How to use the Android Clean Architecture agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions the agent can read when a task matches them. This skill is a single SKILL.md file. Its source describes the module layout, the dependency rule that the domain layer must not depend on data, presentation or any framework, and code patterns for use cases, repositories, mappers, Room, SQLDelight, Ktor, Koin, Hilt, Result-based error handling and Gradle convention plugins.

When installed, the agent follows these patterns when you ask it to build or refactor app code. It is a reference for consistency, not a program that runs on its own.

• Recommended modules: app, core, domain, data, presentation, design-system and an optional feature folder for large projects
• Use cases with operator fun invoke, plain Kotlin domain models and repository interfaces defined in domain
• Data layer patterns for Room (Android), SQLDelight and Ktor (KMP)
• Dependency injection with Koin (KMP friendly) or Hilt (Android only)
• An anti-patterns list, such as no business logic in ViewModels and no GlobalScope

How do you install the android-clean-architecture skill?

Download the ZIP for this skill from this page and unzip it. The skill folder contains SKILL.md. Put that folder in one of these locations:

For a personal install available across projects, place it in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder so it applies only there.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/android-clean-architecture/
• Project only: .claude/skills/android-clean-architecture/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with natural requests?

You do not need a special command. Describe the Android or KMP task in plain language and mention the architecture work you want. The source lists these situations as triggers: structuring Android or KMP modules, implementing use cases, repositories or data sources, designing data flow between layers, setting up Koin or Hilt, and using Room, SQLDelight or Ktor in a layered architecture.

Naming the layers or patterns in your request makes it more likely the agent applies the playbook.

• Set up a new Android app with domain, data and presentation modules following clean architecture.
• Add a GetOrdersByCustomerUseCase and its repository interface in domain, with the implementation in data.
• Refactor this ViewModel so the business logic moves into use cases.
• Set up Koin modules for domain, data and presentation in our KMP project.
• Add a Room entity and mapper so the data layer converts to domain models.

What can you expect the agent to produce?

For a feature request, the agent should place domain models and repository interfaces in the domain module, write use cases that wrap single business operations, implement the repository in data with local and remote data sources, and map entities or DTOs to domain models before they reach the UI. For error handling it can use Result<T> or a sealed type such as the Try and AppError pattern shown in the source.

Check the output against your project's conventions. The source notes the agent may push these patterns over existing ones.

What does the skill need, and is it safe to use?

The skill needs Kotlin and an Android or Kotlin Multiplatform project. It does not run code itself. According to its listing, it runs no scripts and makes no network calls. Its only file is SKILL.md.

The skill only adds instructions to the agent's context. Any code the agent writes still goes through your normal review, build and test process. The source includes example URLs such as https://api.example.com/ as placeholders, so replace them with your own endpoints.

• Requires: Kotlin plus an Android or KMP project
• Runs scripts: no
• Uses network: no
• Files: SKILL.md

Tips for getting good results

Give the agent your existing module names and layers so it fits them rather than inventing a new structure. If your project uses another architecture, such as MVI with its own conventions, say so up front, because this skill is written around clean architecture layers.

Ask for one feature at a time. Small, specific requests make it easier to review the generated domain, data and presentation changes.

• Tell the agent which modules already exist before asking for new ones
• Name the DI framework you use (Koin or Hilt)
• Review dependency directions after each change: domain must not import Android or data classes