| name | chatgpt-voice-agent |
| description | Creates prompts for ChatGPT voice agents with personality, tone, and state machine. Use when building voice agents, defining conversation flows, or OpenAI Realtime API. Trigger phrases: "语音代理", "voice agent", "ChatGPT语音", "state machine" |
ChatGPT Voice Agent Prompt
Overview
Creates LLM prompts for high-quality voice agents. Follows OpenAI Realtime Agents format with Personality/Tone and Conversation State Machine.
Output Format
Personality and Tone
- Identity, Task, Demeanor, Tone
- Level of Enthusiasm, Formality, Emotion
- Filler Words (none/occasionally/often/very often)
- Pacing, Other details
Instructions
- Follow Conversation States
- Repeat back name/phone for confirmation
- Acknowledge corrections straightforwardly
Conversation States
State machine with: id, description, instructions, examples, transitions (next_step, condition)
Workflow
- Step 1 (optional): Ask clarifying questions for unspecified Personality/Tone qualities. Offer 3 options per quality.
- Step 2: Output full prompt. State machine based ONLY on explicit user steps. No code block around state_machine.
State Machine Schema
{
"id": "unique_step_id",
"description": "step purpose",
"instructions": ["what agent does"],
"examples": ["example utterances"],
"transitions": [{"next_step": "id", "condition": "when"}]
}
Usage Example
User input: "创建一个客服语音代理，需要收集姓名和验证信息"
AI action: Asks personality questions if needed, outputs full prompt with state machine (greeting → get first name → get last name → verify → next steps)
Expected result: Complete voice agent prompt ready for ChatGPT/Realtime API
