https://developer.android.com/agents/skills/device-ai/appfunctions/skill

To expose Android app workflows as AppFunctions, you map user tasks like creating notes or sending messages into Kotlin code and metadata configurations, allowing a system agent to discover and execute them on-device without relying on the app UI.

AppFunctions are on-device capabilities that allow system agents to execute app workflows hands-free. You need them when you want an AI assistant to trigger in-app tasks like playing media or creating notes directly, bypassing manual UI interactions.

You test AppFunctions locally using ADB by running commands to list registered functions, execute them, and verify their enable or disable state, ensuring your on-device agent tooling behaves correctly on supported Android devices.

Yes, implementing AppFunctions requires targetSdk 36+ and compileSdk 37+ to support the Jetpack generated Kotlin code and safety constraints needed for system agents to properly discover and invoke your on-device workflows.

You refine KDoc for AppFunctions by explicitly documenting parameters, constraints, and dependencies, which ensures the AI agent accurately interprets the function requirements and safely executes the on-device Kotlin implementation.

The best way to discover high-value workflows is to analyze your Android codebase to identify and recommend user tasks that significantly improve hands-free, AI-driven task execution, then map those specific workflows into Kotlin AppFunction implementations.