Testing Animations Deterministically — autoAdvance = false Or Bust

A Compose animation test that does not pause the clock is by definition flaky. With mainClock.autoAdvance = true (the default), the framework's InfiniteAnimationPolicy throws CancellationException as soon as an indeterminate animation starts, and finite animations finish in a single auto-advanced burst with no observable intermediate state. This skill teaches the five-line recipe that fixes that, plus the runOnUiThread-vs-runOnIdle gotcha that bites every developer who tries it for the first time.

When to use this skill

• The test asserts an intermediate frame of an animation (mid-fade alpha, mid-scroll offset, mid-Crossfade dispose).
• The composable contains an indeterminate animation (LinearProgressIndicator() with no progress argument, rememberInfiniteTransition, looping withFrameMillis).
• The developer reports CancellationException("Infinite animations are disabled on tests") and asks how to make it stop.
• The developer reaches for Thread.sleep(500) to "wait for the animation".
• The developer's animation test passes locally and fails on CI, or vice versa.
• The developer mentions autoAdvance, advanceTimeByFrame, advanceTimeBy, Crossfade, "flaky animation test", or InfiniteAnimationPolicy.

When NOT to use this skill

• The animation is a side effect; the test only cares about the final state. Default autoAdvance = true is faster — see ../synchronizing-with-idle/SKILL.md.
• The clock semantics themselves are unclear (frame model, rounding rules, v1 vs v2 dispatcher). Read ../controlling-the-test-clock/SKILL.md first.
• The condition under test is not Compose state (a Job.isCompleted, a Mockito.verify). Use waitUntil or an IdlingResource from ../synchronizing-with-idle/SKILL.md.
• The test uses screenshot comparison waiting on the RenderThread for ripple/elevation pixels. That is the one legitimate Thread.sleep site (skydoves hot take #7).