Livewire Security Audit

A Claude Code skill that audits Laravel Livewire apps (v3 and v4) for security problems, then optionally fixes them. It exists because Livewire's convenience hides a large client-facing attack surface: every public property round-trips through the browser, every public method is callable from devtools whether or not your UI references it, and anything scalar you put on a public property is readable in page source.

The skill runs a full audit of every component in your app — no sampling — and produces a severity-ranked report where every finding carries quoted evidence, a concrete exploit scenario, and a specific fix. Every component form is inventoried: class-based, v4 single-file and multi-file, and Volt (functional API, class API, and inline @volt fragments). Findings are verified twice before they reach the report: audit subagents flag candidates, then the main session re-reads every flagged line and discards anything that does not hold up. A clean report means your components were checked and came back clean, not that the tool ran out of patience.

What it checks

27 checks across four surfaces. Every engine-behavior claim behind them was verified against the Livewire v4.3.0 source and the official 3.x/4.x docs, and is cited in the reference files.

Client tampering

• LW-01 every public action authorizes; parameters treated as untrusted input
• LW-02 #[Locked] on identity/price/ownership scalars the client must not write
• LW-03 properties that never needed to be public (use #[Computed] or protected)
• LW-04 unlocked properties whose writes drive queries, payments, or permissions
• LW-05 #[On] listeners and the $listeners array — dispatchable from the browser with attacker-controlled payloads
• LW-06 #[Url] properties validated before use
• LW-07 public helper methods that should be protected
• LW-24 lifecycle hooks (updated*/hydrate*/mount) that act on raw client input — the blind spot LW-01 skips

Data exposure

• LW-08 scalars/arrays/plain Collections serialize fully into wire:snapshot in page source
• LW-09 model attributes copied into scalars ($hidden will not save you)
• LW-10 #[Computed(cache: true)] on user-scoped data — one value shared across all users
• LW-11 #[Computed(persist: true)] holding sensitive data with long TTLs
• LW-12 model class names leaking in snapshots (morphMap)
• LW-13 secrets/PII echoed back in validation errors and flash messages
• LW-25 $this->js() expressions that interpolate user data — XSS the snapshot audit can't see
• LW-26 sensitive data shipped to the browser through dispatched events (not in the snapshot)

Standard web surface, Livewire-flavored

• LW-14 file upload validation (per-property rule AND the global temp-endpoint rule)
• LW-15 mass assignment from $this->all() or a whole array property (data.is_admin deep writes)
• LW-16 properties interpolated into raw SQL
• LW-17 {!! !!} on user-controlled data
• LW-18 rate limiting on expensive/abusable actions (core Livewire ships none)
• LW-19 validation that actually constrains (exists: is not ownership)
• LW-27 open redirect from a client-controlled #[Url]/property target

Configuration

• LW-20 custom middleware not registered persistent (route middleware does NOT all re-run on component updates — only Livewire's default persistent list does)
• LW-21 update route exposure
• LW-22 full-page component routes without auth middleware
• LW-23 native type declarations (untyped properties accept any JSON shape the client sends)

Install

In a Claude Code session, add the marketplace and install the plugin:

/plugin marketplace add timgavin/claude-plugins
/plugin install livewire-audit@timgavin

After it installs, run /reload-plugins. The skill is then invoked as /livewire-audit:livewire-audit — Claude Code namespaces plugin skills by plugin name — or just ask in plain language ("audit my Livewire app").

To get new releases automatically, open /plugin, select the timgavin marketplace, and turn on auto-update. Third-party marketplaces have auto-update off by default, so this is a one-time toggle on your end.

Manual install (no plugin system)

To drop the skill in by hand, copy the one folder into your skills directory:

tmp=$(mktemp -d)
git clone --depth 1 https://github.com/timgavin/livewire-audit.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/livewire-audit" ~/.claude/skills/livewire-audit
rm -rf "$tmp"

Or per-project: copy it to .claude/skills/livewire-audit inside the project instead.

