The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses various AWS CLI commands (e.g., describe-cluster, list-cluster-nodes, describe-security-groups) to gather cluster state and executes shell scripts on compute nodes via AWS Systems Manager (SSM) to perform low-level networking and resource checks.
[PRIVILEGE_ESCALATION]: The diagnostic scripts execute shell commands with sudo on remote nodes via SSM, and the remediation guidance suggests commands requiring elevated privileges for managing system services (systemctl), logs (journalctl), and kernel parameters (sysctl).
[DYNAMIC_EXECUTION]: The triage script scripts/triage-cluster.sh dynamically generates shell command payloads that are Base64-encoded for transport and executed on HyperPod nodes using the SSM start-session mechanism with the AWS-StartNonInteractiveCommand document.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external data from cluster events and on-node system logs (dmesg) to diagnose failures and suggest specific user actions.
Ingestion points: Cluster events fetched via the sagemaker:ListClusterEvents API and on-node logs retrieved via SSM in scripts/triage-cluster.sh.
Boundary markers: None explicitly implemented in log parsing logic, although the skill presents proposed commands for user review before execution.
Capability inventory: Subprocess calls for AWS CLI and kubectl are present in triage-cluster.sh, check-efa-sg.sh, and check-vpc-config.sh. Remote shell execution via SSM is used in triage-cluster.sh. Network requests to IMDS are performed in check-node-reachability.sh.
Sanitization: Input validation using regular expressions is applied to the cluster name, region, and node ID parameters in scripts/triage-cluster.sh.

hyperpod-node-debugger