Sync package updates and check QUERY requests for request forgery

.github/workflows/tests.yml

@@ -131,3 +131,15 @@ jobs:
 
       - name: Type-check generated Wayfinder output
         run: pnpm typecheck:wayfinder
+
+      # Telescope's dashboard is a standalone npm package whose devEngines require npm 11.10 or newer.
+      - name: Install npm for Telescope
+        run: npm install --global npm@11.21.0
+
+      - name: Install Telescope dependencies
+        working-directory: src/telescope
+        run: npm ci
+
+      - name: Run Telescope dashboard tests
+        working-directory: src/telescope
+        run: npm test

src/ai/README.md

@@ -27,7 +27,7 @@ Skill discovery is cached for the fixed paths in `ai.skills.cached_paths`. [Relo
 
 The conversation migrations use Hypervel's column types, indexes, and approval claim fields. Existing Laravel conversation data requires a [schema migration](https://hypervel.org/docs/ai-sdk#migrating-conversation-data) before use.
 
-Remote downloads share the HTTP client's destination checks. The protected `UntrustedUrl::validate()`, `resolve()`, and `isBlocked()` helpers and hostname-blocklist constants are removed; `resolveUsing()` remains available. URLs containing embedded credentials are rejected. Downloads are limited to 32 MiB of decoded content by default; configure `ai.remote_files.max_size` to change or disable the limit. See [remote attachments](https://hypervel.org/docs/ai-sdk#remote-attachments).
+Remote downloads share the HTTP client's destination checks. The protected `UntrustedUrl::validate()`, `resolve()`, and `isBlocked()` helpers and hostname-blocklist constants are removed; `resolveUsing()` remains available. URLs containing embedded credentials are rejected. Downloads are limited to 32 MiB of decoded content by default, instead of Laravel's 100 MB; set `ai.remote_files.max_size` in kilobytes to change the limit, or to `null` to disable it. See [remote attachments](https://hypervel.org/docs/ai-sdk#remote-attachments).
 
 ## Known Issues
 

src/ai/config/ai.php

@@ -89,13 +89,13 @@
     |
     | Remote attachments must resolve to public addresses unless their host
     | is listed below. The size limit applies to each decoded download in
-    | bytes. Set max_size to null to allow downloads of any size.
+    | kilobytes. Set max_size to null to allow downloads of any size.
     |
     */
 
     'remote_files' => [
         'allowed_hosts' => [],
-        'max_size' => 32 * 1024 * 1024,
+        'max_size' => 32 * 1024,
     ],
 
     /*

src/ai/src/Files/UntrustedUrl.php

@@ -14,15 +14,17 @@
 use Hypervel\Http\Client\Response;
 use Hypervel\Support\Facades\Http;
 use InvalidArgumentException;
+use Psr\Http\Message\ResponseInterface;
+use RuntimeException;
 use Swoole\Coroutine\CanceledException;
 use Throwable;
 
 class UntrustedUrl
 {
-    public const int DEFAULT_MAX_SIZE = 32 * 1024 * 1024;
-
     protected const int MAX_REDIRECTS = 5;
 
+    public const int MAX_KILOBYTES = 32 * 1024;
+
     // REMOVED: hostname blocklists; the HTTP destination policy checks the resolved addresses.
 
     protected static ?Closure $resolver = null;
@@ -36,16 +38,18 @@ public static function fetch(string $url): Response
     {
         $allowedHosts = config()->array('ai.remote_files.allowed_hosts', []);
         $policy = new RemoteFileDestinationPolicy(static::$resolver);
-        $maxSize = config()->get('ai.remote_files.max_size', static::DEFAULT_MAX_SIZE);
+        $kilobytes = config()->get('ai.remote_files.max_size', static::MAX_KILOBYTES);
 
-        if ($maxSize !== null && (! is_int($maxSize) || $maxSize <= 0)) {
+        if ($kilobytes !== null && (! is_int($kilobytes) || $kilobytes <= 0)) {
             throw new InvalidArgumentException('The remote file maximum size must be a positive integer or null.');
         }
 
+        $limit = $kilobytes === null ? null : $kilobytes * 1024;
+
         for ($hop = 0; $hop <= static::MAX_REDIRECTS; ++$hop) {
             $uri = new Uri($url);
             $options = [];
-            $tooLarge = null;
+            $tooLarge = false;
             $trusted = in_array(rtrim(trim($uri->getHost(), '[]'), '.'), $allowedHosts, true);
 
             if ($trusted) {
@@ -54,20 +58,30 @@ public static function fetch(string $url): Response
                 }
             }
 
