Jaewoong Eum Releases Agent Skills to Automate Jetpack Compose Performance Audits

Jaewoong EumJaewoong Eum

Jaewoong Eum published Compose Performance Skills, a library of 26 open-source agent skills designed to help AI coding assistants optimize Android UI performance. By providing structured operational instructions grounded in primary sources, the library prevents AI hallucinations and enables agents to autonomously diagnose and fix complex Jetpack Compose issues.

Jaewoong Eum, a Google Developer Expert for Android, released Compose Performance Skills—a library of 26 open-source agent skills (structured instructions that teach AI models how to perform specific tasks) for Jetpack Compose. The collection covers stability and recomposition optimizations, following the same format as Google Labs' stitch-skills library for interoperable AI capabilities.
Total skills
26
Skill categories
9
Supported runtimes
Claude Code, Android Studio Agent, Gemini
File format
SKILL.md (Markdown)
License
Apache-2.0
Authoring standard
agentskills.io

While general-purpose models often struggle with niche Android performance metrics, these skills provide grounded operational workflows derived from official documentation and compiler behavior. This shifts the AI from a conversational assistant to a specialized performance engineer capable of parsing compiler reports and applying precise fixes without guessing.

You can install the library into Claude Code, Android Studio Agent, or Gemini using the provided idempotent shell scripts to symlink skills into your project. The repository is available on GitHub under the Apache-2.0 license, allowing you to automate end-to-end performance audits from measurement to verification in release builds.

View the full update on github.com
Jaewoong Eum
Jaewoong Eum
@github_skydoves
X

Published a new open-source project, Compose Performance Skills. ⚡️ This is a curated library of Agent Skills focused on Jetpack Compose performance. Every skill is grounded in primary sources. https://t.co/NZkTHZuRSA

15retweets113likes
View on X

Still wondering? A few quick answers below.

Compose Performance Skills is a curated library of 26 Agent Skills designed to help AI coding assistants optimize Jetpack Compose performance. Unlike human documentation, these are operational instructions grounded in primary sources like the Android Developers documentation and the Compose compiler. They provide agents with specific workflows and code snippets to diagnose and fix UI performance issues.

To use these skills with Claude Code, you must place them in a specific directory. The repository includes a shell script that clones the source and creates symlinks for each skill folder into the required local directory. Once the symlinks are in place and the agent is restarted, it can automatically load the relevant skills based on your prompts.

These skills follow the open Agent Skills standard, making them compatible with several AI runtimes. They are specifically tested for Claude Code, where the files load directly. They also support Android Studio Agent mode and Gemini, which discover skills by scanning project-local directories. Any agent capable of reading Markdown and following YAML-based trigger instructions can potentially use them.

The library covers nine critical categories of Jetpack Compose performance. These include stability, recomposition, lazy layouts, custom modifiers, and side effects. It also provides skills for build configurations like R8 and baseline profiles, as well as measurement tools like Macrobenchmark. An orchestrator skill is included to sequence these individual tasks into a complete end-to-end performance audit.

Yes, the Compose Performance Skills library is an open-source project released under the Apache License 2.0. It was designed and developed by Jaewoong Eum, also known as skydoves. The project is hosted on GitHub, where developers can contribute new skills or report issues. Every skill in the library is grounded in authoritative primary sources to ensure technical accuracy.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update