Laravel Boost

Laravel Boost gives AI coding assistants accurate, project-specific context. This package ships its own guideline and skill, so an assistant working in your codebase writes v8 code instead of guessing — or worse, reproducing v7 patterns it learned from older tutorials.

What ships

File	What it does
resources/boost/guidelines/core.blade.php	The package guideline Boost injects into the assistant's context
resources/boost/skills/sweet-alert-development/SKILL.md	A skill for building alert flows — toasts, inputs, guarded actions

You do not install them separately. Boost discovers guidelines and skills from the packages already in your composer.json:

php artisan boost:install

Why it matters here

Three v8 changes fail quietly rather than loudly, which is exactly the kind of thing an assistant gets wrong:

• html() takes one argument now. A v7-style html($title, $code, $icon) call does not error, because PHP ignores surplus arguments — it simply renders nothing.
• view() takes the view name first; v7 took the title first.
• Alert::input() and Alert::make() never flash themselves, so a chain that forgets flash() shows no alert at all.

The guideline calls all three out, along with the guarded-action attributes and submitTo(), so generated code lands on the v8 API.

Upgrading an existing project

If the assistant is working on a v7 codebase, point it at the migration command rather than letting it rewrite call sites by hand:

php artisan alert:upgrade --dry-run

See the Upgrade Guide for what it covers.

A note on editing the guideline

core.blade.php is compiled by Blade before it reaches the assistant, so anything Blade recognises inside an example has to be escaped — written as @, and directives as @@sweetAlert, @@csrf and so on. An unescaped directive is executed, which is how a guideline can end up shipping a rendered CSRF token instead of the code you meant to show.

The package's test suite renders the guideline the way Boost does and asserts the output is prose, so this cannot regress unnoticed.