Understand why,
change safely.

Certificate → Request → Order → Challenge, resolved automatically across cert-manager, Argo, Keda, and any operator that sets ownerRefs.

Correlated events surface the exact resource and moment a rollout, sync, or issuance went sideways — with a plain-language root cause.

Edit the resource in a typed form or its manifest, and the API server's own dry run returns the exact diff it would apply — defaults and admission mutations included — before you commit it.

Every confirmation spells out the resource, its namespace, and the cluster — dotted in that cluster's environment color, and set in red when the cluster is marked production.

On a cluster marked production, the delete button stays disabled until you type the resource's name. Ergo guesses the environment from the cluster's name and lets you correct the guess.

A pod its controller owns will be recreated; an unowned one is gone for good. For anything else, the sheet says plainly that dependent objects may be garbage-collected.

Pick a replica count outside a HorizontalPodAutoscaler's bounds and Ergo names the bound you crossed, its current metric targets, and the fact that the autoscaler will override your number anyway.

A plain-JSON audit log on your Mac, one file per day and one line per action — failures written down with their reason, not only the changes that worked.

The local MCP tools are built on Ergo's read-only cluster interface, which has no mutating methods to call. Changes live behind a separate interface the agent surface never holds.