Agent Evaluation Harness

A ready-to-use evaluation setup so you can prove an agent still works before you ship a change.

What's included

• A golden-set template and rubric format
• A scoring sheet for success, tool-use, and cost
• A regression-tracking layout
• Guidance on grading by hand and with a second model

Best for

Anyone running an agent in production who is tired of "it seemed fine."

How to use it

1. Fill the golden set with 15-20 real cases.
2. Score your current agent as the baseline.
3. Re-run on every prompt or model change.