What problem does it solve? Agent systems are non-deterministic and can take multiple valid paths to a goal, making traditional pass/fail testing ineffective. This Skill provides rubrics, test set design, and monitoring patterns to systematically measure agent quality, catch regressions, and validate context engineering choices. ## Core Features & Use Cases - Multi-Dimensional Rubrics: Score agent outputs across factual accuracy, completeness, citation accuracy, source quality, and tool efficiency with weighted overall scores. - Evaluation Runner and Test Sets: Build complexity-stratified test sets and run automated evaluation pipelines with pass/fail thresholds and dimension-level summaries. - Production Monitoring: Sample live interactions, track pass rates over time, and trigger alerts when quality drops below warning or critical thresholds. - Use Case: Before deploying a new agent configuration, run it against a stratified test set, compare dimension scores against the baseline, and block the release if the overall score falls below 0.7. ## Quick Start Use the evaluation skill to build a rubric-based test harness that scores my agent's responses on accuracy and completeness.
