The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local scripts (e.g., aggregate_benchmark.py, run_eval.py, run_loop.py) to manage its workflow. These scripts involve standard shell command execution via subprocess for tasks like running evaluation loops, managing background processes (nohup), and interacting with the claude CLI. These operations are restricted to the local environment and are standard for development and benchmarking tools.
[NO_CODE]: The skill leverages existing project-specific scripts and documentation to perform its tasks. It does not introduce external, unverifiable code dependencies or remote execution patterns from untrusted sources. All package management (e.g., npm, pip) is mentioned within the context of teaching users how to use their own environments and is not executed maliciously by the skill itself.

skill-creator