What the agent does
The agent reads how the panel already builds dashboards. In an admin shaped like filamentphp/demo, composer.json pins filament/filament at ^5.0 on laravel/framework ^12 with pestphp/pest ^3, larastan/larastan and laravel/pint, and its test script runs pest --parallel then phpstan analyse at level 8 over app/. app/Providers/Filament/AdminPanelProvider.php discovers pages under app/Filament/Pages and widgets under app/Filament/Widgets, so a new page needs no registration. There are three dashboards: Dashboard.php (the welcome page), ShopDashboard.php at routePath shop with a filtersForm of startDate, endDate, orderStatuses and productCategory and eight widgets, and HrDashboard.php at hr with five. Fourteen widgets live in app/Filament/Widgets, and each has a test under tests/Filament/Widgets.
The patterns are concrete. ShopKpisStats extends StatsOverviewWidget, uses InteractsWithPageFilters to read the page's dates, and returns Stat::make() entries with a description, a sparkline from ->chart() and a colour; CustomerGrowthChart extends ChartWidget, returns line from getType() and builds month labels from the filter range in getData(); FlaggedOrders extends TableWidget with a full column span and a query, columns and a default sort. The blog has what a dashboard needs: App\Models\Blog\Post with a published_at date cast, author(), postCategory() and a comments() morph, and App\Models\Comment with an is_visible boolean. PostFactory gives published_at a range from six months back to a month ahead, so scheduled posts exist, and CommentFactory flips is_visible at random.
The agent adds app/Filament/Pages/BlogDashboard.php at routePath blog with the next navigationSort and a filtersForm of startDate, endDate and a postCategory select from PostCategory::pluck('name', 'id'), and four widgets in app/Filament/Widgets: BlogKpisStats with published posts, scheduled posts (a published_at in the future), comments, and hidden comments awaiting review, each with a monthly sparkline; PostsPerMonthChart, a bar chart over the filter range; PostsByCategoryChart, a doughnut; and CommentsAwaitingReview, a table widget of comments where is_visible is false, sorted oldest first, with a link to the commented post. Every widget reads the page filters the way ShopKpisStats does, and the navigation icon resolves through DemoIconAlias with a Heroicon fallback like the other two dashboards.
Tests follow the existing files one for one: a tests/Filament/Widgets/<Widget>Test.php per widget that creates posts and comments with the factories and asserts Livewire::test() renders, and tests/Filament/Pages/BlogDashboardTest.php that requests BlogDashboard::getUrl() and asserts 200, as ShopDashboardTest does. tests/Pest.php already wraps everything under tests/Filament in RefreshDatabase and signs in a user. The agent runs composer test, which is Pest in parallel plus PHPStan at level 8, and vendor/bin/pint --test, and pushes only when all three are clean.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "A Linear issue labelled agent asks for a Blog dashboard in the admin panel: a page at /admin/blog with date and category filters, KPI stats for published posts, scheduled posts, comments and comments awaiting review, a posts-per-month chart, a posts-by-category chart, and a table of comments awaiting review. Read composer.json, app/Providers/Filament/AdminPanelProvider.php, app/Filament/Pages/ShopDashboard.php, app/Filament/Pages/HrDashboard.php, app/Filament/Widgets/ShopKpisStats.php, app/Filament/Widgets/CustomerGrowthChart.php, app/Filament/Widgets/FlaggedOrders.php, app/Models/Blog/Post.php, app/Models/Comment.php, database/factories/Blog/PostFactory.php, database/factories/CommentFactory.php, tests/Pest.php, tests/Filament/Pages/ShopDashboardTest.php and one widget test first. Add app/Filament/Pages/BlogDashboard.php with a filtersForm and four widgets under app/Filament/Widgets that read the page filters the way ShopKpisStats does, with a test per widget and a page test in the same style as the existing ones. Change no existing page or widget. Run composer test and vendor/bin/pint --test until both pass, and open a pull request that links the issue and lists the widgets.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request on its own branch. It adds app/Filament/Pages/BlogDashboard.php, four files under app/Filament/Widgets, four tests under tests/Filament/Widgets and tests/Filament/Pages/BlogDashboardTest.php, and changes nothing that exists; the Shop and HR dashboards and their widgets are untouched. The description quotes the issue, lists each widget with the query behind it, and names the route.
