What problem does it solve? Converting design specs from Figma, Google Stitch, or Claude-generated design documents into a consistent, maintainable Kotlin Compose theming system is manual and error-prone. This Skill automates the extraction of colors, typography, spacing, and shapes into a three-layer token architecture (Primitive, Semantic, Component) with generated Kotlin code. ## Core Features & Use Cases - Multi-Source Token Extraction: Accepts exactly one seed source per run — Google Stitch (via MCP), Figma (via MCP or exported W3C DTCG tokens.json), or Claude Design specs/artifacts — and rejects all other inputs with clear guidance. - Three-Layer Token Architecture: Generates Primitive objects, Semantic data classes with dark/light variants, and per-Component token classes wired through CompositionLocal providers and a Material 3 bridge. - Screen Recreation Pipeline: Enumerates screens from the source, tracks build status in a tracking file, recreates each screen in order, and delegates implementation to a kta-compose-developer agent. - Validation & Auditing: Includes scripts to detect project configuration, audit token consistency, and find hardcoded values in existing Compose code. - Use Case: A developer receives a Figma file URL, invokes the skill, and gets a complete theme module with token files, P0 components (Surface, Text, Background, Buttons), recreated screens, and a token showcase screen. ## Quick Start Ask the assistant to generate Compose design tokens from your Figma file URL, Stitch project link, or Claude design spec path, then confirm which screens to implement.