AI Agent Steering & Readiness
Learn how Kolt is built from the ground up to be "AI-Agent-Ready" and how you can use its context templates to steer AI coding assistants.
Overview
Modern developers frequently pair-program with agentic AI models (such as Google Gemini, Anthropic Claude, Cursor, and GitHub Copilot). However, large KMP codebases with complex Gradle build logic often confuse AI models, leading to compilation issues, build cache pollution, and incorrect platform imports.
Kolt solves this by design. Every project bootstrapped with Kolt includes standardized AI context files. These templates instantly align an AI agent's reasoning with the codebase structure, coding guidelines, build commands, and platform constraints.
Core Principles of Kolt's AI-Readiness
🔌 Classloader Isolation
Convention Gradle plugins and Vanniktech publishing are strictly separated. AI agents can easily reason about dependencies without triggering configuration-time classpath collisions.
📑 Declarative Versioning
No duplicated version numbers. All versions are read dynamically from TOML catalogs and generated into static class Constants, preventing agents from adding invalid dependencies.
🧠 Context Auto-Discovery
Standardized roots like AGENTS.md and CLAUDE.md are auto-discovered by LLM agents, ensuring they respect naming conventions, package layout, and testing protocols without manual input.
📦 Scaffolded Steering
A built-in Gradle task automatically scaffolds steering markdown files into new consumer modules, saving developers from manually copying architecture rules.
Downloadable Steering Templates
Download these standard steering files and place them in your repository root or project config directories to instruct your AI assistants:
CLAUDE.md (Developer Quickstart)
Defines primary build commands, test execution shortcuts, and syntax styling guidelines.
AGENTS.md (Architecture & Coding Guide)
Autorun architectural rules, catalog definitions, and monorepo modular structures.
KOLT.md (Full Monorepo Guidelines)
Detailed guide describing convention plugin registry, extensions DSL, and platform variants.
CODING_STANDARDS.md (Kotlin & Flow System)
Rules for expect/actual structures, coroutine context handlers, flows, and type safety.
ARCHITECTURE.md (Data Flow & State System)
Detailed guidelines on offline synchronization, repository patterns, and MVI structures.
TESTING.md (Unit & Coroutine Validation)
Guidelines for mocking dependencies, using coroutine test scopes, and Turbine flow testing.
How to Scaffold Steering in Your Project
If you have already applied the Kolt Gradle plugins to your project, you can generate all the above steering templates directly in your repository root by running the built-in scaffolding Gradle task:
./gradlew scaffoldKoltResourcesThis generates a localized version of the files in your project, populated with your custom version variables and settings. AI agents will automatically pick these up during coding sessions.