# Laravel agent skills and the rules Laravel wants your agent to follow > laravel/agent-skills is the Laravel foundation's collection of agent skills, published as plugins for Claude Code and Cursor, covering code review, starter kit upgrades, Laravel Cloud deployment and Nightwatch configuration. The interesting design is in the constraints the skills impose, particularly the rule against auto-merging customised files. **laravel/agent-skills** — Laravel official collection of agent skills - Repository: https://github.com/laravel/agent-skills - Website: https://skills.laravel.cloud/ - Stars: 730 · Forks: 27 - Language: Shell - License: not declared - Published: 2026-10-21 · Updated: 2026-10-21 · Language: en - Canonical page: https://hysenlabs.com/projects/laravel-agent-skills ## Four plugins, one per product, not one bundle laravel/agent-skills is described as Laravel's official collection of agent skills, available as plugins for Claude Code and Cursor. The structure of the repository matters more than it might appear, because the installation instructions are not a single command. For Claude Code, the marketplace is added and then three plugins are installed: ```bash /plugin marketplace add laravel/agent-skills /plugin install laravel@laravel /plugin install laravel-cloud@laravel /plugin install laravel-nightwatch@laravel ``` For Cursor, the instruction is to search for Laravel in the Cursor plugin marketplace panel and install the plugins you need. The split into separate plugins is the point. The Laravel plugin is a bundle of Laravel-focused agents and skills, the Laravel Cloud plugin is a single skill for deploying to Laravel Cloud, the Laravel Nightwatch plugin is a skill for configuring Nightwatch, and the Starter Kit Upgrade skill lives inside the Laravel plugin. Four different concerns, four different installs, and a developer who only writes code and never deploys to Cloud does not need the Cloud plugin at all. The repository layout matches that split. There are separate top level directories for `laravel/`, `laravel-cloud/`, `laravel-lsp/` and `laravel-nightwatch/`, plus `.claude-plugin/` and `.cursor-plugin/` at the root, which are the manifest directories for the two host tools. The stated distribution is a hosted site at skills.laravel.cloud, so the skills are browsable as a catalogue rather than only as files in a repository. ## The review agent only looks at what you just changed The `laravel-simplifier` agent is the piece of this collection that applies to every Laravel developer, and its scope is narrower than the name suggests. The agent reviews recently modified PHP and Laravel code and refines it for clarity, consistency and maintainability without changing functionality. The listed behaviours are: applies Laravel conventions and PSR-12 standards, reduces unnecessary complexity and nesting, improves naming and readability, and focuses on recently modified code by default. That last clause is the important one. The default scope is the recent changes, not the whole application. This is a deliberate design choice and it is worth understanding why before you enable it on a large codebase. A refactoring agent pointed at an entire application is a machine for producing an enormous diff, and an enormous diff is something a reviewer cannot read. Pointed at the last few commits, it produces a small, reviewable set of changes, and the value is in the consistency it applies rather than in the volume of the edits. It also means the agent's output is only as good as the repository's own conventions are discoverable, which is why the skill is bundled with Laravel knowledge rather than being a generic style checker. The constraint that it does not change functionality is what makes the output safe to accept. A simplifier that preserved behaviour can be merged after a human read; a simplifier that quietly renamed a public method or altered a query cannot. The scope of the edits, and the promise about behaviour, are what turn this from an agent that edits into an agent that suggests. ## Starter kit upgrades are one feature per branch, and never a merge The `starter-kit-upgrade` skill exists to solve a problem specific to the Laravel ecosystem, and the constraints it lists are the most interesting part of the whole repository. The skill selectively pulls upstream improvements from a Laravel starter kit into a project bootstrapped from one. The starter kits named are `vue-starter-kit`, `react-starter-kit`, `svelte-starter-kit` and `livewire-starter-kit`, which is the list of front end stacks a Laravel application can be bootstrapped from today. What it does is pick specific features, for example toast notifications or security fixes, and not full version bumps. So the unit of work is a feature, not a version. How it does that is where the constraints live. Each feature is applied one at a time on a dedicated branch, each as its own commit. It never auto-merges customised files or silently overwrites manifests and lockfiles. And it re-runs your tests, typecheck and build to verify behaviour preservation. Read that list as a set of guardrails against a specific failure. A starter kit project accumulates local changes, and the naive way to update it is to pull the whole upstream kit and let the version control system fight it out, which produces conflict markers in files the developer deliberately