:yay: Hello, On a non-trivial Compose Multiplatfo...

# feed

[Ali Sadeghi]

yay Hello,

On a non-trivial Compose Multiplatform codebase, AI assistance tends to degrade the architecture over time. Context is lost between prompts, so each step is unaware of the decisions made before it, and the structure gradually diverges from its original design.

🐕*KMPilot* is a template I built to keep that under control. It provides a set of skills for the distinct stages of feature work, design, build, verify, test, and review, so each stage operates under its own constraints rather than a single open-ended prompt. A feature specification is persisted to disk and survives a context reset, and a pre-tool hook rejects any edit that attempts to bypass the pipeline. The outcome is a complete build that remains architecturally consistent across Android and iOS from one codebase.

I have documented it against a real build: one article on the design rationale, and one full walkthrough with every prompt and timing recorded.

I would value your feedback on the approach, and on how you are addressing this problem in your own work.

Repo: github.com/ThisIsSadeghi/KMPilot
Why it exists: proandroiddev.com/design-a-screen-get-a-clean-architecture-feature-and-keep-ai-generated-kmp-code-from-drifting…?…
Full walkthrough: proandroiddev.com/seven-claude-code-commands-one-kotlin-multiplatform-app-on-android-and-ios…?…