README

Laravel Auditor

Laravel Auditor equips an existing AI coding agent with a specialized, evidence-based methodology and toolset for auditing Laravel applications.

Docs: mrpunyapal.github.io/laravel-auditor

This is early 0.1.x. Install it as a development dependency, wire it into Claude / Codex / Cursor / Boost (or another agent), and ask that agent to audit. It is not a one-click scanner or an autonomous product. Finding quality depends on the agent following the skill.

Requirements: PHP 8.3+, Laravel 12 or 13. Command names, config keys, MCP tool names, finding fields, and rule IDs stay compatible across 0.1.x.

Why it exists

AI agents can already read a Laravel codebase. They still tend to:

guess at framework behavior instead of collecting project facts
report style opinions as high-severity issues
invent vulnerabilities from uncommon patterns
skip verification before reporting a serious finding

Laravel Auditor gives the agent a repeatable workflow and deterministic project context so it is less likely to guess. Findings are only as good as the agent that follows the skill.

Supported agent workflow

The package is agent-agnostic. Audit knowledge lives once and is adapted thinly for:

Codex
Claude Code
Gemini CLI
other agents that consume project instructions, skills, guidelines, or MCP tools

When Laravel Boost is installed, Auditor integrates through Boost's third-party guidelines and skills. When Boost is absent, php artisan auditor:install --agents=... publishes standalone agent resources.

The first five minutes should look like this:

Install the package as a development dependency
Run boost:install or auditor:install --agents=claude_code (or your agent)
Open that AI agent
Ask it to use the laravel-audit skill
The agent produces structured findings; you can render them with auditor:report

Installation

Install the package as a development dependency:

composer require --dev mrpunyapal/laravel-auditor

With Laravel Boost

If the application already uses Laravel Boost, expose Auditor's guidelines and skills through Boost:

php artisan boost:install

After package updates:

php artisan boost:update

Do not run auditor:install just to duplicate Boost setup. Boost consumes resources/boost/guidelines and resources/boost/skills from this package directly.

Standalone (no Boost)

php artisan auditor:install --agents=claude_code

Pass the agent you actually use (opencode, claude_code, cursor, copilot, gemini, codex, junie, zed). Interactive runs ask. Non-interactive runs with no --agents, no config, and no project markers wire nothing.

The installer is idempotent and safe. It:

detects the Laravel application context
detects whether Laravel Boost is installed
publishes agent skills and guidelines to .ai/
asks which AI agent(s) the project uses (non-interactive runs resolve from --agents, then laravel-auditor.agents config, then project detection)
writes thin AGENTS.md, CLAUDE.md, GEMINI.md, Cursor, Copilot, Codex, Junie, and Zed adapters only when those files are missing
copies the laravel-audit skill into the selected agent's native skills directory
registers the laravel-auditor MCP server in the selected agent's config (except Gemini)
publishes finding/report schemas and an example findings file
publishes config/laravel-auditor.php when it is missing
reports what it created or left unchanged

Useful options:

php artisan auditor:install --dry-run
php artisan auditor:install --force
php artisan auditor:install --agents=opencode,claude_code

--agents restricts wiring to the listed agent keys (opencode, claude_code, cursor, copilot, gemini, codex, junie, zed). Non-interactive runs resolve agents from --agents, then laravel-auditor.agents config, then project detection. When none of those resolve, no agents are wired.

--force refreshes Auditor-owned resources. It does not overwrite unrelated user-owned files unless you explicitly ask it to refresh an existing adapter.

You can also publish resources with Artisan:

php artisan vendor:publish --tag="laravel-auditor"
php artisan vendor:publish --tag="laravel-auditor-config"
php artisan vendor:publish --tag="laravel-auditor-resources"

What it audits

The current 0.1.x catalog focuses on six domains. The package does not execute these checks itself — the agent does:

Domain	Looks for
Security	Authorization gaps, mass assignment, sensitive data, unsafe redirects, file handling, committed secrets, debug exposure
Performance	N+1 risks, expensive request-lifecycle work, missing indexes when query evidence exists
Architecture	Boundary violations, duplicated logic, unnecessary abstractions — without cargo-cult repository/service advice
Database	Schema/relationship mismatches, destructive migrations, nullability risks
Testing	Missing meaningful coverage, weak tests, missing authorization tests
Laravel conventions	Version-inappropriate APIs, reinvented framework features, lifecycle misuse

The package also detects ecosystem signals (Livewire, Filament, Inertia, Pest, PHPUnit, Tailwind, queues) so later rule packs can attach cleanly. A rule ships only when it can meet the evidence-first standard.

