Default output: return only the result, blockers, and required evidence. Omit preambles, process narration, repeated context, confidence scores, and follow-up offers. Use at most five bullets unless a required artifact or schema needs more.

Eval Harness

Overview

A systematic framework for evaluating agent performance. Measures accuracy, efficiency, and reliability across defined test scenarios. Enables data-driven decisions about agent quality and improvement.

When to Use

• Before deploying agent changes to production
• Comparing different agent configurations
• Identifying weaknesses in agent behavior
• Tracking agent quality over time
• Validating prompt improvements

Evaluation Dimensions

1. Accuracy

Does the agent produce correct outputs?