ayush016/android-lead-agent-skills

android-lead-agent-skills is a package of AI coding skills written for the role of an Android lead engineer. The idea, as the README puts it, is that Claude, GitHub Copilot, Gemini, and Cursor already know what Hilt, Room, and Jetpack Compose are, but on different days they make different architectural choices, so a codebase slowly drifts. This repo installs once and acts as a team standards document for the AI, so every session in every developer's editor and every CI run produces code that follows the same patterns. The skill is MIT-licensed and targets Kotlin 2.0.21, Compose BOM 2024.09.00, full Material 3, and a minimum Android SDK of 26. A companion sample app called luminary-android is linked from the top of the README, and a quick-start install section is provided. The comparison table near the top sets out what the skill changes. Without it, the AI might use six to eight Material colour roles with hardcoded hex values, a generic CircularProgressIndicator for loading, a flat var isLoading and var error and var data for state, and a single StateFlow for everything (which causes toast events to replay on rotation). With the skill, the AI produces all 29 Material 3 colour roles in a three-layer token architecture, skeleton-shimmer loading that mirrors content shape, a sealed interface with Loading, Ready(data), and Error(message), and SharedFlow with replay zero for one-time events. UI ships are blocked by a 29-item binary checklist. Coverage spans eight collapsible sections in the README. Architecture and state covers MVVM, sealed interfaces, Hilt component layering, and clean-architecture import rules. Theming details a primitive-to-semantic-to-component token system and the Material Theme Builder workflow. Shared element transitions includes four complete Compose patterns (grid-to-hero, card expansion, FAB morph, text continuation), scope threading guidance, and predictive back support. Motion and animation gives spring tuning, stagger formulas, and choreography rules. Navigation covers type-safe routes with @Serializable, multi-module rules, and deep-link wiring. The remaining sections cover adaptive layouts via WindowSizeClass and NavigationSuiteScaffold, coroutines and Flow patterns with Turbine testing, a data layer built around stale-while-revalidate reads and an outbox pattern for writes with Room and DataStore, and security with EncryptedSharedPreferences, BiometricPrompt, and certificate work. Each section links to its own references markdown file inside the repo.