PHPackages                             promptphp/intercept-injection-guard - PHPackages - PHPackages  [Skip to content](#main-content)[PHPackages](/)[Directory](/)[Categories](/categories)[Trending](/trending)[Leaderboard](/leaderboard)[Changelog](/changelog)[Analyze](/analyze)[Collections](/collections)[Log in](/login)[Sign up](/register)

1. [Directory](/)
2. /
3. [Security](/categories/security)
4. /
5. promptphp/intercept-injection-guard

ActiveLibrary[Security](/categories/security)

promptphp/intercept-injection-guard
===================================

Prompt injection guard middleware for Laravel AI agents.

v0.1.0(today)00MITPHPPHP ^8.4.0

Since Jun 19Pushed todayCompare

[ Source](https://github.com/promptphp/intercept-injection-guard)[ Packagist](https://packagist.org/packages/promptphp/intercept-injection-guard)[ Docs](https://intercept.promptphp.com)[ RSS](/packages/promptphp-intercept-injection-guard/feed)WikiDiscussions main Synced today

READMEChangelog (4)Dependencies (2)Versions (2)Used By (0)

promptphp/intercept-injection-guard
===================================

[](#promptphpintercept-injection-guard)

`PromptInjectionGuard` is a Laravel AI SDK agent middleware that detects common prompt injection attempts before an agent prompt reaches the AI provider.

It can block, log, warn, sanitize, or fully delegate handling to a custom callback.

> This middleware is a lightweight heuristic guard. It is designed to catch common prompt injection patterns, not to guarantee complete protection against every possible attack.

Features
--------

[](#features)

- Detects common prompt injection attempts.
- Supports custom regex patterns and merges custom patterns with the default patterns by default.
- Can replace the default patterns entirely.
- Supports `block`, `log`, `warn`, and `sanitize` actions.
- Decodes HTML entities and URL-encoded input.
- Logs safely using prompt hashes by default.
- Supports optional prompt previews in logs.
- Supports fully custom detection handling with a callback.

Supported actions
-----------------

[](#supported-actions)

ActionBehaviour`block`Throws a `PromptInjectionGuardException` and stops the prompt.`log`Logs the detection and allows the prompt to continue.`warn`Prepends a security warning and allows the prompt to continue.`sanitize`Removes the matched injection content, prepends a warning, and allows the prompt to continue.The recommended default action is `block`.

Configuration
-------------

[](#configuration)

No configuration is required. The middleware works out of the box using safe internal defaults.

```
new PromptInjectionGuard()
```

By default, this will:

- use the `block` action
- use the built-in prompt injection patterns
- merge any custom patterns with the built-in patterns
- normalise prompts before scanning
- avoid logging prompt previews

### Optional global config

[](#optional-global-config)

Intercept supports an optional shared config file:

```
config/intercept.php

```

This config file is used for global middleware defaults across the Intercept package.

You may publish it with:

```
php artisan vendor:publish --tag=intercept-config
```

### Configuration priority

[](#configuration-priority)

Configuration is resolved in this order:

```
constructor value > config value > internal middleware default

```

That means a constructor value always wins over the published config.

For example, if your config says:

```
'injection_guard' => [
    'action' => 'block',
],
```

You can still override it for a specific agent:

```
public function middleware(): array
{
    return [
        new PromptInjectionGuard(
            action: 'log',
        ),
    ];
}
```

In this case, the middleware will use `log` for that agent, even though the global config says `block`.

### Partial config is supported

[](#partial-config-is-supported)

You do not need to define every option in `config/intercept.php`.

For example, this is valid:

```
'injection_guard' => [
    'action' => 'log',
],
```

All missing options will fall back to the middleware's internal defaults.

Usage
-----

[](#usage)

Simply register and use the middleware on a Laravel AI agent.

```