improve-animations

Find the highest-value fixes in motion that should remain. Adapted from Emil Kowalski; source and compatibility. Audit/plan requests return findings without source edits. Save plans only when requested. Explicit implementation requests continue through animate and the project's delivery gates, in the current context; no automatic executors, agents or worktrees.

Audit

1. Map the requested surface, installed motion library/version, component primitives, existing tokens, input paths and product context. Identify expected frequency without inventing usage counts. Respect documented tradeoffs; confirm each cited location.
2. Load animate and its craft reference. Audit purpose/frequency, easing/duration, physicality/origin, interruption, rendering cost, accessibility, token cohesion and genuine missed opportunities. Apply these axes inline. For unnecessary motion, route to remove-animations rather than polishing it.
3. Vet findings against actual state changes and existing decisions. Distinguish reproducible defects from craft suggestions and unmeasured performance suspicions. Rank by user impact and effort; show severity, category, file:line, evidence and fix. An empty result is valid; do not force additions or fixed finding counts.
4. When plans are requested, make each independently executable: current revision, problem/current excerpt, exact targets and token values, repository exemplar, ordered edits, scope/dependencies, rollback and focused verification commands. Include rapid reversal, keyboard/touch, reduced motion and a real feel-check; name any device evidence still needed. Revalidate stale plans against current code.

For the sequence remove-animations -> improve-animations -> animation-performance, subtract first, assess retained motion second, then profile the remaining frame problems. The sequence preserves the requested endpoint; an audit does not authorize fixes.

Load animation-performance for frame bottlenecks; static code inspection cannot prove smoothness. Load dependencies through the dependency-loading contract.