Google's Android team explains the philosophy behind Android Skills — context snippets injected into AI coding agents to fill knowledge gaps in state-of-the-art models. Skills are only created when models demonstrably lack knowledge of fast-moving APIs (e.g., AGP 9, Navigation 3, Camera APIs, Perfetto SQL). Each skill injects 100–200 tokens into every task context, so hoarding basic skills wastes tokens. Skills are validated against internal evals before release, and the Android Knowledge Base (accessible via Android CLI) should be the first resort before installing any skill. Community-built skills from trusted contributors are highlighted, and users are warned against AI-generated skill repositories. Skills are designed to become obsolete as models improve, with evals run against new model releases to determine retirement timing.
Table of contents
Why are there so few official skills?Evaluating skillsUsing the Android Knowledge Base (Android Studio or Android CLI)Why are pull requests disabled?When do core or basic skills make sense?Where can I find core skills?Goal: deprecationQuestions this post answers
How many tokens does an Android Skill inject into my AI agent context?
Each installed Android Skill injects 100–200 tokens into the baseline context of every task you start. If the skill actually activates during a task, that count can quickly jump into the thousands. This means installing many basic skills is both counterproductive and expensive — only install skills for topics your model genuinely lacks knowledge on. Android developers managing agent token budgets track skill updates and deprecations on daily.dev.
Why doesn't the Android team accept pull requests for new official Android Skills?
The Android Skills evaluation framework depends on internal infrastructure that cannot be open-sourced. Without that infrastructure, there would be no way to re-evaluate incoming PR changes. Instead, the team actively monitors community feedback — bugs, optimization suggestions, and new skill requests should be filed as GitHub issues at github.com/android/skills. Developers contributing to or following the Android Skills project can track community discussions on daily.dev.
When does the Android team retire an Android Skill?
Skills are retired when state-of-the-art models no longer need them. The team runs the skill's evals whenever new models are released — if the evals pass without the skill active, the skill is kept around for a few more months to allow users to transition, then deprecated. The expectation is that skills built around new APIs will become obsolete as models improve. Developers building on evolving Android APIs can follow model and skill deprecation news on daily.dev.