What problem does it solve? Large React feature folders often become a single fat component tree controlled by mode, readOnly, and variant flags, forcing every host (page, portal, share, micro-app) to ship code it never uses. This Skill guides the restructuring of such domains into independently mountable atoms with sunk state, so each host ships only what it imports. ## Core Features & Use Cases - Atom Grain Analysis: Defines the right split granularity — the smallest unit a host is allowed not to mount — with concrete too-coarse/too-fine examples. - State Sinking Rules: Specifies which state (fetch, view model, transient UI, mutations) moves into atoms and which stays on the assembler, preventing hidden module-graph coupling. - Kit Taxonomy: Classifies pieces as domain primitives, data access, read/write/workflow atoms, domain slots, host seams, and assemblers, with ownership rules for each. - Use Case: A Conversation feature must ship to both the in-app page and a lightweight share host. Apply the 10-step procedure to extract read atoms, isolate the shared SWR hook, and rewrite each host as a pure import list, then prove the cut with a module trace. ## Quick Start Ask the AI to split the heavy feature folder into composable atoms with sunk state so the light host only imports read atoms by deep path.
