The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads data from the Kubernetes environment which could contain untrusted content such as log messages or event descriptions.
Ingestion points: The skill ingests data through kubectl get pods, kubectl get events, and kubectl exec outputs.
Boundary markers: No specific boundary markers or instructions are provided to the agent to treat tool output as untrusted data.
Capability inventory: The skill uses the Bash tool to run kubectl commands for triage and deployment rollbacks.
Sanitization: The instructions include manual steps for users to redact evidence, but no automated sanitization is applied to the data processed by the agent.
[COMMAND_EXECUTION]: The instructions utilize kubectl to interact with the cluster, including executing commands within pods (nvidia-smi) and performing deployment rollbacks.
[EXTERNAL_DOWNLOADS]: The skill provides a link to the official CoreWeave status page for monitoring service health.

coreweave-incident-runbook