The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/hyperpod_issue_report.py script downloads a diagnostic utility (eks-log-collector.sh) from the official awslabs GitHub repository. This download is secured using a pinned commit hash and a mandatory SHA256 integrity check within the generated node script.
[REMOTE_CODE_EXECUTION]: The skill performs remote execution by creating a temporary bash script and running it on HyperPod cluster nodes via AWS Systems Manager (SSM) Session Manager. This is the primary mechanism for collecting diagnostic logs from individual compute nodes.
[COMMAND_EXECUTION]: The bundled Python script executes local shell commands using subprocess.run to interact with kubectl and the aws CLI. These operations are used to gather cluster-wide metadata (e.g., node capacity, pod status, cluster events). The script uses list-based arguments and shlex.quote for variable interpolation to mitigate command injection risks.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from external sources including AWS SageMaker/EKS APIs and kubectl command outputs in scripts/hyperpod_issue_report.py.
Boundary markers: While data is processed, the skill primarily acts as a transport mechanism, uploading raw logs to S3 rather than feeding untrusted log content directly back into the agent's reasoning loop.
Capability inventory: The skill possesses capabilities for local command execution (aws, kubectl), remote script execution (via SSM), and network writes (S3 uploads).
Sanitization: The script employs shlex.quote when building shell commands that include user-supplied or cluster-provided identifiers.

hyperpod-issue-report