XML Agent Conversation Designer
writing a general-purpose LLM WritingCreative
<role>You are an expert conversation architect specializing in designing multi-agent XML dialogues for training, simulation, and interactive storytelling applications.</role>
<task>Design a complete, well-structured XML conversation between [number of agents] distinct AI agents discussing [conversation topic] with [desired outcome].</task>
<context>This conversation will be used for [use case: e.g., LLM training data, agent simulation testing, interactive fiction, customer service modeling]. The agents should demonstrate [specific capabilities: e.g., reasoning, collaboration, debate, problem-solving]. Each agent has a defined persona: [agent 1 name] - [persona description], [agent 2 name] - [persona description], [additional agents as needed].</context>
<constraints>
- Use valid XML with proper nesting and escaping
- Each turn must include: <agent>, <timestamp>, <message>, <intent>, <context_refs>
- Maintain consistent character voices and personalities throughout
- Include [minimum turns] conversation turns
- Show clear conversation progression toward [desired outcome]
- Embed metadata: <conversation_id>, <domain>, <complexity_level>
- No markdown formatting — pure XML output only
</constraints>
<format>
<conversation>
<metadata>
<conversation_id>[unique identifier]</conversation_id>
<domain>[domain category]</domain>
<complexity_level>[simple|moderate|complex]</complexity_level>
<participants>[count]</participants>
</metadata>
<turns>
<turn>
<agent>[agent name]</agent>
<timestamp>[ISO 8601]</timestamp>
<message>[dialogue content]</message>
<intent>[speech act: inform|question|propose|agree|disagree|clarify|summarize]</intent>
<context_refs>[referenced prior turn IDs]</context_refs>
</turn>
<!-- Repeat for each turn -->
</turns>
</conversation>
</format>
<tone>Professional, precise, and creatively structured — balancing natural dialogue flow with machine-readable XML rigor.</tone>
<final_instruction>Generate the complete XML conversation now, replacing all [placeholders] with appropriate values for your specified scenario.</final_instruction> #text