android-wcag-guide

Android (Compose) WCAG 2.2 AA — quick reference

Platform-level accessibility for Jetpack Compose: the androidx.compose.ui.semantics APIs and the AA criteria they satisfy. Framework-agnostic — a design system layered on top may bake some of this in, but the techniques here are the platform baseline.

When this skill fires:

1. Decide what the screen reader should perceive, in what order, and how state changes are announced — then reach for the matching semantics API below.
2. Source every user-facing label from stringResource, never a literal.
3. Load the references/*.md file matching the question, not everything upfront.

Essentials

• Label non-text content - meaningful graphics get a localized contentDescription; decorative ones get null, see references/labelling.md
• Control reading order and headings - traversalIndex, isTraversalGroup, heading(), collection semantics, see references/focus-order.md
• Announce state and status - stateDescription, liveRegion, announceForAccessibility, CustomAccessibilityAction, onClickLabel, see references/state-and-announcements.md
• Meet contrast and never rely on color alone - 4.5:1 text / 3:1 UI, pair color with icon or label, see references/contrast-and-color.md
• Scale text and size touch targets - sp text that honors fontScale, ≥48dp targets via minimumInteractiveComponentSize, see references/text-and-targets.md
• Assert and enforce accessibility - semantics + a11y-text assertions, ATF (where present), lint + CI, see references/testing.md

Gotchas

• contentDescription is a localized user-facing string — source it from stringResource, never a literal. testTag is a separate dev-facing concern; never reuse one as the other.
• An empty semantics(mergeDescendants = true) { } with no contentDescription collapses children into raw concatenated order and drops the intended label — always set the consolidated description.
• contentDescription has an accumulating merge policy: setting it on a mergeDescendants parent does not replace children, it appends to whatever the children already expose (badge counts, labelled icons). When children carry their own descriptions, use clearAndSetSemantics { } to get one clean label instead of a doubled-up one.
• graphicsLayer { alpha = 0f } (or any fade) hides a node visually but leaves it in the accessibility tree, so TalkBack still reads it — common cause of a collapsing header announcing its title twice. Gate the hidden copy with Modifier.semantics { hideFromAccessibility() } (older Compose: invisibleToUser()).
• Putting the same text in an Assertive liveRegion and a contentDescription makes TalkBack read it twice — pick one.
• clickable { } on a Row/Box/Card exposes no role by default, so TalkBack does not announce it as actionable — pass clickable(role = Role.Button) { } (or set role in semantics). Without onClickLabel it also announces only a generic "double tap to activate" — name the action verb.
• heading() is not implied by text size or font weight — set it explicitly on screen titles and section headers, or heading navigation finds nothing.
• Don't put controls (tabs, buttons, chips) in the heading rotor — heading() is for content sections only; tabs already carry Role.Tab. Conversely, in a list of cards, mark each item a heading so users skim with the heading gesture instead of swiping every card and its button.

Example — one focusable, labelled, actionable node for a composite tile

val recap = stringResource(R.string.event_recap, eventTitle, date)   // localized, prebuilt
val newLabel = stringResource(R.string.badge_new)

Row(
    modifier = Modifier
        .clickable(onClickLabel = stringResource(R.string.action_open_event)) { onOpen() }
        .clearAndSetSemantics {                       // replace children: announce as ONE node
            role = Role.Button
            contentDescription = recap
            if (isNew) stateDescription = newLabel
        },
) {
    Icon(painterResource(R.drawable.ic_calendar), contentDescription = null)  // decorative
    Column {
        Text(text = eventTitle)
        Text(text = date, style = MaterialTheme.typography.bodySmall)
    }
}

Progressive Disclosure

Each reference is a trigger — read it only when the user's intent matches; do not preload everything.

• Read references/labelling.md - Load when adding or fixing contentDescription, deciding decorative vs meaningful, or consolidating a composite control with clearAndSetSemantics / mergeDescendants (WCAG 1.1.1, 4.1.2).
• Read references/focus-order.md - Load when the screen reader reads elements in the wrong order, when distinguishing accessibility traversal order from keyboard focus order vs composition order, when adding headings / traversal grouping / list-collection semantics, or for programmatic focus, bring-into-view, focus-first-error, and IME traversal (WCAG 1.3.1, 1.3.2, 2.4.3, 2.4.6, 2.4.7, 3.3.1, 2.1.1).
• Read references/state-and-announcements.md - Load when a toggle/selection state isn't announced, when announcing loading / errors / transient events, or when adding custom or named actions (WCAG 4.1.2, 4.1.3, 2.1.1).
• Read references/contrast-and-color.md - Load when checking contrast ratios or when meaning is carried by color alone (WCAG 1.4.1, 1.4.3, 1.4.11).
• Read references/text-and-targets.md - Load when text doesn't scale with system font size, content clips at large scale, or tap targets are too small (WCAG 1.4.4, 1.4.10, 2.5.5, 2.5.8).
• Read references/testing.md - Load when writing accessibility tests, enabling automated a11y checks, or wiring lint / CI enforcement.