customised and, worse, silently changes a lockfile and moves every dependency at once. The skill refuses both. The dedicated branch per feature and the one commit per feature are what make the result reviewable, since each change can be looked at on its own and reverted on its own. The refusal to touch manifests and lockfiles without saying so is what keeps the dependency graph stable while a feature is cherry picked in. And re-running the tests, the typecheck and the build is what turns the whole operation from a copy operation into a verified one. ## Laravel Cloud and Nightwatch, and what the skills will not do for you The remaining two plugins cover the two Laravel products that a developer touches only occasionally, and each is a single skill rather than a bundle of agents. The Laravel Cloud skill is for deploying and managing applications on Laravel Cloud using the `cloud` CLI. It guides deployment workflows, covering a first deploy, existing applications and environment setup. It manages infrastructure resources, with databases, caches, domains, buckets and secrets named as the categories. It covers build and deploy commands, queues and scheduling, and billing or usage queries. It includes operational checklists for multi-step tasks, and it follows the CRUD command patterns of the Cloud CLI. The Nightwatch skill is for configuring Laravel Nightwatch data collection, sampling rates, filtering rules and redaction policies, and it bundles the Nightwatch MCP server for browsing issues, viewing stack traces, updating statuses and adding comments. Its four capabilities are setup and configuration, sampling and filtering to manage data volume, PII redaction to protect sensitive data, and event optimisation for production workloads. The two plugins are shaped differently because the problems are different. Cloud is a sequence of steps where the order matters, hence the checklists and the CRUD patterns. Nightwatch is a configuration surface where the hard part is deciding how much to collect and what to redact, hence the emphasis on sampling rates and redaction policies. Both are worth noting for what they do not include. The Cloud skill does not create your infrastructure for you or decide your deployment strategy. The Nightwatch skill does not tell you which events to instrument; it configures the tool that carries them. And the PII redaction capability is the one to look at first if you are deploying to a service that will ship data to a third party, since that is a decision you make once and then live with. ## Installing a single skill without the whole plugin There is a third install path, and it is the one that changes the granularity of the whole thing. Individual skills can be installed directly from the repository with the `npx skills add` command, pointed at a specific directory rather than a plugin: ```sh # Laravel Cloud npx skills add https://github.com/laravel/agent-skills/tree/main/laravel-cloud/skills/deploying-to-cloud # Laravel Nightwatch npx skills add https://github.com/laravel/agent-skills/tree/main/laravel-nightwatch/skills/configure-nightwatch # Starter Kit Upgrade npx skills add https://github.com/laravel/agent-skills/tree/main/laravel/skills/starter-kit-upgrade ``` The URLs are directory paths inside the repository, which tells you the internal layout: each plugin directory contains a `skills/` directory, and each skill is a subdirectory of that. The names of the skill directories are the things you install, and they are named after the task, `deploying-to-cloud`, `configure-nightwatch`, `starter-kit-upgrade`. This is the path for a developer who wants one capability without the rest of the plugin. If your only interest is pulling a security fix out of a starter kit, you can install that single skill and nothing else, and you avoid pulling in the Cloud and Nightwatch skills you will never use. It is also the path that makes the collection composable for a team. A team that has standardised on one agent host can install what it needs from one repository, and the fact that the skills are plain directories in a git tree means they can be read, reviewed and pinned like any other source. ## How the skills are meant to be triggered The usage section is short and it shows a difference in how the four things activate. The Laravel agent is invoked after a coding session, and the example prompt is a plain request to review recent changes using the laravel agent. The Laravel Cloud skill triggers automatically when you ask about deploying or managing Laravel Cloud resources, and the examples are deploying an app to Laravel Cloud and setting up a database and cache for the staging environment. The Laravel Nightwatch skill triggers when you ask about configuring Nightwatch, with examples of configuring sampling rates for production and setting up PII redaction. The Starter Kit Upgrade skill triggers when you ask to pull upstream improvements into a project bootstrapped from a Laravel starter kit, with examples of syncing the latest toast notification feature from the Vue starter kit and pulling the 2FA autofocus fix from a starter kit upstream. The distinction between the first and the other three is the useful part. The Laravel agent is a thing you invoke, because reviewing recent changes is a deliberate act with a defined scope. The other three are things that trigger, because