App templates used to be code. In the era of AI code generation, app templates can now be prompts — very detailed prompts, very structured prompts — agent skills.
Skills provide procedural knowledge. This is what makes them super useful in this era of AI code generation. Before now, if you wanted your project to be set up in a particular manner, you created a template project with actual code from which you could always start new projects. Now, you only need to describe the procedures and steps that AI agents need to follow when setting up your projects.
Why Skills Are the New Templates
Another useful property of skills is that they are task-specific and only load when relevant, making them better for specialized workflows. This means you no longer need to provide all the instructions your agents need before firing them off — with skills, they automatically load the appropriate instructions per task.
It also doesn't hurt that agent skills can contain code snippets alongside procedural instructions, giving agents both the 'what' and the 'how' for any given task.
Introducing Relai
To put this idea of skills as the new school templates to test, we're building a simple platform that allows users to select templates from which they can build out their applications. We call the platform Relai (Reliable AI), pronounced like the word 'relay.'
On Relai, you enter a name and a short description of what you want to build. Afterwards, you are presented with different app templates to choose from — API (Express), API (Nest), Web UI (Next), Web UI (Vite), Mobile UI (React Native), and CLI. After choosing, you can click Download to get the project on your local machine.
Alternatively, you can use the CLI tool: npm i -g relai, then relai init with your project name, description, and platform, followed by relai apps create to add specific app types with templates.
Project Structure
The generated project follows a monorepo structure. The apps folder holds all apps added to the project, giving AI agents a holistic view — a project can contain multiple apps such as an API, a web UI, a mobile UI, and a marketing website.
Agent skills are located in the .claude/skills/ path because the current platform upon which Relai is running is Claude Code. The skills folder contains a host of agent skills related to the chosen framework — skills like implement-authentication, generate-crud-endpoints, optimize-db-schema, and add-stripe-subscriptions.
The idea is to have the user choose higher-level options like type of app, features, and themes, and to have those choices translate to agent skills that guide the AI in building out the application.
From Choices to Skills to Code
Each skill encapsulates a complete unit of work — a SKILL.md file with instructions and metadata, optional scripts for executable code, references for documentation, and assets for templates and resources.
The platform also maintains a .relai/ directory with project metadata, architecture information, conversation logs between user and agent, and AI-generated tasks. This gives the system full context about the project's state and history.
With parallel workers, you can run multiple agents simultaneously: relai run --workers=4 --model=claude-opus-4-6 --mode=plan, letting the AI plan and execute across skills concurrently.
Conclusion
By reimagining templates as agent skills rather than boilerplate code, we open the door to a fundamentally more flexible and intelligent way to scaffold applications. Skills are composable, evolvable, and framework-aware — everything that static templates are not.
In the next article, we tackle the task of generating high-fidelity code — which will be tested by generating the code multiple times to see how similar the outputs are as we build out Relai.