Improved consensus to treat a step properly as both a request + response, and initial step includes Claude's assessment.
Improved prompt to not request for code when it's a general business decision
This commit is contained in:
Fahad
@@ -48,8 +48,9 @@ CONSENSUS_WORKFLOW_FIELD_DESCRIPTIONS = {
|
||||
"steps 2+ are for processing individual model responses."
|
||||
),
|
||||
"total_steps": (
|
||||
"Total number of steps needed. This equals 1 (your analysis) + number of models to consult + "
|
||||
"1 (final synthesis)."
|
||||
"Total number of steps needed. This equals the number of models to consult. "
|
||||
"Step 1 includes your analysis + first model consultation on return of the call. Final step includes "
|
||||
"last model consultation + synthesis."
|
||||
),
|
||||
"next_step_required": ("Set to true if more models need to be consulted. False when ready for final synthesis."),
|
||||
"findings": (
|
||||
@@ -182,7 +183,7 @@ class ConsensusTool(WorkflowTool):
|
||||
"IMPORTANT: This workflow enforces sequential model consultation:\\n"
|
||||
"- Step 1 is always your independent analysis\\n"
|
||||
"- Each subsequent step processes one model response\\n"
|
||||
"- Total steps = 1 (your analysis) + number of models + 1 (synthesis)\\n"
|
||||
"- Total steps = number of models (each step includes consultation + response)\\n"
|
||||
"- Models can have stances (for/against/neutral) for structured debate\\n"
|
||||
"- Same model can be used multiple times with different stances\\n"
|
||||
"- Each model + stance combination must be unique\\n\\n"
|
||||
@@ -435,15 +436,16 @@ of the evidence, even when it strongly points in one direction.""",
|
||||
self.initial_prompt = request.step
|
||||
self.models_to_consult = request.models or []
|
||||
self.accumulated_responses = []
|
||||
# Set total steps: 1 (Claude) + len(models) + 1 (synthesis)
|
||||
request.total_steps = 1 + len(self.models_to_consult) + 1
|
||||
# Set total steps: len(models) (each step includes consultation + response)
|
||||
request.total_steps = len(self.models_to_consult)
|
||||
|
||||
# For all steps (1 through total_steps), consult the corresponding model
|
||||
if request.step_number <= request.total_steps:
|
||||
# Calculate which model to consult for this step
|
||||
model_idx = request.step_number - 1 # 0-based index
|
||||
|
||||
# If this is a model consultation step (2 through total_steps-1)
|
||||
elif request.step_number > 1 and request.step_number < request.total_steps:
|
||||
# Get the current model to consult
|
||||
model_idx = request.current_model_index or 0
|
||||
if model_idx < len(self.models_to_consult):
|
||||
# Consult the model
|
||||
# Consult the model for this step
|
||||
model_response = await self._consult_model(self.models_to_consult[model_idx], request)
|
||||
|
||||
# Add to accumulated responses
|
||||
@@ -458,23 +460,48 @@ of the evidence, even when it strongly points in one direction.""",
|
||||
"model_stance": model_response.get("stance", "neutral"),
|
||||
"model_response": model_response,
|
||||
"current_model_index": model_idx + 1,
|
||||
"next_step_required": request.step_number < request.total_steps - 1,
|
||||
"next_step_required": request.step_number < request.total_steps,
|
||||
}
|
||||
|
||||
if request.step_number < request.total_steps - 1:
|
||||
# Add Claude's analysis to step 1
|
||||
if request.step_number == 1:
|
||||
response_data["claude_analysis"] = {
|
||||
"initial_analysis": request.step,
|
||||
"findings": request.findings,
|
||||
}
|
||||
response_data["status"] = "analysis_and_first_model_consulted"
|
||||
|
||||
# Check if this is the final step
|
||||
if request.step_number == request.total_steps:
|
||||
response_data["status"] = "consensus_workflow_complete"
|
||||
response_data["consensus_complete"] = True
|
||||
response_data["complete_consensus"] = {
|
||||
"initial_prompt": self.initial_prompt,
|
||||
"models_consulted": [
|
||||
f"{m['model']}:{m.get('stance', 'neutral')}" for m in self.accumulated_responses
|
||||
],
|
||||
"total_responses": len(self.accumulated_responses),
|
||||
"consensus_confidence": "high",
|
||||
}
|
||||
response_data["next_steps"] = (
|
||||
"CONSENSUS GATHERING IS COMPLETE. Synthesize all perspectives and present:\n"
|
||||
"1. Key points of AGREEMENT across models\n"
|
||||
"2. Key points of DISAGREEMENT and why they differ\n"
|
||||
"3. Your final consolidated recommendation\n"
|
||||
"4. Specific, actionable next steps for implementation\n"
|
||||
"5. Critical risks or concerns that must be addressed"
|
||||
)
|
||||
else:
|
||||
response_data["next_steps"] = (
|
||||
f"Model {model_response['model']} has provided its {model_response.get('stance', 'neutral')} "
|
||||
f"perspective. Please analyze this response and call {self.get_name()} again with:\n"
|
||||
f"- step_number: {request.step_number + 1}\n"
|
||||
f"- findings: Summarize key points from this model's response\n"
|
||||
f"- current_model_index: {model_idx + 1}\n"
|
||||
f"- model_responses: (append this response to the list)"
|
||||
)
|
||||
else:
|
||||
response_data["next_steps"] = (
|
||||
"All models have been consulted. For the final step, synthesize all perspectives."
