Compose Skill (Compose Kit)

A Compose skill for AI coding agents: seven agent skills that make Claude Code, Codex, Cursor, Copilot and Gemini write better Jetpack Compose and Compose Multiplatform code. One consistent house style, measured on seven models in real agent CLIs.

Works with Claude Code · Codex · Cursor · OpenCode · Copilot · Gemini CLI · Antigravity

Website: compose.avinya.dev · Install guides per agent · Results

+30.8points forGemini 3.8 Flashwith the v6.1 kit (v6 test) +23.1points forGPT-6-Lunawith the v6.1 kit (v6 test) 97.4%of rubric items forDeepSeek V4.1 Flashwith v6.1 (cheap add-on model) 90 + 36agentic runs (panel + add-ons);2 blind graders must agree on every item

What we achieved

• Held-out v5 (120 agentic runs): the kit lifted Gemini 3.8 Flash by +25.6 points and Sonnet 5.5 by +18.6, and the test found a real flaw: on a "match our conventions" task, models restructured too much with the kit (results).
• v6.0.0-preview.1 (2026-10-01): seven skills, one entry tree, four install channels.
• Website live (2026-10-02): compose.avinya.dev, with install guides per agent and a build check that every number on it appears in the evidence files.
• v6 A/B (90 agentic runs): v6.0 vs the v6.1 fixes vs no kit. The "no new UI controls, screens or features" item on conform tasks went from 2 of 10 cells (v6.0) to 8 of 10 (v6.1), and every pre-registered rule held (results).
• v6.1.0 stable (2026-10-05): no model scored below its own no-kit result, so it is released as stable and is the GitHub "Latest" release.
• Cheap models (36 add-on runs): DeepSeek V4.1 Flash went from 82.1% to 97.4% of items with v6.1, and DeepSeek V4 Pro from 74.4% to 87.2% (results).

Why

AI agents write Compose that compiles and then drifts: state ends up in the wrong place, every screen gets its own pattern, error handling changes from file to file, and reviews flag perfectly fine code as blocking. Compose Kit gives the agent one house style plus the reasoning behind it, and only loads the parts the current task needs.

• One consistent style: MVI on a shared BaseViewModel contract, Koin annotations, Navigation 3, feature-owned data/domain/presentation layers, and typed error handling.
• Progressive loading: a ~1.3k-token entry tree picks the task path and the exact reference files. The full kit is about 127k tokens; in the v5 test, 35 of 40 kit runs loaded under 20k.
• Respects your project: in a coherent existing codebase (Hilt, MVVM, Navigation 2, your own base class), the kit follows your pattern and proposes migration separately.
• Checked, not trusted: 11 shell guard scripts catch drift in CI, backed by 90 fixture tests.

What's inside

Skill	What it owns
compose	The entry decision tree (~1.3k tokens): task kind → affected area → exact reference files
compose-architecture	Module graph, MVI/BaseViewModel contract, error tiers, state ownership, naming, Koin DI, Navigation 3, coroutines
compose-feature	A screen or feature slice end to end: contract, ViewModel, UI, navigation, DI and tests; review mode
compose-ui	Composables: Route/Screen split, stability, loading/empty/error states, lists, animation, accessibility, resources
compose-data	Repositories, Ktor, Room, DataStore, Paging 3, offline-first, data-layer tests
compose-project	New projects, adopting the kit, modules, convention plugins, version catalog, CI and agent hooks
compose-platform	commonMain vs expect/actual, iOS/Swift interop, desktop and web targets, platform lifecycle

Also included: project and feature templates, and the guard scripts (compose-architecture/scripts/).

How it works

flowchart TD
    P["Your prompt"] --> E["compose entry tree · ~1.3k tokens"]
    E --> K{"What kind of task?new feature · change · bug fix · reviewconform · setup · question"}
    K --> A{"Which areas does it touch?architecture · feature · uidata · project · platform"}
    A --> F["Load one reference file per affected area"]
    F --> C["Code in the house style"]

Install

Install all seven skills. compose is the entry point, and it hands each task to the other six. Step-by-step pages for each agent are on the website: Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI and Antigravity, OpenCode.

Every option below can install for one project (lives in the repo, so your team gets it too) or globally (for you, in every project on your machine).

npx skills

Needs Node 22.20+. Works with Claude Code, Codex, Cursor, OpenCode, Copilot, Gemini CLI, Antigravity and 70+ other agents.

# This project only (run it in the repo root)
npx skills add Meet-Miyani/compose-skill --skill '*' -a claude-code

# Globally, for every project
npx skills add Meet-Miyani/compose-skill --skill '*' -a claude-code -g

Swap claude-code for your agent: codex, cursor, opencode, github-copilot, gemini-cli, antigravity. Repeat -a to install for several at once. Leave it out and you get an interactive picker.

Agent	This project	Global
Claude Code	.claude/skills/	~/.claude/skills/
Codex, Cursor, OpenCode, Copilot, Gemini CLI, Antigravity	.agents/skills/	the agent's own folder, e.g. ~/.codex/skills/

gh skill

Needs GitHub CLI 2.90+. --pin v6.1.0 fixes the v