Use

From a Claude Code session in your Laravel project, run the audit. The command depends on how you installed it:

• Plugin install: /livewire-audit:livewire-audit
• Manual install: /livewire-audit
• Either way, plain language works too: "audit my Livewire app"

The read-only audit writes a report to livewire-audit-<date>.md in the project root. Append --fix to apply fixes for confirmed findings instead of only reporting them — e.g. /livewire-audit:livewire-audit --fix.

The audit makes no changes to your code and runs no migrations or seeders. If your app happens to be runnable, it will additionally pull a real rendered page and confirm snapshot-exposure findings against the actual wire:snapshot payload; if not, those findings are labeled statically inferred.

Other agents

This is built and tested as a Claude Code skill. It uses the open Agent Skills format (SKILL.md + frontmatter), so other runtimes that support that format — Copilot CLI, Codex, Gemini CLI — can load it too; install it in that tool's skills directory instead of ~/.claude/skills and invoke it by name rather than the /livewire-audit slash command.

One caveat, stated plainly: the audit's accuracy comes from fanning the work out across parallel read-only subagents and then re-verifying every finding in the orchestrating context. That orchestration is written for Claude Code's subagent model. An agent without parallel subagent dispatch can still run the checks, but in a single context and without that verification stage — a degraded run. The checklist, snapshot internals, and fix patterns themselves are pure Livewire/PHP knowledge and carry over unchanged. I have not tested the skill on Codex, Copilot CLI, or Gemini CLI, so treat it as format-compatible there, not verified.

Fix mode applies the smallest sufficient fix per finding (#[Locked] before visibility changes before restructuring), never weakens validation, never commits, and refuses to guess authorization rules — if a fix needs a product decision like "which policy applies here," it collects the question for you instead of inventing an answer.

Sample output

Trimmed (and lightly edited) from a real run against an earlier, smaller revision of the bundled fixture apps — the counts reflect that revision, not the current corpus:

# Livewire Security Audit - v4-app - 2026-06-11
Livewire version: v4 | Components audited: 9 | Findings: 17 (C/H/M/L/I: 5/4/3/3/2)

## Critical

### [LW-08] Live secrets (Stripe secret key, DB host) serialized into the page snapshot - resources/views/components/profile/⚡show.blade.php:22-27
Evidence: `'stripe_key' => config('services.stripe.secret'),` assigned to the public
array property `public $debugInfo = [];`
Exploit: `$debugInfo` is a plain PHP array, so it serializes in full into the
`wire:snapshot` attribute in the page HTML. Anyone who loads the page and view-sources
it reads the live Stripe secret key in plaintext.
Fix: fixes.md "computed-instead-of-public" - debug config values must never sit on a
client-serialized property.

### [LW-16] SQL injection via #[Url] filter interpolated into whereRaw - resources/views/components/feed/⚡index.blade.php:15
Evidence: `->whereRaw("category = '$this->filter'")`
Exploit: `$filter` is hydrated straight from the `?filter=` query string and
interpolated unbound into raw SQL. `/feed?filter=' OR '1'='1` controls the query, and
the route has no auth middleware, so this is reachable unauthenticated.
Fix: fixes.md "bind-not-interpolate" - bound where() plus an allowlist on $filter.

...

## Clean components
- resources/views/components/posts/⚡show.blade.php - model-typed property, authorized
  actions, validated input, escaped output.

## Discarded during verification
10 subagent findings did not survive main-thread verification.

What it does not cover

• General Laravel security outside Livewire (route audits, headers, CSP, CSRF config) — except where it intersects a Livewire surface.
• Livewire v2.
• Static-analysis tooling. This is a judgment-based audit run by a model against your actual code, not an AST scanner with pattern rules.

How the claims were verified

The reference files (references/snapshot-internals.md, references/checklist.md, references/fixes.md) cite the Livewire source file or docs page for every engine-behavior claim — what serializes into snapshots, exactly what #[Locked] throws, how model-typed properties are protected (it is not #[Locked], despite what the docs imply), what happens when a client sends a wrong-typed value, and which middleware actually re-runs on component updates. Verified against Livewire v4.3.0 and the official 3.x/4.x documentation. If you find a claim that is wrong for a newer Livewire version, please open an issue.

License

MIT