Use eval-view when the job is regression-gating an existing agent workflow in CI, not when a user just wants a general observability product or benchmark library. The workflow is bounded: capture a baseline, rerun the same agent scenarios, diff outputs and tool traces, then classify whether the change is safe, flaky, or a release blocker. That scope boundary, behavior-regression review for multi-turn tool-calling agents, gives it a clear skill shape instead of leaving it as a generic eval toolkit listing.