Guides

This section is for the moment something is wrong — or just was. Each guide starts from a symptom you might be staring at and walks the real investigation, command by command, to a diagnosis and a verified outcome. By the end of a guide you can run the same workflow on your own cluster, and you will know which commands answer which questions.

Which guide do I need?

What you’re seeing	Guide
Pods are crashing or won’t start, and you don’t know why	Investigate a broken workload
You shipped a deploy and it never finished rolling out	Your rollout is stuck
Memory or CPU keeps climbing and an OOM kill looks inevitable	Catch resource exhaustion early
A node went down and everything on it is failing at once	A node just died
The incident is over and you need to know what changed before it	What changed before the incident
You’d rather hit quota and capacity limits on your terms than in an outage	Capacity & quota ahead of time

Each guide walks the real workflow, using output captured during live validation drills (abridged, never invented):

• Investigate a broken workload — the bundle-first flow: root-causing a double fault in one call.
• Your rollout is stuck — rollout.stall fires while the old revision still serves; roll back to a verified resolved.
• Catch resource exhaustion early — a memory-leak forecast lands a session 14 minutes before the OOM kill.
• A node just died — storm correlation: one session for a 33-object blast, and its full member lifecycle.
• What changed before the incident — --at post-mortems from a copied sentinel store, offline.
• Capacity & quota ahead of time — the correlated quota incident, the drafted increase request, and the cloud sweep commands.

Every guide ends with a pointer to the matching agent skill in skills/ — the same workflows, packaged for the agents themselves.