hyperpod-issue-report
Pass
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/hyperpod_issue_report.pyallows for the execution of arbitrary shell commands on remote HyperPod nodes via an SSM session when the user provides the--commandargument. This is an intended feature for diagnostic flexibility during troubleshooting sessions. - [EXTERNAL_DOWNLOADS]: For EKS clusters, the skill fetches the
eks-log-collector.shscript from the officialawslabsGitHub repository. The script implements a robust security control by verifying the downloaded file's SHA256 checksum against a hardcoded value (61c8940e9391330a9c67d8cd6720de3de3a1a90021546835f7f54f1fed2efb3f) before execution. - [DATA_EXFILTRATION]: The tool is designed to collect extensive diagnostic data, including cluster logs from
/var/log/aws/clusters/, Slurm configuration files from/opt/slurm/etc/, and Kubernetes resource metadata (e.g., node descriptions, pod lists). All collected data is bundled and uploaded to an S3 bucket specified by the user. - [PROMPT_INJECTION]: The skill includes a post-collection phase where the agent is invited to "Help analyze collected diagnostics." This creates a surface for indirect prompt injection (Category 8) where malicious instructions could be embedded in the collected log files or configuration metadata. Ingestion point: analysis of log files in
SKILL.md. Boundary markers: absent. Capability inventory: script can execute shell commands and call AWS APIs. Sanitization: absent.
Audit Metadata