both viewports.
Switch viewports to see what the agent captures, then run the loop and watch it clear every seeded defect. Nothing here talks to a server.
No horizontal overflow at this width
Each pass scores the rendered page across these defect classes at both widths. A class only clears when the screenshot proves it.
Flags competing headlines, weak primary CTAs, and sections where the eye has nowhere to land first.
Catches collapsed gaps, inconsistent vertical rhythm, and crowding that reads as unfinished.
Measures text/background ratios and raises anything under 4.5:1, before a real user squints at it.
Detects 404 assets and missing sources — the silent breakage that screenshots expose instantly.
Captures at 390px and flags any horizontal scroll, clipped CTA, or element pushing past the viewport.
Compares the rendered page to the brief you wrote, so "looks fine" is graded against intent, not vibes.
It opens a local preview and drives a real browser to capture the rendered DOM at desktop and mobile widths. It reviews the same pixels a visitor would, not the source code.
When a full pass across both viewports surfaces zero visual defects. If a fix introduces a new issue, the next pass catches it — the loop keeps going until it converges.
No. It grades layout and rendering, not content truth. Placeholder copy and stats are yours to replace — the loop only guarantees the page looks right, not that the claims are.
Anything that serves a local preview over HTTP. Astro, Next, plain HTML — if a browser can load it, the loop can screenshot, critique and fix it.
Join the LOOPSCOPE waitlist and put the screenshot-critique-fix loop on your next landing. We’ll send your access when the next batch opens.