test-agent

Test Voice Agent

Before generating any code:

1. Fetch current Simulate Conversation API docs to verify endpoint and parameters
2. Check the knowledge file at .claude/elevenlabs/knowledge/general-elevenlabs-knowledge.md for API patterns
3. If fetch fails, fall back to static knowledge base with staleness warning.

Steps:

1. Detect the agent ID:

   • Check .env or .env.local for ELEVENLABS_AGENT_ID
   • Check project code for agent ID references
   • If not found, ask the user via AskUserQuestion

2. Ask the testing mode via AskUserQuestion:

   • Single conversation — Provide a prompt, run one simulated conversation, show transcript
   • Test suite — Define multiple scenarios, run them all, report pass/fail

3. Single conversation mode:

   • Ask the user for a test prompt (e.g., "I'd like to book an appointment for next Tuesday")
   • Call the ElevenLabs Simulate Conversation API with the agent ID and prompt
   • Display results:
     • Full transcript (user turns + agent turns)
     • Tools invoked by the agent and their results
     • Latency metrics per turn
     • Any errors or unexpected behaviors

4. Test suite mode:

   • Generate a test file with scenario definitions:

     # test_agent.py
     scenarios = [
         {
             "name": "Book appointment",
             "prompt": "I'd like to book an appointment for next Tuesday",
             "expect_tool": "schedule_appointment",
         },
         {
             "name": "Transfer to human",
             "prompt": "I need to speak to a real person",
             "expect_tool": "forward_call",
         },
     ]
     

   • Ask the user to customize scenarios for their agent's capabilities
   • Run all scenarios and report results in a structured table:

     Scenario	Status	Tools Invoked	Latency

   • Highlight any failures or unexpected tool invocations

5. Offer to save test results and set up as a repeatable test command