The Agent Skills Directory

[SAFE]: The skill serves as a technical reference and operational toolkit for infrastructure management. It includes utility scripts, spot_math.py and disk_read_throughput.sh, which are written in plain Python and Bash respectively, performing transparent calculations and hardware benchmarking without external dependencies.
[SAFE]: Implements and recommends robust security configurations, specifically instructing users to bind model servers to loopback addresses, use mandatory API keys, and establish secure SSH tunnels for remote access to private nodes.
[SAFE]: Demonstrates a strong security posture by detailing least-privilege service account configurations and forced-command SSH keys for automated lifecycle components, minimizing the potential impact of credential compromise.
[COMMAND_EXECUTION]: The skill provides templates and instructions for executing standard cloud provider CLI commands and systemd service operations. These actions are fundamental to the skill's primary purpose of managing GPU instance lifecycles and are documented with clear operational rationale.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface related to the ingestion of agent logs and benchmark results. However, its focus is on infrastructure-level durability, salvage, and publication rather than data interpolation, and it provides specific guidelines for keeping raw logs separate from judged results.

spot-gpu-evaluation