Operations and audit

Prerequisites: staff keys declared by each operation (for example ops.read, ops.panic, users.read) and apiRequest. Authentication succeeds independently of authorization; consult the operation's permission list.

Investigate a change or incident

Read listAuditEvents with the documented actor/action/date filters. Include the x-request-id from the affected API response when reporting a problem. Use operations status and incidents to distinguish runtime dependency failure from platform maintenance. Stats endpoints provide aggregate views, not an event replay log.

The user CSV export shares user-list filters but exports only the first 100 matches and ignores page/per_page. Download it rather than parsing JSON:

curl 'https://api.example.com/v1/admin/users/export.csv' \
  -H "Authorization: Bearer $ACCESS_TOKEN" -o users.csv

See exportAdminUsersCsv. The response content type is text/csv with an attachment filename. Bulk user mutations return 200 with per-user results; inspect each item's error instead of treating 200 as all-or-nothing success.

Manage maintenance

getOpsStatus exposes the current state. setOpsPanicMode requires ops.panic. To request degraded mode for a planned maintenance window:

const state = await apiRequest("/v1/admin/ops/panic", {
  method: "POST", body: JSON.stringify({ level: "degraded", reason: "Planned maintenance window" }),
});
console.log(state.panic);

degraded blocks new user checkout and game starts. lockdown blocks other public/user writes as well. Reads, staff routes and the user's session-stop operation stay available. To end maintenance, set level: "off" with a reason. Blocked requests return 503 panic_mode plus Retry-After: 30; an unavailable external dependency has a different error body.

Alerts and jobs

Alert subscriptions specify roles, severity and incident/panic event types. Test alert delivery before relying on it, then inspect incident history for the result. Notification broadcasts can target a segment and contain in-app and/or email channels; announcements.write or email.send is accepted at the route gate, with channel rules in the implementation/reference.

System cron operations require their own permissions. Inspect a job before toggling or running it, and treat a trigger response as job acceptance rather than proof of all downstream work completing. Errors and limits covers recovery and maintenance behavior.