List the current rules:

php artisan auditor:rules
php artisan auditor:rules --domain=security
php artisan auditor:rules --json

0.1.x ships 61 evidence-first rules, including optional Livewire, Filament, Inertia, Sanctum, and Pest packs that only apply when those packages are installed. Queue and DSA rules always apply. The full catalog is in resources/auditor/rules/RULES.md.

php artisan auditor:rules --applicable

Example audit interaction

Ask the agent:

Use the laravel-audit skill to audit this application. Discover the project first, scope the relevant domains, and report only evidenced findings.

The agent should:

Collect project facts (auditor:status or the MCP tools)
Scope the domains that actually apply
Investigate with routes, models, schema, policies, tests, and source
Verify high-severity claims before reporting them
Produce structured findings and a report

Full example prompt

You are auditing the Laravel application in this project using the Laravel Auditor methodology.

Use the laravel-audit skill. Follow its Discover → Scope → Verify → Report workflow.

Start by calling the context MCP tools to gather deterministic facts BEFORE reading code:

project_info — PHP/Laravel versions, database, ecosystem signals

routes — the full route surface

models — all models with fillable/guarded, casts, relationships

migrations — schema changes over time

database_schema — actual tables/columns/indexes

dependencies — installed packages and versions

configuration — config keys in use

policies_authorization — gates, policies, auth middleware

jobs_events_schedules — queues, events, cron

tests — test coverage layout

Scope the relevant domains (e.g., security, database, architecture, testing). Do NOT audit everything superficially — pick the domains with the most risk signal and go deep.

For every potential finding, verify against actual files, routes, or schema. Never report a guess.

Report findings ranked P0–P3, each with: file/route/schema evidence, the rule violated, why it matters, and a concrete fix.

Be read-only. Do not modify any application code.

For a quick Discover-only pass:

Start with a Discover phase only: run all 11 context tools, summarize what this app is (framework versions, database, route surface, model list, test coverage), and flag any immediate red flags in 3-5 bullets. Do not write findings yet.

For a data-structure / ownership pass:

Use the laravel-audit-dsa skill. Inventory subsystems, review them in bounded read-only lanes, then rank P0–P3.

Example finding

{
  "id": "F-2026-0001",
  "rule_id": "AUD-SEC-001",
  "title": "Missing authorization boundary",
  "domain": "security",
  "severity": "high",
  "confidence": "confirmed",
  "status": "open",
  "summary": "Any authenticated user can delete another user's post.",
  "why_it_matters": "The destroy action never authorizes the Post policy.",
  "evidence": [
    {
      "type": "file",
      "reference": "app/Http/Controllers/PostController.php",
      "line": 42,
      "end_line": 48
    }
  ],
  "affected_resources": ["app/Http/Controllers/PostController.php"],
  "symbol": "App\\Http\\Controllers\\PostController@destroy",
  "recommendation": "Authorize the deletion with a PostPolicy or route middleware."
}

Severity: critical, high, medium, low, info.

Confidence: confirmed, high, medium, low.

DSA / subsystem audit

For a read-only, orchestrated pass over data structures, state, algorithms, and ownership, use the laravel-audit-dsa skill. The coordinator inventories every subsystem, sends bounded read-only workers (at most two findings each), then validates, dedupes, and ranks P0–P3.

php artisan auditor:context subsystems
php artisan auditor:report --findings=storage/auditor-findings.json

Reports include a priority synthesis. Set metadata.priority to p0–p3 when ranking explicitly.

