The Agent Skills Directory

[SAFE]: The skill is designed with a strictly read-only constraint, explicitly prohibiting state-mutating commands. Diagnostics are performed via bundled scripts using authenticated AWS CLI calls, ensuring no unauthorized changes are made to the cluster.
[SAFE]: The slurm-diagnose.sh script implements rigorous validation for all parameters (region, cluster name, node names, instance IDs) using specific regular expression patterns before they are used in shell commands, mitigating local injection risks.
[SAFE]: Interaction with cluster nodes via AWS Systems Manager (SSM) utilizes a secure wrapper that Base64-encodes the diagnostic payload and uses jq @sh to safely quote environment variables, preventing remote command injection on the target instances.
[SAFE]: The script includes a sanitization helper that strips ANSI escape sequences and control characters from server-derived data before outputting to the user's terminal, protecting against terminal-based injection attacks.
[SAFE]: The skill follows the principle of least privilege by relying on standard AWS permissions and providing authoritative documentation links for remediation instead of executing autonomous repairs.

hyperpod-slurm-debugger