-            if ($maxSize !== null) {
-                $buffer = Utils::streamFor();
+            if ($limit !== null) {
+                $body = Utils::streamFor();
+
+                // The transport drops a compressed response's Content-Length before this runs, so only decoded lengths are checked.
+                $options['on_headers'] = static function (ResponseInterface $response) use ($limit, &$tooLarge): void {
+                    if ((int) $response->getHeaderLine('Content-Length') > $limit) {
+                        $tooLarge = true;
 
-                // Count decoded bytes at the sink; transfer sizes may describe compressed data.
-                $options['sink'] = FnStream::decorate($buffer, [
-                    'write' => static function (string $chunk) use ($buffer, $maxSize, &$tooLarge): int {
-                        if (strlen($chunk) > $maxSize - $buffer->tell()) {
-                            $tooLarge = new InvalidArgumentException("The remote file exceeds the maximum size of [{$maxSize}] bytes.");
+                        throw new RuntimeException('The remote file declares a body larger than the size limit.');
+                    }
+                };
+
+                // Count decoded bytes at the sink, which also bounds compressed and chunked bodies that declare no decoded length.
+                $options['sink'] = FnStream::decorate($body, [
+                    'write' => static function (string $chunk) use ($body, $limit, &$tooLarge): int {
+                        // The sink appends sequentially, so tell() gives its size without a stat call for every chunk.
+                        if (strlen($chunk) > $limit - $body->tell()) {
+                            $tooLarge = true;
 
                             // A short write aborts the transfer while preserving transport cleanup.
                             return 0;
                         }
 
-                        return $buffer->write($chunk);
+                        return $body->write($chunk);
                     },
                 ]);
             }
@@ -87,7 +101,11 @@ public static function fetch(string $url): Response
             try {
                 $response = $request->get((string) $uri);
             } catch (Throwable $exception) {
-                throw $exception instanceof CanceledException ? $exception : $tooLarge ?? $exception;
+                if ($tooLarge && ! $exception instanceof CanceledException) {
+                    throw new InvalidArgumentException("The remote file exceeds the maximum size of {$kilobytes} kilobytes.");
+                }
+
+                throw $exception;
             }
 
             if (! $response->redirect() || blank($response->header('Location'))) {

src/ai/src/Gateway/Gemini/Concerns/MapsMessages.php

@@ -67,6 +67,11 @@ protected function mapAssistantMessage(AssistantMessage|Message $message, array
             return;
         }
 
+        // A signature never rides on a function call, so a persisted turn opens with the rebuilt thought step that held it...
+        if ($message instanceof AssistantMessage && filled($signature = $message->toolCalls->first()?->thoughtSignature)) {
+            $input[] = ['type' => 'thought', 'signature' => $signature];
+        }
+
         if (filled($message->content)) {
             $input[] = [
                 'type' => 'model_output',
@@ -75,11 +80,6 @@ protected function mapAssistantMessage(AssistantMessage|Message $message, array
         }
 
         if ($message instanceof AssistantMessage) {
-            // A signature never rides on a function call, so a persisted turn rebuilds the thought step that held it...
-            if (filled($signature = $message->toolCalls->first()?->thoughtSignature)) {
-                $input[] = ['type' => 'thought', 'signature' => $signature];
-            }
-
             foreach ($message->toolCalls as $toolCall) {
                 $input[] = [
                     'type' => 'function_call',

src/docs/ai-sdk.md

@@ -1063,16 +1063,16 @@ Downloads inherit your application's outgoing proxy configuration; no separate A
 
 Pinning an untrusted hostname requires cURL with SSL support and libcurl 7.75 or later; see the HTTP client's [destination policy limitations](/docs/{{version}}/http-client#destination-policy-limitations). Explicitly trusted hosts bypass destination checks and do not require hostname pinning.
 
-Downloads are limited to 32 MiB of decoded content by default. You may set a different positive byte limit, or `null` to disable the limit, in your `config/ai.php` file:
+Downloads are limited to 32 MiB of decoded content by default. You may set a different limit in kilobytes, or `null` to disable the limit, in your `config/ai.php` file:
 
 ```php
 'remote_files' => [
     'allowed_hosts' => ['files.internal.example'],
-    'max_size' => 64 * 1024 * 1024,
+    'max_size' => 64 * 1024,
 ],
 ```
 
-The limit also applies to allowed hosts and compressed responses. A blocked destination or oversized download throws an `InvalidArgumentException`. The limit applies to each file; concurrent downloads and preparing files for a provider can use additional memory.
+The limit also applies to allowed hosts and compressed responses. A response that declares a larger `Content-Length` is rejected before its body is downloaded. A blocked destination or oversized download throws an `InvalidArgumentException`. The limit applies to each file; concurrent downloads and preparing files for a provider can use additional memory.
 
