Keep technology logos and images inside text safe from pruning

Media::referencedPaths() is what `content:prune --media` asks before
deleting a file, and it did not know about two kinds of image a page
shows:

- A technology's logo. Every logo on the radar looked orphaned, and one
  prune would have deleted all of them.
- Images inside rich text — a page's body, a project's description,
  decisions and postmortem — whether attached in the panel's editor or
  uploaded through the API and written in as <img src>. They are now
  found in the HTML, in every language, bin included.

The page body's editor also states where attachments go — the media
disk, pages/content, public — instead of relying on Filament's
defaults: the HTML reaches the frontend as it is, and a private file
would be a broken image there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

app/Filament/Resources/Pages/Schemas/PageForm.php

@@ -75,6 +75,15 @@ class PageForm
 
                         RichEditor::make("body.{$locale}")
                             ->label(__('panel.pages.fields.body'))
+                            // Images in the text go where every other upload
+                            // does, public: the HTML travels to the frontend
+                            // as it is, with each image's URL written in, and
+                            // a private file would be a broken image there.
+                            // Media::referencedPaths() finds them in the body,
+                            // so pruning never takes one a page still shows.
+                            ->fileAttachmentsDisk(config('media.disk'))
+                            ->fileAttachmentsDirectory('pages/content')
+                            ->fileAttachmentsVisibility('public')
                             ->columnSpanFull(),
                     ]),
                 ]),

app/Models/Media.php

@@ -183,6 +183,7 @@ class Media extends Model
             [Member::class, 'avatar'],
             [Client::class, 'logo'],
             [Testimonial::class, 'avatar'],
+            [Technology::class, 'logo'],
             [ImageAttachment::class, 'path'],
         ];
 
@@ -198,9 +199,52 @@ class Media extends Model
             $paths = [...$paths, ...$query->pluck($column)->all()];
         }
 
+        // Images placed inside rich text — attached in the panel's editor, or
+        // uploaded through the API and written into a body as `<img src>`. No
+        // column holds their path, so without this they would look orphaned
+        // and `content:prune --media` would delete them from under the page.
+        $richText = [
+            [Page::class, ['body']],
+            [Project::class, ['description', 'decisions', 'postmortem']],
+        ];
+
+        foreach ($richText as [$model, $columns]) {
+            $model::query()
+                ->withTrashed()
+                ->select(['id', ...$columns])
+                ->chunkById(200, function ($records) use ($columns, &$paths): void {
+                    foreach ($records as $record) {
+                        foreach ($columns as $column) {
+                            foreach ($record->getTranslations($column) as $html) {
+                                $paths = [...$paths, ...static::pathsEmbeddedIn((string) $html)];
+                            }
+                        }
+                    }
+                });
+        }
+
         return array_values(array_unique(array_filter($paths)));
     }
 
+    /**
+     * The stored paths of every image on the media disk a piece of HTML links
+     * to, whether written with the disk's absolute URL or its path alone.
+     *
+     * @return array<int, string>
+     */
+    public static function pathsEmbeddedIn(string $html): array
+    {
+        $base = (string) parse_url(Storage::disk((string) config('media.disk', 'public'))->url(''), PHP_URL_PATH);
+
+        if ($base === '' || ! str_contains($html, $base)) {
+            return [];
+        }
+
+        preg_match_all('#'.preg_quote(rtrim($base, '/'), '#').'/([^"\'\s?\#<>]+)#', $html, $matches);
+
+        return array_map('rawurldecode', $matches[1]);
+    }
+
     /**
      * Whether any record still points at this file.
      *