Use PinchBench when the job is to benchmark OpenClaw coding agents on repeatable real-world tasks before changing models, prompts, or harness behavior, not when a user simply wants a generic leaderboard or agent framework. The operator workflow is crisp: run the benchmark suite, score task outcomes, compare results, and decide whether a candidate model or change is ready. That scope boundary, OpenClaw agent benchmarking before rollout, makes this a clear skill instead of a plain benchmark product listing.