# yaml-language-server: $schema=https://promptfoo.dev/config-schema.json
# Strands Agents SDK Example - Weather Assistant Evaluation
#
# This configuration evaluates a Strands agent that provides weather information
# and temperature conversions. It demonstrates:
# - Using a Python provider (file://) to integrate custom agents
#   https://promptfoo.dev/docs/providers/python/
# - Testing tool usage with contains-any assertions
#   https://promptfoo.dev/docs/configuration/expected-outputs/#contains
# - Using llm-rubric for semantic evaluation of responses
#   https://promptfoo.dev/docs/configuration/expected-outputs/model-graded/
# - Performance assertions (latency, cost)
#   https://promptfoo.dev/docs/configuration/expected-outputs/#latency
#
# Strands Agents SDK by AWS: https://github.com/strands-agents/sdk-python
# Strands Documentation: https://strandsagents.com/

description: Strands Agents SDK evaluation

# The prompt template - {{query}} is replaced with test case variables
prompts:
  - '{{query}}'

# Provider configuration pointing to our Python agent
# The file:// prefix tells promptfoo to load a Python module
# Format: file://<script>:<function>
providers:
  - id: 'file://agent_provider.py:call_api'
    label: 'Strands Weather Agent'
    config:
      # Model ID passed to the Strands agent (can be changed to gpt-4o, etc.)
      model_id: 'gpt-4o-mini'

# Default assertions applied to all test cases
# https://promptfoo.dev/docs/configuration/guide/#default-test-cases
defaultTest:
  assert:
    # Performance: Response should complete within 30 seconds
    - type: latency
      threshold: 30000

# Test cases to evaluate the agent's responses
tests:
  # Test 1: Basic weather query - verifies the get_weather tool works
  - description: 'Weather query for New York'
    vars:
      query: "What's the weather like in New York?"
    assert:
      # Check that response contains expected weather data
      - type: contains-any
        value: ['New York', 'Sunny', '72']
      # Use LLM to verify response quality
      - type: llm-rubric
        value: 'Response should include weather information for New York City'

  # Test 2: Another city to verify tool handles different inputs
  - description: 'Weather query for London'
    vars:
      query: 'Tell me the current weather in London'
    assert:
      - type: contains-any
        value: ['London', 'Cloudy', '58']
      # Verify response mentions temperature (custom JavaScript assertion)
      # https://promptfoo.dev/docs/configuration/expected-outputs/#javascript
      - type: javascript
        value: 'output.includes("°F") || output.includes("degrees")'

  # Test 3: Test with icontains (case-insensitive)
  - description: 'Weather query for Tokyo'
    vars:
      query: "What's the weather in Tokyo?"
    assert:
      # Case-insensitive check
      - type: icontains
        value: 'tokyo'
      - type: icontains
        value: 'clear'
      # Verify response is not too short (at least 20 chars)
      - type: javascript
        value: 'output.length >= 20'

  # Test 4: Multi-tool usage - tests both get_weather AND convert_temperature
  # This verifies the agent can chain multiple tools together
  - description: 'Weather with temperature conversion'
    vars:
      query: "What's the weather in New York? Also tell me the temperature in Celsius."
    assert:
      - type: llm-rubric
        value: 'Response should include New York weather (72°F) AND convert it to Celsius (approximately 22°C)'
      # Verify both temperature units appear in response
      - type: javascript
        value: '(output.includes("°F") || output.includes("Fahrenheit")) && (output.includes("°C") || output.includes("Celsius"))'

  # Test 5: Unknown city - verify graceful handling
  - description: 'Weather for unknown city'
    vars:
      query: "What's the weather in Atlantis?"
    assert:
      # Should still return a response (default weather)
      - type: icontains
        value: 'atlantis'
      # Should not error out
      - type: not-contains
        value: 'Error'