Accessibility and layout audit
The free command: unnamed, tiny, off-screen, overlapping, duplicate and truncated controls.
TestedOK reads the accessibility tree to decide what to tap. The same read shows what the app got wrong on that screen: controls with no name, tap targets too small to hit, controls off the edge or on top of each other, two controls a screen reader cannot tell apart, labels cut short. No model, no account, no credits.
testedok audit com.example.shopAccessibility and layout audit · com.example.shop · iPhone 17 Pro
the screen it opens on
✘ button at 355, 60 (top right, beside "Cart") has no name
▫ button "Add" at 114, 74 is 28×28; the minimum is 44×44
≡ 2 buttons named "Edit"; a screen reader cannot tell them apart
1 unnamed, 1 tiny, 1 duplicate · 14 controls
1 unnamed, 1 tiny, 1 duplicate across 1 screen with 14 controls.
Unnamed: a screen reader announces these as just their role …
Tiny: Apple asks for 44×44 pt and Android for 48×48 dp …With an app id it launches the app and audits the first screen. Without one, it audits whatever is on screen right now, so navigate to the screen you care about and run it. --android audits the attached Android device.
The whole app
testedok audit com.example.shop --explore --max-screens 20--explore walks the app the way explore does — tabs first, then rows and buttons, never anything that deletes, pays, sends or signs out, never typing — and audits every screen it reaches, named by the taps that led there. --budget 120 caps the walk in seconds.
What it checks
| Mark | Finding | What it means |
|---|---|---|
| ✘ | unnamed | No accessible name. VoiceOver and TalkBack announce it as "button". Labels TestedOK inferred from nearby text or a view id do not count: no screen reader reads them. |
| ▫ | tiny | A tap target under 44×44 pt (iOS) or 48×48 dp (Android). |
| → | offscreen | Part of the control is past the screen's left or right edge. Vertical clipping is scrolling and is not reported. |
| ▣ | overlap | Two controls share more than 30% of the smaller one's area, neither containing the other. |
| ≡ | duplicate | Two controls on one screen with the same role and name. |
| … | truncated | A label ending in an ellipsis: the text did not fit. |
Each kind ends the report with the fix in SwiftUI and Compose terms.
Some things that look like findings are not, and are left out: the previous screen that iOS parks off the left edge during a push; a system switch, which the platform hits generously; a line of text with a button trait inside a row, which is hit through the row; a small control floating over a big one, such as a bar's button over a list.
In CI
testedok audit com.example.shop --explore --json > audit.json
testedok audit com.example.shop --explore --strict--json prints the screens and findings as JSON (each finding with its kind, role, label, point and detail). --strict exits non-zero when there is any finding, for a job that keeps the audit clean. Over MCP, audit_app does the same for a coding agent, which can then fix the labels it names.
Why it matters to your tests too
The same nameless and duplicate controls are what make flows brittle: the model sees (unlabeled button) or two "Edit"s and has to guess. Naming them fixes the screen reader and the test at once.
What it does not check
Colour contrast, text size and motion need pixels, not the tree; they are not here. A screen is audited only if the run reached it: the unit is screens seen, never "your app".