Visual checks
Code reached tells you a browser ran your component. It cannot tell you the button ended up behind the modal. Two checks look at the rendered page itself.
Audit the rendered page
verifaied audit loads a page of your running app at phone, tablet and desktop widths (375, 768 and 1280 pixels) and measures it. It is free and uses no AI. At each width it reports:
- content that spills outside the viewport,
- text below the WCAG AA contrast minimum,
- images that failed to load,
- errors the browser console logged.
The audit drives a real browser, so it needs the optional verifaied[audit] extra. Pass the URL of the page, and --viewports to choose your own widths.
# One-off install: the audit drives a real browser, so Playwright ships # as an optional extra rather than a dependency of every command. pip install 'verifaied[audit]' playwright install chromium # Point it at the running app. Three widths by default: 375, 768, 1280. verifaied audit http://localhost:5173 # Pick your own widths, and say who's driving for the session ledger: verifaied audit https://staging.example.com --viewports 320,1440 --driver agent # Exit codes: 0 clean, 3 findings recorded, 2 could not run or record.
Findings are saved on the branch with a full-page screenshot for each width. Only the most recent audit counts, so after a fix, run it again: a clean pass clears what the last one found.
| Exit code | Meaning |
|---|---|
0 | Clean: nothing found at any width. |
3 | Findings were recorded. |
2 | The audit could not be run or could not be recorded. |
1 | Usage or configuration error. |
Get a verdict on a screenshot
verify_ui is a tool your coding agent calls.Paid plan The agent takes a screenshot of what it just changed, says what it intended, and verifAIed judges the result.
verify_ui( repository="my-org/my-app", intent="The error state shows a Retry button under the message", target="src/components/ErrorPanel.tsx::ErrorPanel", screenshot_base64=<base64 PNG>, )
The answer is one of three, with the reasons:
pass: the screenshot shows what the agent intended.fail: it does not.unclear: the screenshot did not show the part of the page the intent was about. The agent should capture the right thing and ask again. It is never a pass.
Without an intent, the tool runs a general review instead: layout, hierarchy, legibility, consistency and obviously broken states, returned as a list of defects.
Every verdict is saved on the branch. Pass a target and it is also saved against that component.
Where the results show
- The Visual Verdicts card on a branch's Browser tests page lists each verdict.
- Other activity on the same page lists each audit and each verdict as a session, with its screenshots beside the findings.
- A function's page shows the verdicts recorded against that component.
Three switches decide whether these results block a branch from being done: Visually unreviewed, Intent mismatch and Visual defects. See Definition of done.