AI-ready development guide

Laravolt v7 is AI-ready because its conventions make work easier to describe, review, and test.

The goal is not to let a coding agent improvise an application. The goal is to give humans and agents a shared map: where code belongs, which APIs are safe to extend, and what evidence proves the change works.

Start with a clear feature contract

Before asking an agent to code, write the feature in Laravolt terms:

Resource: PurchaseOrder
Screens: index, create, edit, show
Actions: submit, approve, reject
Permissions: purchase-order.read/create/update/approve/reject
Validation: StorePurchaseOrderRequest, UpdatePurchaseOrderRequest
UI: PrelineForm for forms, Suitable for listing
Tests: feature tests for validation and permissions, browser smoke for happy path

This turns a vague prompt into a reviewable checklist.

Preferred project layout

Keep custom application code in standard Laravel locations:

app/
  Actions/
  Http/
    Controllers/
    Requests/
  Jobs/
  Models/
  Policies/
  Providers/
resources/views/
routes/web.php
tests/Feature/
tests/Browser/

Use Laravolt packages for platform surfaces: forms, tables, menus, access control, workflow, media, and generated admin modules. Do not bury domain logic inside Blade views or anonymous callbacks.

Promptable conventions

Good Laravolt prompts mention the exact platform primitive to use:

• “Use PrelineForm and a FormRequest.”
• “Use Suitable for the list page and keep query logic out of Blade.”
• “Register the sidebar menu with permission metadata.”
• “Add positive and negative permission tests.”
• “Dispatch slow side effects to a queued job.”
• “Mark unknown APIs as TODO instead of inventing them.”

Bad prompts ask for broad output without boundaries:

Build an inventory module with AI.

Better:

Build a Laravolt v7 inventory module for Product.
Create controller, FormRequests, policy, menu entry, Suitable list, PrelineForm create/edit views,
and tests for product.read/create/update permissions.
Do not add destructive actions without policy checks.

Safe extension points

Prefer these extension points when automating work:

If an agent needs to change something outside this table, ask it to explain why before accepting the patch.

Review checklist for AI-generated code

Review generated Laravolt code with this checklist:

• [ ] Routes are named and protected by auth/authorization.
• [ ] Controllers are thin and delegate business work.
• [ ] Forms use FormRequest validation.
• [ ] Blade views do not contain query or permission logic.
• [ ] Menu entries include permission metadata.
• [ ] Destructive buttons have backend authorization, not only UI hiding.
• [ ] Tests cover allowed and denied paths.
• [ ] Docs or comments explain any non-obvious convention.

Minimal evidence contract

Every agent task should return evidence, not just a summary:

Return changed files, tests run, build result, screenshots if UI changed,
and any TODOs that require human verification.

For release work, require at least:

composer test
bun run build
php artisan route:list

Use smaller targeted gates during development, then full gates before tagging a release.

What to read next

• AI-ready platform — the product-level mental model.
• Forms overview — predictable form APIs for humans and agents.
• Browser testing — smoke tests that catch integration drift.