Every AI coding assistant I’ve used writes decent, generic Laravel out of the box. Ask it for a new endpoint and you get a controller with business logic in it, a service that returns a JsonResponse, a request class with rules() instead of storeRules() — all of it syntactically fine, all of it wrong for a codebase that’s spent this entire series explaining why those specific choices matter. The assistant isn’t broken. It just doesn’t know your architecture exists, because nothing told it, and it defaults to the most common pattern on the internet, which is not the same thing as the correct pattern for your codebase.
For a while my answer was re-explaining it every session. Paste the base classes, paste the conventions, hope the context window holds long enough to finish the feature. That works exactly once per conversation and costs real tokens and real patience doing it.
The Pro tier ships a set of context files — instructions written specifically for AI coding assistants, describing every base class, every rule, every “never override this method, override that one instead” — so the assistant knows the architecture before you’ve asked it a single question. Ask for a new module and it reaches for BaseController’s response helpers instead of a raw response()->json(). Ask for validation and it writes storeRules() and updateRules() instead of a single rules() with a conditional buried inside. It isn’t smarter. It’s briefed, the same way a new engineer is briefed on day one instead of being expected to reverse-engineer the conventions from old pull requests.
## Always Extend MyNewBase
Incorrect:
class Feature extends PlainClass {}
Correct:
class Feature extends MyNewBase {}That’s the actual shape of a rule file — plain markdown, an “Incorrect” and a “Correct” example, nothing clever. The assistant reads it before it writes a line, the same way a new hire reads the conventions doc before their first pull request.
This is the part of the package that made me take agentic development seriously as something that has to be designed for, not bolted on after the fact. An assistant that writes architecture-compliant code by default is the difference between AI making your team faster at building the thing you’ve already decided to build correctly, and AI making your team faster at accumulating exactly the kind of inconsistency this whole series exists to prevent. The context files are the difference between those two outcomes, and writing them well — keeping them accurate as the architecture evolves — is exactly the kind of ongoing maintenance that belongs in a paid, actively supported tier instead of a static file you download once and watch go stale.
I built this for myself first, the same way I built everything else in this package first. I wanted an assistant that already knew the rules, so every session started from “what’s the feature” instead of “let me remind you how we do things here” for the third time this week. If you’re already paying for an AI coding assistant, this is the difference between it writing code you accept and code you have to fix.
Get it at https://packagist.org/packages/rupeshstha/core-foundation — part of the core-foundation Pro tier, and the final post in the core walkthrough. How Agentic Skills work and how to extend them for your own base classes: Agentic Skills (Laravel Boost)
What’s next is entirely about whether this is the right call for you.