evaluation

This skill focuses on building robust evaluation frameworks specifically designed for agent systems. Unlike traditional software, agents are dynamic, non-deterministic, and often lack single correct answers. This skill provides methods to evaluate agent performance, validate context engineering choices, measure improvements, and catch regressions before deployment. It supports building quality gates for agent pipelines, comparing different agent configurations, and continuously evaluating production systems. The core concept is to judge agents on achieving right outcomes while following reasonable processes, accounting for multiple valid paths.

What It Does

Builds evaluation frameworks for agent systems, incorporating multi-dimensional rubrics, LLM-as-judge methodologies, and human evaluation to ensure quality and continuous improvement.

When To Use

Use this skill when testing agent performance, validating context engineering, measuring improvements, catching regressions, comparing configurations, and evaluating production systems.

Installation

Copy SKILL.md to your skills directory