Collecting project facts

You do not need MCP to inspect the app. Dump any collector from Artisan:

php artisan auditor:context --list
php artisan auditor:context project_info
php artisan auditor:context routes --output=storage/auditor-routes.json

Or from PHP:

use LaravelAuditor\Facades\LaravelAuditor;

LaravelAuditor::collect('models');

MCP tools

php artisan auditor:mcp

Example Claude Code registration:

claude mcp add -s local -t stdio laravel-auditor php artisan auditor:mcp -q

When Laravel Boost is installed, the same context collectors are also registered automatically as read-only tools inside Boost's laravel-boost MCP server (via boost.mcp.tools.include), so no extra setup is needed there.

Tools:

Tool	Returns
`project_info`	PHP/Laravel versions, database engine, ecosystem signals, source layout
`routes`	Methods, URIs, names, actions, middleware
`models`	Tables, fillable/guarded, casts, relationships
`migrations`	Migration files
`database_schema`	Tables, columns, indexes (read-only)
`dependencies`	Direct Composer requirements and versions
`configuration`	Config keys and a small set of non-secret values
`policies_authorization`	Gates, policies, auth middleware
`jobs_events_schedules`	Jobs, events/listeners, schedules
`tests`	Framework, test case counts (feature/unit), file layout
`subsystems`	Ownership-bounded inventory for a DSA-style coordinator audit

These tools are read-only. They return structured facts, not unfiltered source dumps.

Reporting and diagnostics

php artisan auditor:status
php artisan auditor:report
php artisan auditor:report --example
php artisan auditor:report --format=json
php artisan auditor:report --format=text
php artisan auditor:report --findings=storage/auditor-findings.json --output=storage/auditor-report.md
php artisan auditor:report --findings=storage/auditor-findings.json --format=sarif
php artisan auditor:ci --findings=storage/auditor-findings.json --fail-on=high

Finding and report JSON schemas live in resources/auditor/schema. See the findings docs.

auditor:report does not invent findings. The agent produces findings; the command renders them as Markdown, JSON, or CLI text with project facts, domain scope, counts, key risks, evidence, and recommendations.

There is no web dashboard. Reports are CLI, Markdown, JSON, or SARIF.

Architecture

src/
  Audit/          rules, findings, evidence, domains, report renderers
  Context/        read-only Laravel collectors used by MCP and reports
  Console/        install, status, rules, report, mcp
  MCP/            stdio MCP server
  Support/        Boost detection
resources/
  auditor/        agent-agnostic skills, guidelines, and rules
  boost/          third-party Boost guidelines and skills

Audit knowledge stays agent-neutral. AGENTS.md / CLAUDE.md adapters only point at that source of truth.

Configuration

Publish the config file to change the default domain list, extra rule directories, or standalone resource target:

return [
    'domains' => [
        'security',
        'performance',
        'architecture',
        'database',
        'testing',
        'conventions',
    ],
    'rules' => [
        // base_path('auditor/rules'),
    ],
    'resources_target' => '.ai',
    'agents' => [],
    'context' => [
        'composer_audit' => false,
        'test_listing' => false,
    ],
    'report' => [
        'format' => 'markdown',
    ],
];

Trustworthiness

The agent is instructed to:

say when evidence is incomplete
distinguish confirmed findings from hypotheses
avoid inventing package or runtime behavior
avoid claiming exploitability without evidence
avoid recommending upgrades only because a package is old
keep style preferences out of high-severity findings

A short, evidenced report is the intended product.

Compatibility

Laravel Auditor is meant to sit in an application as a development tool for a long time. Releases stay backward compatible: command names, config keys, MCP tool names, finding fields, and rule IDs are not removed or renamed. Breaking changes are rare and would require a major version with an explicit changelog note.

Not in 0.1.x

Deferred work lives in the future scope docs. That includes automatic fixes, historical baselines, a web dashboard, legacy/standalone runners, organization policy packs, and deeper ecosystem rule packs.

Changelog

Please see CHANGELOG for more information on what has changed recently.

Contributing

Thank you for considering contributing to Laravel Auditor! Please review our contributing guide to get started.

mrpunyapal / laravel-auditor

Maintainers

Package info

Fund package maintenance!

Statistics

Security