AI-drafted incident suggestions from telemetry

This one is the biggest quality-of-life feature we've ever shipped.

You're at dinner. Your API starts returning 500s. Your uptime monitor catches it. Under the old flow, you'd have to open your laptop, navigate to the incident panel, write a title, pick a severity, write a body, and click post. Ten minutes of typing in the middle of a fire.

Now, LiveStatus drafts it for you.

Here's how it works:

1. Uptime monitor sees two consecutive failed checks for a service
2. A suggestion row is created with the failure metadata (duration, status code, response time)
3. Your dashboard shows an amber banner at the top of the incidents panel: "We noticed API has been down for 3 minutes. Draft an incident?"
4. Click Draft. LiveStatus calls Claude via OpenRouter to generate a real incident title, severity classification, and body text based on what the monitor observed
5. You review, edit anything you want, click Post. The real incident is created, subscribers notified, Slack pinged. 30 seconds total.

The AI never auto-posts. You always review. And it only fires after 2+ consecutive failures, not one blip, so a flaky network check doesn't drag you into your dashboard.

This is the kind of thing that exists because LiveStatus is the only status page product with the telemetry and the LLM integration in the same place. Statuspage.io can't do this because they don't watch your services. Datadog can watch but can't communicate to customers. LiveStatus does both.

Shipped with 22 new tests covering the detection logic, JSON parsing of LLM responses, and the full accept-to-incident flow.