android/skills: Strengthening Android agent workflows with SKILL.md
Adds SKILL.md workflows to Android agents, focusing on development tasks where LLMs underperform.
GitHub android/skills Updated 2026-10-17 Branch main Stars 8.1K Forks 536
Kotlin Python Android agents Android CLI SKILL.md

🧭 Decision Guide

Why trending now: The README emphasizes AI-optimized modular SKILL.md files, the open-standard agent skills format, and Android workflows where LLMs underperform. The project also has release v1.0.15 and gained 36 stars that day. The material supports these signals but cannot establish a single reason for trending.

Try it if you

  • You want Gemini or Antigravity to perform specialized tasks with Android best-practice guidance.
    The README's “Android skills” section says the skills help agents understand and execute Android development patterns following developer.android.com guidance.
  • You need Android CLI to install skills into a project root or for selected agents.
    The README's “Install Android skills” section provides android skills add, --project, and --agent options.
  • Your Android workflow uses Claude Code or Codex plugins.
    The README's “Install as an agent plugin” section provides installation commands for Claude Code and Codex.

Skip it if you

  • You mainly need basic Jetpack Compose best practices rather than Android areas where LLMs perform poorly.
    The README's “Android skills” section explicitly says it does not prioritize well-established areas such as basic Jetpack Compose best practices.
  • Your team requires direct public submission of code, skills, or changes.
    The README's “Contributing” section explicitly states, “Public contributions are not accepted at this time.”
  • You have no Android agent environment and do not use Android CLI, Claude Code, or Codex.
    The README lists installation paths only for Android CLI, Gemini, Antigravity, Claude Code, and Codex.

Requirements

  • Android CLI is required; the README presents it as the way to install a specific skill.
  • If no agent is specified and no existing agent directories are present, the README says installation goes to ~/.gemini/antigravity/skills for Gemini and Antigravity.
  • To install all skills, use the README-defined --all option; by default, only the android-cli skill is installed.

First step (verbatim from README)

android skills add --all

Watch out

  • android skills add --all installs all Android skills for every detected agent.
    The README's “Install Android skills” section says --all adds all skills and that omitting --agent targets all detected agents.
  • When targeting a project, use the README's --project option with a project root.
    The README defines --project as “Path to a project root in which to install the skills”.
  • After installation, you do not manually select a skill; the agent activates relevant skills automatically.
    The README's “Activate skills” section says the agent automatically activates skills relevant to the task.

Not stated in the README

  • The README does not specify a minimum Android CLI version or installation method.
  • The README does not list which specific Android skills are included in v1.0.15.
  • The README provides no evaluation metrics, accuracy figures, or comparison with other agent instructions.
  • The README does not define compatibility beyond Gemini, Antigravity, Claude Code, and Codex.
  • The README does not explain which functions are implemented in Kotlin versus Python.

💡 Deep Analysis

6
No I mainly maintain backend or iOS projects, do not have Android CLI, and do not use Gemini, Claude Code, Codex, or Antigravity for Android development; is this project suitable for my daily coding work?
For: A developer mainly maintaining backend or iOS projects without an Android CLI or Android agent workflow

No, it is not suitable because its knowledge, installation flow, and activation model are centered on Android development and compatible agents.

  • The README defines the output as Android-specific modular agent instructions and resources grounded in official Android guidance.
  • The installation entry point is Android CLI, while the plugin paths explicitly name Claude Code and Codex; these are not generic backend or iOS integrations.
  • The project deliberately focuses on specialized Android cases where LLMs underperform rather than broad general-purpose capabilities.
  • Apache License 2.0 permits reuse, but the license does not turn Android skills into iOS, backend, or general software-engineering skills.
  • Android skills: AI-optimized, modular instructions ... on Android development
  • Install Android skills: Use Android CLI to install a specific skill
  • Install as an agent plugin: Claude Code and Codex
  • Android skills: development focuses on use cases and workflows where evaluations show LLMs underperform
  • Project data: Apache License 2.0
Not stated in the README：The README does not say whether iOS, backend, or other non-Android skills are planned.；The README does not say whether these skills can be loaded by coding tools that do not support the Agent Skills standard.
Yes I use Gemini or Antigravity for Android development and only want an agent to perform R8 analysis in one project rather than install every skill; is this project suitable for me?
For: An Android developer using Gemini or Antigravity who needs R8 analysis in a single Android project

Yes, it is suitable because the README directly provides an R8-analysis skill and project-scoped installation.

  • r8-analyzer can be added independently, so you do not need to install every Android skill.
  • --project=. installs the skill for the current project, which is appropriate when the project’s agent should discover it locally.
  • The skill uses SKILL.md to provide task specifications and specialized context, and the agent activates relevant skills automatically.
  • However, it is an agent-instruction resource rather than R8 itself, so it cannot guarantee correct configuration; the README says AI results must be double-checked.
  • Install Android skills: `android skills add r8-analyzer --project=.`
  • Android skills: markdown files (SKILL.md) that provide a technical specification of a task
  • Activate skills: Your agent automatically activates skills that are relevant to your task
  • Disclaimer: AI can make mistakes, so always double-check the results
android skills add r8-analyzer --project=.
Not stated in the README：The README does not specify which R8 rules are covered, AGP compatibility ranges, or evaluation accuracy for r8-analyzer.；The README does not specify required Gemini or Antigravity versions.
Yes I maintain an Android CLI integration environment with multiple existing agents and want to install all Android skills at once while letting the CLI choose locations from existing agent directories; is this project suitable?
For: A tooling engineer responsible for Android CLI and agent integration who needs to install skills for multiple existing agents

Yes, it is suitable because Android CLI supports bulk installation and installs for detected agents when no agent is specified.

