feat: ai

.agents/skills/configuring-horizon/references/metrics.md

@@ -0,0 +1,21 @@
+# Metrics & Snapshots
+
+## Where to Find It
+
+Search with `search-docs`:
+- `"horizon metrics snapshot"` for the snapshot command and scheduling
+- `"horizon trim snapshots"` for retention configuration
+
+## What to Watch For
+
+### Metrics dashboard stays blank until `horizon:snapshot` is scheduled
+
+Running `horizon` artisan command does not populate metrics automatically. The metrics graph is built from snapshots, so `horizon:snapshot` must be scheduled to run every 5 minutes via Laravel's scheduler.
+
+### Register the snapshot in the scheduler rather than running it manually
+
+A single manual run populates the dashboard momentarily but will not keep it updated. Search `"horizon metrics snapshot"` for the exact scheduler registration syntax, which differs between Laravel 10 and 11+.
+
+### `metrics.trim_snapshots` is a snapshot count, not a time duration
+
+The `trim_snapshots.job` and `trim_snapshots.queue` values in `config/horizon.php` are counts of snapshots to keep, not minutes or hours. With the default of 24 snapshots at 5-minute intervals, that provides 2 hours of history. Increase the value to retain more history at the cost of Redis memory usage.

.agents/skills/configuring-horizon/references/notifications.md

@@ -0,0 +1,21 @@
+# Notifications & Alerts
+
+## Where to Find It
+
+Search with `search-docs`:
+- `"horizon notifications"` for Horizon's built-in notification routing helpers
+- `"horizon long wait detected"` for LongWaitDetected event details
+
+## What to Watch For
+
+### `waits` in `config/horizon.php` controls the LongWaitDetected threshold
+
+The `waits` array (e.g., `'redis:default' => 60`) defines how many seconds a job can wait in a queue before Horizon fires a `LongWaitDetected` event. This value is set in the config file, not in Horizon's notification routing. If alerts are firing too often or too late, adjust `waits` rather than the routing configuration.
+
+### Use Horizon's built-in notification routing in `HorizonServiceProvider`
+
+Configure notifications in the `boot()` method of `App\Providers\HorizonServiceProvider` using `Horizon::routeMailNotificationsTo()`, `Horizon::routeSlackNotificationsTo()`, or `Horizon::routeSmsNotificationsTo()`. Horizon already wires `LongWaitDetected` to its notification sender, so the documented setup is notification routing rather than manual listener registration.
+
+### Failed job alerts are separate from Horizon's documented notification routing
+
+Horizon's 12.x documentation covers built-in long-wait notifications. Do not assume the docs provide a `JobFailed` listener example in `HorizonServiceProvider`. If a user needs failed job alerts, treat that as custom queue event handling and consult the queue documentation instead of Horizon's notification-routing API.

.agents/skills/configuring-horizon/references/supervisors.md

@@ -0,0 +1,27 @@
+# Supervisor & Balancing Configuration
+
+## Where to Find It
+
+Search with `search-docs` before writing any supervisor config, as option names and defaults change between Horizon versions:
+- `"horizon supervisor configuration"` for the full options list
+- `"horizon balancing strategies"` for auto, simple, and false modes
+- `"horizon autoscaling workers"` for autoScalingStrategy details
+- `"horizon environment configuration"` for the defaults and environments merge
+
+## What to Watch For
+
+### The `environments` array merges into `defaults` rather than replacing it
+
+The `defaults` array defines the complete base supervisor config. The `environments` array patches it per environment, overriding only the keys listed. There is no need to repeat every key in each environment block. A common pattern is to define `connection`, `queue`, `balance`, `autoScalingStrategy`, `tries`, and `timeout` in `defaults`, then override only `maxProcesses`, `balanceMaxShift`, and `balanceCooldown` in `production`.
+
+### Use separate named supervisors to enforce queue priority
+
+Horizon does not enforce queue order when using `balance: auto` on a single supervisor. The `queue` array order is ignored for load balancing. To process `notifications` before `default`, use two separately named supervisors: one for the high-priority queue with a higher `maxProcesses`, and one for the low-priority queue with a lower cap. The docs include an explicit note about this.
+
+### Use `balance: false` to keep a fixed number of workers on a dedicated queue
+
+Auto-balancing suits variable load, but if a queue should always have exactly N workers such as a video-processing queue limited to 2, set `balance: false` and `maxProcesses: 2`. Auto-balancing would scale it up during bursts, which may be undesirable.
+
+### Set `balanceCooldown` to prevent rapid worker scaling under bursty load
+
+When using `balance: auto`, the supervisor can scale up and down rapidly under bursty load. Set `balanceCooldown` to the number of seconds between scaling decisions, typically 3 to 5, to smooth this out. `balanceMaxShift` limits how many processes are added or removed per cycle.

.agents/skills/configuring-horizon/references/tags.md

@@ -0,0 +1,21 @@
+# Tags & Silencing
+
+## Where to Find It
+
+Search with `search-docs`:
+- `"horizon tags"` for the tagging API and auto-tagging behaviour
+- `"horizon silenced jobs"` for the `silenced` and `silenced_tags` config options
+
+## What to Watch For
+
+### Eloquent model jobs are tagged automatically without any extra code
+
+If a job's constructor accepts Eloquent model instances, Horizon automatically tags the job with `ModelClass:id` such as `App\Models\User:42`. These tags are filterable in the dashboard without any changes to the job class. Only add a `tags()` method when custom tags beyond auto-tagging are needed.
+
+### `silenced` hides jobs from the dashboard completed list but does not stop them from running
+
+Adding a job class to the `silenced` array in `config/horizon.php` removes it from the completed jobs view. The job still runs normally. This is a dashboard noise-reduction tool, not a way to disable jobs.
+
+### `silenced_tags` hides all jobs carrying a matching tag from the completed list
+
+Any job carrying a matching tag string is hidden from the completed jobs view. This is useful for silencing a category of jobs such as all jobs tagged `notifications`, rather than silencing specific classes.