hyperpod-performance-debugger

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local shell script, scripts/perf-snapshot.sh, to gather diagnostic data. This script uses the AWS CLI to interact with SageMaker, CloudWatch, and FSx services.
  • [REMOTE_CODE_EXECUTION]: The perf-snapshot.sh script performs remote execution of diagnostic commands (e.g., nvidia-smi, iostat, dmesg, df) on remote SageMaker HyperPod cluster nodes using aws ssm start-session. This functionality is essential for its diagnostic purpose and is properly scoped to the targeted cluster nodes.
  • [EXTERNAL_DOWNLOADS]: The skill references external performance testing benchmarks (nccl-tests) and technical documentation from trusted sources, specifically from the awslabs organization and official Amazon AWS documentation domains.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes information from external AWS API responses and remote node system logs.
  • Ingestion points: scripts/perf-snapshot.sh (collects data from aws sagemaker, aws cloudwatch, and output from remote SSM commands).
  • Boundary markers: None present in the data processing flow.
  • Capability inventory: Local shell execution, AWS CLI operations, and remote command execution on managed cluster instances via SSM.
  • Sanitization: The script implements strict regex-based validation for user-supplied arguments (cluster identifiers and region) and utilizes jq with the @sh filter to properly escape command payloads before they are executed on remote nodes.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 12:02 PM
Security Audit — agent-trust-hub — hyperpod-performance-debugger