The checks are the proof. The repository's tests workflow runs on the pull request with PHP 8.4 and SQLite, builds the assets and runs composer test:pest; the phpstan workflow runs composer test:phpstan at level 8. The trajectory at GET /o/{org}/sessions/{id}/messages keeps the local Pest, PHPStan and Pint output beside the commands, so a reviewer sees the suite ran on the new widgets rather than reading that it did.
Before merging, a reviewer opens the page in a local checkout seeded with composer setup to see the charts with data, checks the definition of a scheduled post against how their editors use published_at, and decides whether comments awaiting review belong on this dashboard or on the comment resource. The format workflow that auto-commits Pint fixes runs only on the default branch, so the pull request has to be clean on its own, which is why the agent runs Pint before pushing.
Run it on every Linear issue labelled agent
Connect Linear once from Settings → Connections, then add an automation to the Project that holds the panel with this prompt and a linear trigger on linear.issue.labeled, naming the label in config.labels. The trigger appends the labelled issue after the prompt and fills in no placeholder, so the agent reads the widgets and the route from the issue text. Each labelled issue starts its own run in its own sandbox and ends in its own pull request, up to the automation's max_runs_per_hour.
A handler on your own server can do the same: when Linear's webhook reports the label, it sends POST /o/{org}/sessions with the repository and https://linear.app/<workspace>/issue/<ID> in resources, which puts the issue in the prompt through your Linear connection, with a request_id that is a UUID v5 of the issue id so a redelivered webhook returns the run it already started.
The use case this job follows, with its automation and trigger: Turn a Linear label into a PR.
Why run it on Rush?
A Filament dashboard is judged by whether it renders with real data, and that means a seeded database, a built asset bundle and the full Pest and PHPStan run for every change to a query. On a laptop that loop takes a running database, the right PHP and a Node build, and a level 8 PHPStan pass that is easy to skip when it is slow. In a Rush sandbox the agent sets that up once, runs composer test after each widget, reads the failure and fixes the query, and pushes only when the suite and the type analysis are clean.
The pull request carries its proof: the repository's own tests and phpstan workflows run on it, and the transcript keeps the local output beside the commands. A second issue, for an authors dashboard or a tag cloud widget, is a second sandbox and a second pull request at the same time.
Frequently asked questions
Does it register the page or the widgets anywhere?
No. AdminPanelProvider discovers pages under app/Filament/Pages and widgets under app/Filament/Widgets by namespace, so new classes in those directories appear on their own; the page lists its widgets in getWidgets() the way ShopDashboard does. The agent changes no provider and no existing page.
What counts as a scheduled post?
A post whose published_at is after now, which is the only state the model records: Post casts published_at as a date and has no status column. PostFactory sets published_at anywhere from six months back to a month ahead, so the tests have both kinds. If your editors use a separate status, the prompt can name it and the stat changes accordingly.
Will the filters work like the Shop dashboard's?
Yes. BlogDashboard uses the same HasFiltersForm trait with startDate, endDate and a postCategory select, and each widget reads them through InteractsWithPageFilters as ShopKpisStats and CustomerGrowthChart do, falling back to the last twelve months when the dates are empty. The category filter applies to posts and to the comments table through the commented post.
Why does PHPStan matter for a dashboard?
Because composer test runs it at level 8 over app/ and the phpstan workflow runs it on every pull request, so a widget with an untyped closure or a wrong return type fails the check even when it renders. The agent runs it locally before pushing and fixes the types rather than lowering the level.
Can it change the existing dashboards too?
Not in this run. The prompt says to change no existing page or widget, so the Shop and HR dashboards keep their widgets and sort order. A follow-up issue can move a widget onto the welcome Dashboard page or rename the navigation group once the team has seen the new page.