♿ Accessibility
WCAG AA/AAA contrast ratios, touch-target sizes, focus-state visibility, and text legibility at real rendered sizes.
Your agent ships UI fast — but it can't see what it built. Agent Design Check analyzes screenshots and screen recordings and returns concrete, prioritized findings — contrast, spacing, hierarchy, accessibility, flow — in plain text or JSON your agent can apply on the next pass.
$ npm install -g agent-design-check
$ agent-design-check register you@company.com
# check your inbox, confirm, done — 10 free analyses
$ agent-design-check analyze ./checkout-flow.mp4
Launching soon · Contact us — 10 free analyses at launch · No credit card · Node 18+
Not a vague score — a prioritized worklist you can act on today.
Analyzing checkout-flow.mp4 (0:42, 6 key screens) ...
HIGH 01 Primary CTA "Place order" fails WCAG AA contrast
(#8A8F98 on #F4F4F5 → 2.6:1, needs 4.5:1)
HIGH 02 Error state for card number appears below the fold
on 13" screens — users in the recording scrolled past it
MEDIUM 03 3 different spacing scales on the payment form
(12 / 14 / 18px) — pick one rhythm
MEDIUM 04 Step indicator says "Step 2 of 4" but the flow has 5 steps
LOW 05 "Submit" and "Place order" used for the same action
on consecutive screens
5 findings · 2 high · full report: ./agent-design-check-report.html
9 free analyses remaining — run `agent-design-check upgrade` anytime
One global package, no config files. Works on macOS, Linux, and Windows with Node 18+.
agent-design-check register you@email.com sends a confirmation link. Click it and your terminal is authenticated — no passwords, no dashboard required.
Point it at a screenshot, a folder of screenshots, or a screen recording. Recordings get frame-by-frame analysis plus flow-level findings.
WCAG AA/AAA contrast ratios, touch-target sizes, focus-state visibility, and text legibility at real rendered sizes.
Inconsistent margins, broken alignment grids, mixed spacing scales, and elements that crowd or float.
Competing CTAs, headings that don't rank, and screens where nothing tells the user what to do first.
Inconsistent terminology, vague button labels, jargon, and error messages that don't say how to recover.
From recordings: dead ends, backtracking, steps that take too long, and state changes users visibly missed.
--json output your coding agent can consume directly in its loop, and a --ci flag to fail builds on high-severity findings.
Don't want to run anything? Send it to us — we'll do the QA for you.
Send us your screenshots and screen recordings directly and get back a prioritized findings report: accessibility violations, layout and spacing issues, hierarchy problems, confusing copy, and flow dead-ends — each with a concrete recommended fix.
$0
$12/month
at launch: agent-design-check upgrade
Screenshots as PNG, JPG, or WebP, and screen recordings as MP4, MOV, WebM, or GIF. For recordings, Agent Design Check samples key frames and also analyzes the interaction flow between them — so it can flag things a single screenshot never shows, like backtracking or missed state changes.
Agent Design Check is launching soon. Contact us with your email and we'll notify you the moment it ships — you'll get 10 free analyses at launch, no credit card required.
Yes — our automated QA service accepts screenshots and screen recordings directly. Contact us to send your captures and you'll receive a prioritized findings report, no installation required.
Run agent-design-check register you@email.com. You'll get a confirmation email; clicking the link authenticates your terminal. No password, no separate dashboard signup. Your 10 free analyses start immediately.
The CLI tells you how many free analyses remain after each run. When you hit zero, subscribe to Pro with agent-design-check upgrade (opens a checkout page) or from the pricing page. Your history and settings carry over.
Uploads are processed for analysis and deleted after your report is generated, unless you enable history tracking on the Pro plan. See the privacy policy for details.
Yes — use agent-design-check analyze ./screens --ci --json. It exits non-zero when findings at or above your configured severity threshold are present, which makes failing a PR on UI regressions a one-line addition to your pipeline.
Want the QA service, a heads-up when the CLI ships (10 free analyses at launch), or just have a question? Leave a message — a human replies.
We'll get back to you within one business day.