Calendar Agent - SemanticTest

Overview

This example demonstrates end-to-end testing of an AI calendar agent that can:

Check user availability
Create calendar events
Handle conflicts
Send notifications

Complete Test Suite

test.json

{
  "name": "Calendar Agent Tests",
  "version": "1.0.0",
  "context": {
    "USER_ID": "user123",
    "TEST_DATE": "2024-01-20"
  },
  "tests": [
    {
      "id": "schedule-meeting",
      "name": "Schedule Team Meeting",
      "pipeline": [
        {
          "id": "simulate-ai-request",
          "block": "MockData",
          "config": {
            "data": {
              "userMessage": "Schedule a team meeting for tomorrow at 2pm with john@example.com",
              "aiResponse": {
                "message": "I'll check availability and schedule the meeting for you. Let me look at your calendar for tomorrow at 2 PM.",
                "toolCalls": [
                  {
                    "name": "check_user_conflicts",
                    "args": {
                      "userId": "${USER_ID}",
                      "date": "${TEST_DATE}",
                      "startTime": "14:00",
                      "endTime": "15:00"
                    }
                  },
                  {
                    "name": "create_event",
                    "args": {
                      "userId": "${USER_ID}",
                      "summary": "Team Meeting",
                      "start": "${TEST_DATE}T14:00:00",
                      "end": "${TEST_DATE}T15:00:00",
                      "attendees": ["john@example.com"],
                      "location": "Conference Room A"
                    }
                  },
                  {
                    "name": "send_notification",
                    "args": {
                      "to": ["john@example.com"],
                      "subject": "Meeting Invitation: Team Meeting",
                      "type": "calendar_invite"
                    }
                  }
                ]
              }
            }
          },
          "output": "agentData"
        },
        {
          "id": "validate-tools",
          "block": "ValidateTools",
          "input": {
            "from": "agentData.aiResponse.toolCalls",
            "as": "toolCalls"
          },
          "config": {
            "expected": ["check_user_conflicts", "create_event", "send_notification"],
            "order": true
          },
          "output": "toolValidation"
        },
        {
          "id": "validate-response",
          "block": "LLMJudge",
          "input": {
            "text": "${agentData.aiResponse.message}",
            "toolCalls": "${agentData.aiResponse.toolCalls}",
            "expected": {
              "expectedBehavior": "The AI should: 1) Check for scheduling conflicts first, 2) Create the calendar event with correct details (tomorrow at 2 PM, attendee john@example.com), 3) Send notifications to attendees. The response should be friendly and confirm the action."
            }
          },
          "config": {
            "model": "gpt-4o-mini",
            "criteria": {
              "accuracy": 0.4,
              "completeness": 0.4,
              "relevance": 0.2
            }
          },
          "output": "judgement"
        }
      ],
      "assertions": {
        "toolValidation.passed": true,
        "toolValidation.score": { "gte": 1.0 },
        "judgement.score": { "gte": 0.8 }
      }
    },
    {
      "id": "conflict-handling",
      "name": "Handle Scheduling Conflict",
      "pipeline": [
        {
          "id": "simulate-conflict",
          "block": "MockData",
          "config": {
            "data": {
              "userMessage": "Schedule a meeting tomorrow at 2pm",
              "aiResponse": {
                "message": "I found that you already have a meeting scheduled from 2:00-3:00 PM tomorrow. Would you like me to suggest alternative times or would you prefer to reschedule your existing meeting?",
                "toolCalls": [
                  {
                    "name": "check_user_conflicts",
                    "args": {
                      "userId": "${USER_ID}",
                      "date": "${TEST_DATE}",
                      "startTime": "14:00",
                      "endTime": "15:00"
                    }
                  }
                ]
              }
            }
          },
          "output": "conflictData"
        },
        {
          "id": "validate-conflict-handling",
          "block": "LLMJudge",
          "input": {
            "text": "${conflictData.aiResponse.message}",
            "toolCalls": "${conflictData.aiResponse.toolCalls}",
            "expected": {
              "expectedBehavior": "The AI should check for conflicts first, detect the scheduling conflict, inform the user about the existing meeting, and offer alternatives or ask for next steps. It should NOT create a conflicting event."
            }
          },
          "config": {
            "model": "gpt-4o-mini"
          },
          "output": "conflictJudgement"
        },
        {
          "id": "verify-no-double-booking",
          "block": "ValidateTools",
          "input": {
            "from": "conflictData.aiResponse.toolCalls",
            "as": "toolCalls"
          },
          "config": {
            "forbidden": ["create_event"],
            "expected": ["check_user_conflicts"]
          },
          "output": "toolCheck"
        }
      ],
      "assertions": {
        "conflictJudgement.score": { "gte": 0.8 },
        "toolCheck.passed": true
      }
    },
    {
      "id": "multi-attendee",
      "name": "Schedule with Multiple Attendees",
      "pipeline": [
        {
          "id": "simulate-multi-attendee",
          "block": "MockData",
          "config": {
            "data": {
              "aiResponse": {
                "toolCalls": [
                  {
                    "name": "check_user_conflicts",
                    "args": {
                      "userId": "${USER_ID}",
                      "date": "2024-01-21",
                      "startTime": "10:00",
                      "endTime": "11:00"
                    }
                  },
                  {
                    "name": "create_event",
                    "args": {
                      "userId": "${USER_ID}",
                      "summary": "Project Planning",
                      "start": "2024-01-21T10:00:00",
                      "end": "2024-01-21T11:00:00",
                      "attendees": [
                        "alice@example.com",
                        "bob@example.com",
                        "charlie@example.com"
                      ]
                    }
                  }
                ]
              }
            }
          },
          "output": "multiData"
        },
        {
          "id": "validate-attendees",
          "block": "ValidateTools",
          "input": {
            "from": "multiData.aiResponse.toolCalls",
            "as": "toolCalls"
          },
          "config": {
            "validateArgs": {
              "create_event": {
                "userId": "${USER_ID}"
              }
            }
          },
          "output": "attendeeValidation"
        }
      ],
      "assertions": {
        "attendeeValidation.passed": true
      }
    }
  ]
}

