Developers can guide AI agents to handle common Android pitfalls accurately
This repository reduces errors when agents process complex Android patterns and ensures generated code aligns with developer guidelines. Based on evaluations showing where AI struggles, it provides technical specifications for those areas in Markdown files. You can install only the skills you need, and the agent will automatically reference the relevant guidelines when performing related tasks.
Skills focused on AI weaknesses
The repository excludes basic areas where AI already performs well and focuses on use cases and workflows where evaluations show lower performance. For example, it does not cover basic Jetpack Compose usage but includes technical specifications for complex patterns that agents find difficult. Each skill exists as a SKILL.md file, providing the LLM with information about specific domains and workflows to serve as a basis for its decisions.
Installation via CLI and plugins
Use the Android CLI to install the required skills in the current directory. To install a specific skill, run the command android skills add r8-analyzer --project=.. To install all skills, use android skills add --all. If no agent directory exists, it installs to Gemini and Antigravity by default. If you use Claude Code or Codex, you can also install via the plugin marketplace. After installation, the agent automatically finds and applies the relevant skill when a related task is requested.
Verification and contribution
Since AI can make mistakes, you must always verify the generated results. This repository does not accept public contributions; feedback or requests for new skills are accepted via GitHub issues. The project follows the Apache License 2.0 and adheres to Google's open source community guidelines. It is suitable for teams or individual developers adopting agent-based development workflows, but the user remains responsible for final verification of AI output.
