Agent skills are the missing layer between generic AI and your codebase. Here's how our Android team built 13 of them and what we learned.

Medium Engineering is a publication on Medium.com that provides insights, best practices, and technical articles from Medium's engineering team. Covering topics such as software architecture, backend development, and infrastructure management, Medium Engineering offers  insights into the engineering practices and technologies used at Medium. Developers can learn about Medium's engineering culture, technology stack, and software development processes, and gain insights from Medium's experiences in building and scaling a popular content platform.

Medium Engineering

Medium's small Android team shares how they use Cursor's AGENTS.md and custom 'skills' (Markdown runbooks stored in .agents/skills/) to make AI coding assistants generate code that matches their team's conventions. They built 13 skills across four categories: scaffolding (new screens, modules, use cases), migration (Material 2→3, ViewModel patterns), pattern enforcement (feature flags, analytics, referrer tracking), and workflow automation (deeplinks, translations, unit tests). Key lessons: skills are living documents that improve iteratively, hyper-specificity beats generality, and consistency across a small team is the primary win. They also describe how agents can write new skills by analyzing existing PRs, and outline future plans for scoped rules and automated release diff commands.

Making AI Write Android Code Our Way: A Practical Guide to Agent Skills

Skills: Step-by-Step Playbooks for the AI

Get Pierrick CAEN ’s stories in your inbox