The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides reference guides (references/environment-capture.md) that contain standard shell commands (e.g., lscpu, sysctl, nvidia-smi, free, lsblk) for capturing hardware and software context. These commands are necessary for documenting the environment to ensure benchmark reproducibility.
[COMMAND_EXECUTION]: The isolation strategy guide (references/test-case-design.md) suggests using the cpupower utility, which may require sudo privileges, to fix the CPU governor for variance control. This is a standard and documented procedure for minimizing environmental noise in high-rigor benchmarks.
[SAFE]: The skill incorporates statistical analysis templates using standard Python libraries and emphasizes best practices such as warmup iterations, representative workloads, and reporting variance. The overall design focuses on methodological soundness and transparency.

benchmark-runner