multi-assertion.yaml

# Multi-Assertion — Multiple checks on a single agent response
# Demonstrates how to validate different aspects of one response.

name: "Multi-Assertion"
description: "Apply multiple assertions to verify response quality"

agent:
  adapter: openai
  model: gpt-4o-mini
  system: "You are a math tutor. Always show your work step by step."

steps:
  - send: "What is 15% of 200?"
    assert:
      # Must contain the correct answer
      - type: contains
        value: "30"
      # Should show working/steps
      - type: regex
        pattern: "(step|first|multiply|percent)"
      # Should not be excessively long
      - type: max-tokens
        value: 500
      # Should not contain wrong answers
      - type: not-contains
        value: "25"
      # LLM-as-judge: is the explanation clear?
      - type: llm-judge
        criteria: "The response clearly explains how to calculate 15% of 200"
        threshold: 0.8