Kubernetes operations
See cluster inventory, health, logs and diagnostics in one place.
Available views
| View | Status | Benefit |
|---|---|---|
| Clusters and workloads | Available | See the current state quickly. |
| Pod logs and diagnostics | Available | Find the cause of a problem faster. |
Read-only inventory
Inventory and diagnostics are scoped, authorized, bounded, and audited. Reads should report cluster identity, namespaces, workloads, pods, health, and safe metadata without leaking Secret values, bearer tokens, kubeconfigs, or raw dashboard endpoints.
- ✓Use the least-privileged ClusterPilot role and cluster credential boundary.
- ✓Apply namespace, label, size, and time limits before retrieving logs or inventory.
- ✓Preserve correlation IDs when escalating a read failure.
- ✓Do not turn a diagnostics endpoint into a generic kubectl proxy.
Sensitive mutation contract
High-risk operations require separate capability gates, authorization, admission preview, target restrictions, bounded runtime, audit, and evidence. v0.0.5 does not provide a general-purpose remote shell or unrestricted cluster mutation plane.
| Risk | Required controls |
|---|---|
| Remote command | Explicit target, command policy, no shell interpolation, bounded input/output/time, isolated execution. |
| Port forwarding | Approved service/port, short-lived session, origin and user binding, complete audit. |
| Secret/RBAC change | Typed operation, least privilege, diff preview, dual control where required, redacted evidence. |
| Helm/generic mutation | Immutable artifacts, server-side validation, conflict detection, health and rollback contract. |
Pod diagnostics
Start with workload status, conditions, events, container state, image identity, readiness, and bounded redacted logs. InvalidImageName, repeated @sha256:, pull-secret failures, unschedulable nodes, and failing probes require different recovery paths.