 If a file requires HTTP authentication, download it using the HTTP client's authentication methods and pass the downloaded content or local path to the SDK. The attachment's `withHeaders` method configures the upload to the AI provider, not the remote download.
 
@@ -1642,7 +1642,7 @@ When using Anthropic, additional provider-specific options may be passed to the
 <a name="file-storage-tools"></a>
 ### File Storage Tools
 
-The `FileStorage` tool factory allows you to give agents access to a Hypervel [filesystem disk](/docs/{{version}}/filesystem). The `all` method returns tools that allow the agent to list, read, inspect, generate URLs for, write, delete, and copy files on the given disk:
+The `FileStorage` tool factory allows you to give agents access to a Hypervel [filesystem disk](/docs/{{version}}/filesystem). The `all` method returns tools that allow the agent to list, read, inspect, generate URLs for, write, delete, copy, and move files on the given disk:
 
 ```php
 use Hypervel\Ai\Tools\FileStorage;
@@ -1826,6 +1826,9 @@ To refine search results based on user location, use the `location` method:
 );
 ```
 
+> [!NOTE]
+> Groq only supports web search on its GPT-OSS models, and ignores the `max`, `allow`, and `location` methods.
+
 <a name="web-fetch"></a>
 #### Web Fetch
 
@@ -1921,6 +1924,9 @@ When using OpenAI or Azure, you may make [stored files](#files) available to the
 ]);
 ```
 
+> [!NOTE]
+> Groq only supports code execution on its GPT-OSS models.
+
 <a name="sub-agents"></a>
 ### Sub-Agents
 

src/docs/blade.md

@@ -1881,6 +1881,8 @@ You may push to a stack as many times as needed. To render the complete stack co
 </head>
 ```
 
+The `@stack` directive renders the content that has been pushed at the moment it is reached. Content pushed afterwards, such as from a component rendered below the `@stack` directive in the same layout component, will not be included. If you need to push from such components, place the `@stack` directive after them, for example at the end of the `<body>` tag.
+
 If you would like to prepend content onto the beginning of a stack, you should use the `@prepend` directive:
 
 ```blade

src/docs/csrf.md

@@ -31,7 +31,7 @@ Without CSRF protection, a malicious website could create an HTML form that poin
 
 If the malicious website automatically submits the form when the page is loaded, the malicious user only needs to lure an unsuspecting user of your application to visit their website and their email address will be changed in your application.
 
-To prevent this vulnerability, we need to inspect every incoming `POST`, `PUT`, `PATCH`, or `DELETE` request for a secret session value that the malicious application is unable to access.
+To prevent this vulnerability, we need to inspect every incoming `POST`, `PUT`, `PATCH`, `DELETE`, or `QUERY` request for a secret session value that the malicious application is unable to access.
 
 <a name="preventing-csrf-requests"></a>
 ## Preventing CSRF Requests
@@ -147,4 +147,4 @@ Hypervel stores the current CSRF token in an encrypted `XSRF-TOKEN` cookie that
 This cookie is primarily sent as a developer convenience since some JavaScript frameworks and libraries, like Angular and Axios, automatically place its value in the `X-XSRF-TOKEN` header on same-origin requests.
 
 > [!NOTE]
-> For JavaScript requests, read the `XSRF-TOKEN` cookie and send its value in the `X-XSRF-TOKEN` header on same-origin `POST`, `PUT`, `PATCH`, and `DELETE` requests.
+> For JavaScript requests, read the `XSRF-TOKEN` cookie and send its value in the `X-XSRF-TOKEN` header on same-origin `POST`, `PUT`, `PATCH`, `DELETE`, and `QUERY` requests.

src/docs/http-tests.md

@@ -70,8 +70,6 @@ $response = $this->query('/search', ['filter' => 'active']);
 $response = $this->queryJson('/search', ['filter' => 'active']);
 ```
 
-Because these requests are simulated, they aren't affected by the [HTTP/1.1 limitation on `QUERY` routes](/docs/{{version}}/routing#available-router-methods) that applies in production.
-
 Instead of returning a `Hypervel\Http\Response` instance, test request methods return an instance of `Hypervel\Testing\TestResponse`, which provides a [variety of helpful assertions](#available-assertions) that allow you to inspect your application's responses:
 
 ```php tab=Pest