Google Expands Android Developer Tooling with CLI Device Streaming and Grounded AI Skills

As the landscape of mobile app development continues to evolve rapidly with the integration of artificial intelligence, Google is introducing a series of major updates aimed at empowering Android developers regardless of their preferred tools, models, or command-line interfaces. The overarching goal of these enhancements is to streamline the creation of high-quality, polished Android applications by bridging the gap between cutting-edge AI capabilities and official platform standards. Central to this latest release are significant updates to the Android CLI command-line tooling—most notably the inclusion of remote Android Device Streaming—alongside an expansion of the official Android skills repository and a deep dive into specialized frameworks like Wear OS Compose Material 3.

Android CLI for Command-Line Development

Modern development workflows increasingly rely on automation, command-line interfaces, and AI-assisted agents to accelerate production cycles. To support this shift, the Android CLI has been designed as a robust foundation to make command-line-driven Android development smoother and more efficient. The tooling is engineered to be agnostic, meaning it can seamlessly support virtually any AI agent, LLM, or third-party tool chosen by a developer to build for the Android ecosystem.

Working in tandem with the designated android-cli skill, AI agents can leverage the Android CLI to handle a wide array of heavy lifting. This includes creating, building, testing, and managing complex Android projects from scratch. Furthermore, the CLI assists developers in setting up pristine development environments, spinning up and executing emulators, and running comprehensive test suites directly from the terminal. By centralizing these operations into a cohesive command-line utility, Google aims to reduce friction and allow developers to maintain momentum without constantly context-switching between graphical user interfaces and terminal windows.

Android Device Streaming Now Available in CLI

One of the persistent challenges in mobile application development is ensuring that software performs reliably across the vast and fragmented spectrum of physical hardware and operating system variations. While emulators provide a solid baseline, rigorous testing on real devices remains vital for catching hardware-specific glitches, performance bottlenecks, and OS-level quirks. However, developers do not always have immediate physical access to every device model required for comprehensive QA.

To solve this problem, Android Device Streaming is now fully integrated into the Android CLI, bringing remote physical hardware access directly into the terminal environment. Through a secure ADB over SSL connection, AI agents can interact with real physical devices as though they were plugged directly into the developer’s workstation via a physical USB cable. This capability completely transforms terminal-based workflows, allowing agents to spin up remote devices, deploy fresh builds, harvest logs and traces, and even capture headless screenshots—all without leaving the command line.

Getting started with the feature involves linking the project, instructing the agent to query available remote devices, and selecting the target hardware for testing. This remote capability ensures that AI-driven workflows can validate application behavior on actual consumer hardware early and often, catching edge cases that might otherwise slip past standard emulator-based testing.

Grounding Agents with Android Skills

While large language models possess vast amounts of general programming knowledge, they frequently struggle with platform-specific standards, rapidly changing APIs, and the nuances of modern framework updates. To address this limitation and prevent hallucinations or outdated code generation, Google continues to expand its official Android skills repository.

Android skills are essentially structured sets of instructions encapsulated within SKILL.md files designed to ground AI agents using official, up-to-date guidance straight from developer.android.com. Rather than depending on a model’s static training cutoff date, these skills supply precise, fresh documentation, API references, code samples, and architectural patterns directly into the agent’s working context.

With more than twenty distinct skills currently available in the repository, developers can equip their AI agents to tackle intricate, specialized development tasks with confidence. The engineering philosophy behind these skills emphasizes rigorous evaluation, ensuring that every piece of guidance provided to an AI model is thoroughly tested against real-world scenarios. Managing these skills across different projects and individual agent directories is handled smoothly through the Android CLI. Moreover, because the skills are designed to be entirely environment-agnostic, they function seamlessly across a wide variety of setups—whether a developer is writing code inside Android Studio, utilizing Antigravity, or pairing with third-party coding agents like Claude and Codex.

Skill Spotlight: Wear Compose Material 3

Developing applications for Wear OS presents unique design and engineering challenges that differ substantially from traditional smartphone development. Smartwatches require careful consideration of round viewports, rotary input mechanisms, ambient display modes, and aggressive power conservation strategies. Developers must also adopt specific architectural patterns, such as prioritizing the TransformingLazyColumn component and organizing layouts within AppScaffold and ScreenScaffolds containers.

Without explicit, specialized guidance, standard LLMs often lack the contextual understanding required to implement these distinct Wear OS patterns correctly. To bridge this knowledge gap, Google introduced the Wear Compose Material 3 skill (wear/wear-compose-m3), designed to give AI agents deep insight into modern watch face and app development.

Early adopters across the industry have already reported dramatic productivity gains after integrating the Wear Compose Material 3 skill into their workflows. For instance, the engineering team at FotMob utilized the skill to modernize their existing Wear M3-based application. Their migration efforts included updating multiple lists to use TransformingLazyColumn with proper ScreenScaffold content padding, implementing ListHeader titles, applying SurfaceTransformation to cards and buttons, adopting theme typography, and configuring accurate Wear previews.

The resulting code compiled successfully on the first pass and was thoroughly verified on the emulator for smooth scrolling, rotary input behavior, edge morphing, and right-to-left language support. Crucially, the migration allowed the team to completely eliminate their legacy wrapper code along with accumulated rotary and focus boilerplate. According to the developers, the skill successfully caught subtle mistakes that the underlying AI model would have otherwise missed—such as forgetting to forward the ScreenScaffold content padding into the active list, and mistakenly using hardcoded sp values instead of proper theme typography.

Reflecting on the experience, Roy Solberg, Android Tech Lead at FotMob, noted that a single skill deployed over the course of a single afternoon enabled the migration of eight distinct lists while wiping out an entire pile of custom rotary code. As Google continues to refine both its CLI tooling and its growing repository of grounded AI skills, the integration of intelligent automation into everyday mobile development promises to make building high-quality Android and Wear OS applications faster and more reliable than ever before.