agent-eval-design

Use when designing evaluations for AI agents, skills, routers, prompts, tool-use policies, or multi-step workflows: task sets, rubrics, graders, hard negatives, regression cases, traces, and acceptance thresholds. Do NOT use for application test planning (use `testing-strategy`), skill-library health tooling (use `skill-infrastructure`), or live debugging of a failed run (use `debugging`).