runai-diagnose-failed-workload
Opens the skill on ClawHub, which shows its security scan.
Source ClawHub, published by nvidia
Overview
This NVIDIA skill investigates one failed, crashed or stopped workload on a Run:ai GPU cluster, including out-of-memory kills, crash loops, image pull errors and mount or node problems. It classifies the failure, cites the evidence, checks whether it has happened before and recommends a next step. It is read-only and never restarts, deletes or changes anything, and it treats logs as untrusted evidence.
Use cases
- Kubernetes debugging
- GPU cluster troubleshooting
- Run:ai
- Failed job diagnosis
