prompt-evaluation

name	prompt-evaluation
description	Design, test, and improve prompts, agents, tool-use policies, model routing rules, and evaluation suites for internal AI products.

Prompt And Agent Evaluation

Purpose

Improve prompt and agent behavior with testable criteria instead of subjective impressions.

Workflow

1. Define task scope, expected outputs, failure modes, and users.
2. Create representative examples and adversarial cases.
3. Specify scoring rubrics: correctness, citation quality, formatting, safety, latency, and tool use.
4. Compare models, prompts, and routing rules on the same set.
5. Record versioned results and recommended changes.

Output

Return:

• Evaluation dataset outline.
• Scoring rubric.
• Test cases.
• Recommended prompt/model changes.
• Regression checks.