Agent Evaluation

Use this skill when the work is deciding how an AI agent should be measured, not when the work is simply building the feature itself.

Read references/grader-selection.md when you need help picking grader types, benchmark families, or score dimensions for a specific agent surface.

Read references/ops-and-calibration.md when you need harness design, transcript review, CI gates, sampling policy, saturation checks, or production monitoring guidance.

When to use this skill