What problem does it solve? Kubernetes storage failures on Huawei Cloud CCE—such as PVC Pending, FailedAttach, FailedMount, NFS timeouts, OBS 403 errors, and read-only filesystems—are hard to diagnose because evidence is scattered across PVCs, PVs, StorageClasses, VolumeAttachments, CSI logs, and cloud-side EVS/SFS/OBS metadata. This Skill correlates all of that read-only evidence into a ranked root-cause report. ## Core Features & Use Cases - Lifecycle-stage diagnosis: Classifies failures into provisioning, binding/topology, attach, mount, runtime I/O, network path, credential, and teardown stages with ranked causes, evidence, and counter-evidence. - Dual-plane evidence collection: Combines hcloud cluster and cloud-storage metadata (EVS, SFS, SFS Turbo, OBS, VPC, security groups) with read-only kubectl-cce queries of PVCs, PVs, VolumeAttachments, Events, and bounded CSI logs. - Strict safety boundaries: Read-only only—no mutations, finalizer removal, force detach, or exec; remediation is handed off to the auto-remediation skill after user confirmation. - Use Case: A Pod is stuck in ContainerCreating with FailedMount events on an SFS Turbo volume. The Skill inspects Pod events, CSI logs, DNS, routes, and security groups, then reports an NFS network-path root cause with confidence and next verification steps. ## Quick Start Ask the assistant to diagnose why PVC data-pvc in namespace default of CCE cluster <cluster-id> in region cn-north-4 is stuck in Pending and produce a root-cause report.
