honcho/tests/unified/test_cases/agentic_fde_custom_dialecti...

68 lines
2.1 KiB
JSON

{
"description": "Test that custom dialectic rules modify chat response behavior",
"steps": [
{
"step_type": "create_session",
"session_id": "dialectic_rules_test",
"peer_configs": {
"user": {
"observe_me": true,
"observe_others": false
},
"assistant": {
"observe_me": false,
"observe_others": true
}
}
},
{
"step_type": "add_messages",
"session_id": "dialectic_rules_test",
"messages": [
{
"peer_id": "user",
"content": "I started a new job at Google last week as a senior engineer."
},
{
"peer_id": "assistant",
"content": "Congratulations on the new role! That's exciting."
},
{
"peer_id": "user",
"content": "Thanks! I'm working on their cloud infrastructure team."
}
]
},
{
"step_type": "wait",
"target": "queue_empty",
"timeout": 120,
"flush": true
},
{
"step_type": "set_agent_config",
"description": "Configure dialectic to respond in bullet points only",
"dialectic_rules": "CRITICAL: Always format your response as exactly 3 bullet points using '•' characters. Never use prose paragraphs. Be extremely concise - each bullet should be under 10 words."
},
{
"step_type": "query",
"description": "Test that response follows bullet point format",
"target": "chat",
"session_id": "dialectic_rules_test",
"observer_peer_id": "assistant",
"observed_peer_id": "user",
"input": "What do you know about this user's job?",
"assertions": [
{
"assertion_type": "llm_judge",
"prompt": "Does this response mention the user's job at Google or on a cloud infrastructure team? It should reference their employment."
},
{
"assertion_type": "llm_judge",
"prompt": "Is this response formatted primarily as bullet points (using • or - or * characters) rather than flowing prose paragraphs? The response should have a list-like structure."
}
]
}
]
}