version: "1.0"
description: "Research agent JSON format instructions"
Each template starts with a blank line (preserving the leading \n from
the original Python triple-quoted strings) so that appending the template
to a user prompt produces clean spacing.
round_1_format: |
Respond with valid JSON only — no markdown, no explanation:
{
"summary": "Concise summary of your findings (2-3 paragraphs).",
"key_points": ["Key insight 1", "Key insight 2", "Key insight 3"],
"perspective": "Your unique perspective on this topic.",
"confidence": 0.8
}
review_format: |
Respond with valid JSON only:
{
"questions": [
"What specific aspect needs deeper investigation?",
"How does X relate to Y based on other agents' findings?"
],
"target_agent_ids": ["agent-id-or-null", "agent-id-or-null"]
}
round_2_format: |
Respond with valid JSON only:
{
"summary": "Refined summary of your deeper research.",
"key_points": ["Refined point 1", "Refined point 2"],
"perspective": "Your evolved perspective after reviewing shared knowledge.",
"confidence": 0.85
}
report_format: |
Respond with valid JSON only:
{
"title": "Your Individual Research Report Title",
"perspective_summary": "High-level summary of your perspective.",
"key_insights": ["Insight 1", "Insight 2"],
"analysis": "Detailed analysis text (2-4 paragraphs).",
"metaphors_or_analogies": ["Compelling analogy 1"],
"open_questions": ["Unresolved question 1"],
"full_text": "Complete report text.",
"sections": [
{
"heading": "Section Title",
"source_agent_id": null,
"content": "Section content here.",
"subsections": []
}
]
}
clarify_format: |
Respond with valid JSON only:
{
"response": "Clear, concise answer to the clarification question."
}