the skill recognises the shape of the request. That is the difference between a tool you call and a capability that activates, and it is why the prompts for the three automatic skills are phrased as requests rather than commands. It also means the naming of the skills is doing descriptive work. `starter-kit-upgrade` is named for the operation and `deploying-to-cloud` for the task, so the trigger can be inferred from the request without the user needing to know a command name. ## Where relying on skills for framework knowledge is the wrong bet There are three limits to this collection, and the first is the one that matters most for a team. A skill is a set of instructions that the agent reads. It is not a guarantee about what the agent does. The distinction matters because the most attractive use of this collection is to ask the agent to apply Laravel conventions without the developer knowing the conventions well enough to check the output. That is a reasonable use for a developer who knows Laravel and wants faster review, and a poor use for a developer who is learning the framework, because the review then becomes a black box that is either trusted or not trusted as a whole. The second limit is the scope of the trigger. The Laravel agent focuses on recently modified code by default, so a codebase with a large body of pre-existing inconsistency will come out half refactored. That is usually what you want, and occasionally it produces a diff that mixes your new work with unrelated cleanup, which is a review burden. The third is the dependency on the host tool. These are plugins for Claude Code and Cursor, with an `npx skills add` path for individual skills. The value is tied to the agent's ability to read and follow them, and the collection is written against those two hosts specifically. A team standardising on a third agent host would be depending on how well that host implements the skills format, which is not something this repository controls. The practical advice is to use the skills as a second pair of eyes on code you understand, not as a substitute for understanding it, and to keep the review agent's scope tight so the diff stays reviewable. The Starter Kit Upgrade skill is the exception to that caution, because its constraints exist precisely to keep a machine from doing damage. ## Conclusion laravel/agent-skills fits a Laravel developer who runs Claude Code or Cursor on a real application and wants the agent to follow framework conventions rather than generic advice, or who is about to pull upstream changes into a starter kit project and wants that pull to be surgical. It is the wrong tool if you want one agent that does everything, since the collection is split across four separate plugins for four separate products, and it is not a substitute for knowing Laravel yourself. Before installing, verify three things: which of the four plugins you actually need, since the Laravel, Laravel Cloud, Laravel Nightwatch and Starter Kit Upgrade skills are separate, and the review agent's default scope, because it focuses on recently modified code by default rather than the whole application. The repository has no recorded licence in the metadata and publishes no GitHub releases, and the last push was on 2026-10-17. ## FAQ ### What is laravel/agent-skills? It is Laravel's official collection of agent skills, available as plugins for Claude Code and Cursor. The repository holds four separate pieces: the Laravel plugin with the `laravel-simplifier` agent and the `starter-kit-upgrade` skill, a Laravel Cloud plugin, a Laravel Nightwatch plugin, and a `laravel-lsp/` directory. ### How do I install the Laravel agent skills in Claude Code? Add the marketplace with `/plugin marketplace add laravel/agent-skills`, then install the plugins you want with `/plugin install laravel@laravel`, `/plugin install laravel-cloud@laravel` and `/plugin install laravel-nightwatch@laravel`. In Cursor, search for Laravel in the plugin marketplace panel. ### Can I install a single Laravel skill without the whole plugin? Yes, with the `npx skills add` command pointed at a skill directory in the repository, for example `npx skills add https://github.com/laravel/agent-skills/tree/main/laravel/skills/starter-kit-upgrade`. Individual skills such as `deploying-to-cloud` and `configure-nightwatch` can be installed the same way. ### What does the laravel-simplifier agent actually review? It reviews recently modified PHP and Laravel code and refines it for clarity, consistency and maintainability without changing functionality, applying Laravel conventions and PSR-12 standards. It focuses on recently modified code by default rather than the whole application. ### How does the starter-kit-upgrade skill avoid overwriting my changes? It picks specific features rather than full version bumps, applies one feature at a time on a dedicated branch with its own commit, never auto-merges customised files or silently overwrites manifests and lockfiles, and re-runs your tests, typecheck and build to verify behaviour preservation. ## Sources - [Issues](https://github.com/laravel/agent-skills/issues) - [laravel/agent-skills on GitHub](https://github.com/laravel/agent-skills) - [Project website](https://skills.laravel.cloud/) - [README](https://github.com/laravel/agent-skills/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/laravel-agent-skills