Compose Performance Skills

If you have used android/skills, you already know the shape of an Agent Skill: a Markdown file an agent loads on demand to perform one focused task. That library covers Android development at large. This repository goes narrow and deep on a single axis: Jetpack Compose performance. The skills here cover stability, recomposition, lazy layouts, custom modifiers, side effects, baseline profiles, R8, hot reload with Compose HotSwan, and the measurement loop that ties them together. The goal is to give a coding agent enough operational knowledge to diagnose and fix Compose performance issues without guessing, and to give the developer reading the diff a citation trail back to primary sources.

What this repo is

This is a curated library of Agent Skills focused on Jetpack Compose performance. Every skill is grounded in primary sources: Android Developers documentation, the Compose compiler, posts by Ben Trengrove, Chris Banes, Manuel Vivo, and the skydoves open source projects, and blog posts. Every API reference is pinned to a version.

The skills follow the open Agent Skills standard published at agentskills.io, the same SKILL.md format used by Anthropic's Skills API, Android Studio Agent mode, and Gemini. The library was iterated against Claude Code, where the SKILL.md files load directly; other compatible runtimes have not been individually end to end tested.

Note that the Android CLI itself manages only Google's first party skill catalog (android skills add --skill ); this repo is a community library outside that catalog and is loaded by agent runtimes from project local directories rather than installed via the CLI.

What is a Skill

A Skill is a single Markdown file (SKILL.md) plus optional references/ material that teaches an agent how to perform one focused task. It declares trigger vocabulary in YAML frontmatter and a numbered workflow in the body. The agent reads the frontmatter, decides whether the skill applies to the current task, then follows the workflow inside SKILL.md step by step.

A skill is not documentation for humans. It is operational instructions for an LLM. That changes the writing voice: terse, imperative, RIGHT and WRONG snippet pairs, MUST and MUST NOT directives in bold caps, and a Verification checklist that proves the work was done. Prose is kept short. Code samples carry the load.

Directory layout

compose-performance-skills/
├── README.md                 # this file
├── INDEX.md                  # symptom to skill lookup table
├── LICENSE                   # Apache-2.0
└── //
    ├── SKILL.md              # the skill (required)
    └── references/           # optional, one level deep
        └── .md

Conventions:

• Category: lowercase, one of stability, recomposition, lists, modifiers, side-effects, measurement, build, audit, hot-reload.
• Slug: gerund form, lowercase, hyphenated, max 64 chars (`diagnosing-compose