  • --all adds all Android skills, which fits a one-step deployment across the environment.
  • Without --agent, the CLI installs skills for all detected agents, matching an environment with multiple existing agent directories.
  • --project can change the scope to a project root, showing that both project-level and user-level installation are supported.
  • The README only describes detection behavior, not diagnostic APIs; if no directories exist and no agent is specified, the default locations are Gemini and Antigravity directories.
  • Install Android skills: `android skills add --all`
  • Options: `--agent` installs for a comma-separated list of agents; if omitted, skills are installed for all detected agents
  • Options: `--project` specifies a project root in which to install the skills
  • Install Android skills: without existing agent directories or specified agents, skills are installed for Gemini and Antigravity
android skills add --all
Not stated in the README：The README does not list the complete set of agents that the CLI can detect.；The README does not explain how path conflicts between agents are handled or show detection-failure output.
Yes I use Claude Code and Codex in the same Android workflow and do not want to maintain separate prompts and configurations; can this project integrate with both agents through plugins?
For: An Android developer using both Claude Code and Codex who wants Android-specific skills through plugins

Yes, it is suitable because the README provides separate plugin installation paths for Claude Code and Codex, while the skills follow an open standard.

  • Claude Code has a dedicated /plugin install command, while Codex uses separate marketplace and plugin commands.
  • Skills are expressed as Markdown SKILL.md task specifications rather than separate prompt sets for one model.
  • Agents discover and activate relevant skills automatically, allowing both tools to use the same Android-oriented resources.
  • However, the README does not promise identical behavior across the two agents or list compatible versions; successful plugin installation does not guarantee every skill will auto-activate.
  • Install as an agent plugin / Claude Code: `/plugin install android-skills --marketplace android/skills`
  • Install as an agent plugin / Codex: `codex plugin marketplace add android/skills` and `codex plugin add android-skills@android-skills`
  • Android skills follow the open-standard agent skills
  • Activate skills: Your agent automatically activates skills that are relevant to your task
/plugin install android-skills --marketplace android/skills
Not stated in the README：The README does not specify minimum Claude Code or Codex versions or a per-skill compatibility matrix.；The README does not say whether both plugin entry points expose exactly the same skill set.
Yes I lead an Android project that uses AI agents and want skills installed at the project root so team members use them with the same codebase instead of relying on individual user directories; is this project suitable?
For: A team lead who wants Android skills shared with the codebase through project-level installation instead of machine-by-machine setup

Yes, it is suitable because the README explicitly supports installing skills into a specified project root with --project.

  • Project-scoped installation binds the skill scope to the Android codebase, matching a shared repository workflow.
  • --agent can target particular agents; without it, behavior depends on agents detected in each environment, so teams using different agents may get different results.
  • Without --all, only the android-cli skill is installed, so the team must decide whether to install the complete set.
  • The project provides update commands, but the README does not say whether project skill files should be committed or how a team should pin skill versions.
  • Options: `--project` specifies a project root in which to install the skills
  • Install Android skills: if omitted, only the `android-cli` skill will be installed
  • Options: `--agent` accepts a comma-separated list of agents
  • To update all skills or specific ones: `android skills update --all`
android skills add --all
Not stated in the README：The README does not describe the project-level directory layout, whether skill files should be committed to Git, or how teammates should keep versions aligned.；The README does not say whether skills support version pinning or rollback.
It depends I maintain a Wear Compose Material 3 project and want the agent to follow a dedicated Wear workflow rather than rely only on basic Jetpack Compose knowledge; is this project suitable?
For: An Android developer maintaining a Wear Compose Material 3 project who wants an agent to use a dedicated Wear workflow

It depends, because the README explicitly names a wear-compose-m3 skill but does not show its coverage or compatibility details.

  • wear-compose-m3 appears directly in the update command, indicating that Wear Compose Material 3 is managed as a distinct skill.
  • The project focuses on specialized Android use cases where LLMs underperform rather than basic Jetpack Compose practices, which aligns with a specialized Wear workflow.
  • Automatic activation requires the skill to be installed and the target agent to support the relevant discovery mechanism.
  • If your task exceeds the skill’s actual APIs, versions, or device scenarios, the README provides no evidence that the agent will handle it reliably.
  • Install Android skills: `android skills update wear-compose-m3`
  • Android skills: We aren't prioritizing well-established areas where LLMs are already proficient, such as basic Jetpack Compose best practices
  • Activate skills: Your agent automatically activates skills that are relevant to your task
  • Options: If omitted, only the `android-cli` skill will be installed
android skills update wear-compose-m3
Not stated in the README：The README does not list supported Wear Compose, Material 3, or Android SDK versions for `wear-compose-m3`.；The README does not say whether the skill covers UI, navigation, testing, or the complete Wear app workflow.

✨ Highlights

  • Provides AI-optimized modular SKILL.md instructions for Android
  • Android CLI supports installation by skill, project, and agent
  • Supports Gemini, Antigravity, Claude Code, and Codex
  • Has 8,076 GitHub stars and release v1.0.15

🔧 Engineering

  • Uses SKILL.md to provide Android agents with task-specific specifications
  • android skills add supports installing individual or all Android skills
  • Uses --project for a project root and --agent to select agents
  • Claude Code and Codex can install it as an agent plugin

⚠️ Risks

  • Omitting --all installs only the android-cli skill
  • Without --agent, installation targets all detected agents
  • The README explicitly says AI can make mistakes and results need checking
  • Public contributions are not accepted; feedback goes through GitHub issues

👥 For who?

  • Developers using Gemini or Antigravity for Android tasks
  • Teams needing Android CLI to install project-level skills
  • Android developers using Claude Code or Codex plugin mechanisms
  • Users needing only basic Jetpack Compose guidance are outside the focus