Setup

Create a .env file with your OpenAI API key:

.env

OPENAI_API_KEY=sk-proj-your-key-here

Running the Tests

npx semtest test.json

Expected output:

✅ Calendar Agent Tests
  ✅ schedule-meeting: Schedule Team Meeting (1.5s)
     ✅ toolValidation.passed = true
     ✅ toolValidation.score = 1.0
     ✅ judgement.score = 0.89

  ✅ conflict-handling: Handle Scheduling Conflict (1.2s)
     ✅ conflictJudgement.score = 0.91
     ✅ toolCheck.passed = true

  ✅ multi-attendee: Schedule with Multiple Attendees (0.8s)
     ✅ attendeeValidation.passed = true

3 tests passed, 0 failed (3.5s total)

Test Breakdown

Test 1: Basic Scheduling

Validates the happy path where the agent successfully schedules a meeting:

Check conflicts - Agent must check availability first
Create event - Agent creates the meeting with correct details
Send notifications - Agent notifies attendees

{
  "expected": ["check_user_conflicts", "create_event", "send_notification"],
  "order": true  // Tools must be called in this order
}

Test 2: Conflict Handling

Validates the agent handles conflicts gracefully:

Detects existing meetings
Informs user about conflicts
Does NOT create conflicting events
Offers alternatives

{
  "config": {
    "forbidden": ["create_event"],  // Should NOT create event
    "expected": ["check_user_conflicts"]
  }
}

Test 3: Multiple Attendees

Ensures the agent correctly handles group meetings:

Validates attendee list is preserved
Checks userId is correct
Verifies all required fields

Key Patterns

Tool Order Validation

{
  "block": "ValidateTools",
  "config": {
    "expected": ["check_conflicts", "create_event"],
    "order": true  // Enforce execution order
  }
}

Forbidden Actions

{
  "config": {
    "forbidden": ["delete_event", "cancel_all"]
  }
}

Argument Validation

{
  "config": {
    "validateArgs": {
      "create_event": {
        "userId": "user123"
      }
    }
  }
}

Integration with Real Calendar API

Replace MockData with actual API calls:

{
  "pipeline": [
    {
      "block": "HttpRequest",
      "input": {
        "url": "${CALENDAR_API}/ai/chat",
        "method": "POST",
        "headers": {
          "Authorization": "Bearer ${env.API_KEY}"
        },
        "body": {
          "message": "Schedule a meeting tomorrow at 2pm",
          "userId": "${USER_ID}"
        }
      },
      "output": "apiResponse"
    },
    {
      "block": "JsonParser",
      "input": "${apiResponse.body}",
      "output": { "parsed": "agentData" }
    },
    {
      "block": "ValidateTools",
      "input": {
        "from": "agentData.toolCalls",
        "as": "toolCalls"
      },
      "config": {
        "expected": ["check_user_conflicts", "create_event"]
      }
    }
  ]
}

Advanced Scenarios

Testing Retry Logic

{
  "pipeline": [
    {
      "id": "attempt-schedule",
      "block": "HttpRequest",
      "input": {
        "url": "${API_URL}/schedule"
      },
      "output": "result"
    },
    {
      "block": "Loop",
      "config": {
        "target": "attempt-schedule",
        "maxIterations": 3,
        "condition": {
          "path": "result.status",
          "operator": "notEquals",
          "value": 200
        }
      }
    }
  ]
}

Testing Time Zones

{
  "tests": [
    {
      "id": "timezone-handling",
      "pipeline": [
        {
          "block": "MockData",
          "config": {
            "data": {
              "toolCalls": [{
                "name": "create_event",
                "args": {
                  "start": "2024-01-20T14:00:00-08:00",  // PST
                  "end": "2024-01-20T15:00:00-08:00"
                }
              }]
            }
          }
        }
      ]
    }
  ]
}

Next Steps

Tool Validation

Learn advanced tool validation

Multi-turn Conversations

Test conversational AI agents

Error Handling

Handle errors gracefully

ValidateTools Block

Complete tool validation docs

​Overview

​Complete Test Suite

​Setup

​Running the Tests

​Test Breakdown

​Test 1: Basic Scheduling

​Test 2: Conflict Handling

​Test 3: Multiple Attendees

​Key Patterns

​Tool Order Validation

​Forbidden Actions

​Argument Validation

​Integration with Real Calendar API

​Advanced Scenarios

​Testing Retry Logic

​Testing Time Zones

​Next Steps