The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional content, best practices, and documentation for agent evaluation. It does not provide any scripts, tools, or automated behaviors.
[EXTERNAL_DOWNLOADS]: The 'Quick start' section references well-known, industry-standard benchmarking packages (swebench, webarena) to be installed via pip. These are established tools in the AI development community.
[COMMAND_EXECUTION]: The skill provides example command-line instructions and CI/CD configuration snippets (such as pytest and python -m swebench.harness.run_evaluation) as part of its educational content. These are meant to be executed by a developer in a controlled environment and are not triggered autonomously by the skill.

agent-benchmarking