With an AI agent

BuilderKit ships an agent skill that teaches your coding agent how to work with it: how to start a new application, how to install BuilderKit into an existing one, how to update it, and how to find and adapt blocks. You install the skill once, then describe what you want to build.

01. Check your GitHub access

The skill reads BuilderKit straight from the private repository, so your agent has to run as the same GitHub account that has access to it. Sign in with gh auth login or make sure your SSH key is set up, and confirm that cloning the repository works from your terminal.

02. Install the skills

Run this in your project folder to install the BuilderKit skill. It works with Claude Code and any other CLI agent that supports skills:

npx skills add builderkit/builderkit

Then install the official Angular skills. They give your agent current Angular guidance to pair with the BuilderKit one:

npx skills add angular/skills

03. Tell the agent what to build

With the skills in place, describe what you need in plain language:

• "Create a new Angular app with BuilderKit for a project management tool."
• "Add a settings page with account and notification sections using BuilderKit blocks."

Before the first task, the agent brings BuilderKit into your project by cloning the repository content into a .builderkit folder, keeping your own repository untouched. That folder is ignored by git and disposable, much like node_modules. If it goes missing or falls behind, the agent clones it again. Everything after that is a local read: new applications are started from the Spark template, existing ones get BuilderKit installed into them, and pages are composed from the block index at .builderkit/blocks/INDEX.md.

In sandboxed environments an agent may not be able to reach the network on its own. If that happens, it will ask you to approve the clone or to run it yourself. It is a single step, once per project.