|
||||
f"- findings: Summarize key points from this model's response"
|
||||
)
|
||||
|
||||
# Add accumulated responses for tracking
|
||||
response_data["accumulated_responses"] = self.accumulated_responses
|
||||
|
||||
return [TextContent(type="text", text=json.dumps(response_data, indent=2))]
|
||||
|
||||
# Otherwise, use standard workflow execution
|
||||
@@ -520,6 +547,7 @@ of the evidence, even when it strongly points in one direction.""",
|
||||
"verdict": response.content,
|
||||
"metadata": {
|
||||
"provider": provider.get_provider_type().value,
|
||||
"model_name": model_name,
|
||||
},
|
||||
}
|
||||
|
||||
@@ -627,8 +655,90 @@ of the evidence, even when it strongly points in one direction.""",
|
||||
else:
|
||||
response_data["consensus_workflow_status"] = "ready_for_synthesis"
|
||||
|
||||
# Customize metadata for consensus workflow
|
||||
self._customize_consensus_metadata(response_data, request)
|
||||
|
||||
return response_data
|
||||
|
||||
def _customize_consensus_metadata(self, response_data: dict, request) -> None:
|
||||
"""
|
||||
Customize metadata for consensus workflow to accurately reflect multi-model nature.
|
||||
|
||||
The default workflow metadata shows the model running Claude's analysis steps,
|
||||
but consensus is a multi-model tool that consults different models. We need
|
||||
to provide accurate metadata that reflects this.
|
||||
"""
|
||||
if "metadata" not in response_data:
|
||||
response_data["metadata"] = {}
|
||||
|
||||
metadata = response_data["metadata"]
|
||||
|
||||
# Always preserve tool_name
|
||||
metadata["tool_name"] = self.get_name()
|
||||
|
||||
if request.step_number == request.total_steps:
|
||||
# Final step - show comprehensive consensus metadata
|
||||
models_consulted = []
|
||||
if self.models_to_consult:
|
||||
models_consulted = [f"{m['model']}:{m.get('stance', 'neutral')}" for m in self.models_to_consult]
|
||||
|
||||
metadata.update(
|
||||
{
|
||||
"workflow_type": "multi_model_consensus",
|
||||
"models_consulted": models_consulted,
|
||||
"consensus_complete": True,
|
||||
"total_models": len(self.models_to_consult) if self.models_to_consult else 0,
|
||||
}
|
||||
)
|
||||
|
||||
# Remove the misleading single model metadata
|
||||
metadata.pop("model_used", None)
|
||||
metadata.pop("provider_used", None)
|
||||
|
||||
else:
|
||||
# Intermediate steps - show consensus workflow in progress
|
||||
models_to_consult = []
|
||||
if self.models_to_consult:
|
||||
models_to_consult = [f"{m['model']}:{m.get('stance', 'neutral')}" for m in self.models_to_consult]
|
||||
|
||||
metadata.update(
|
||||
{
|
||||
"workflow_type": "multi_model_consensus",
|
||||
"models_to_consult": models_to_consult,
|
||||
"consultation_step": request.step_number,
|
||||
"total_consultation_steps": request.total_steps,
|
||||
}
|
||||
)
|
||||
|
||||
# Remove the misleading single model metadata that shows Claude's execution model
|
||||
# instead of the models being consulted
|
||||
metadata.pop("model_used", None)
|
||||
metadata.pop("provider_used", None)
|
||||
|
||||
def _add_workflow_metadata(self, response_data: dict, arguments: dict[str, Any]) -> None:
|
||||
"""
|
||||
Override workflow metadata addition for consensus tool.
|
||||
|
||||
The consensus tool doesn't use single model metadata because it's a multi-model
|
||||
workflow. Instead, we provide consensus-specific metadata that accurately
|
||||
reflects the models being consulted.
|
||||
"""
|
||||
# Initialize metadata if not present
|
||||
if "metadata" not in response_data:
|
||||
response_data["metadata"] = {}
|
||||
|
||||
# Add basic tool metadata
|
||||
response_data["metadata"]["tool_name"] = self.get_name()
|
||||
|
||||
# The consensus-specific metadata is already added by _customize_consensus_metadata
|
||||
# which is called from customize_workflow_response. We don't add the standard
|
||||
# single-model metadata (model_used, provider_used) because it's misleading
|
||||
# for a multi-model consensus workflow.
|
||||
|
||||
logger.debug(
|
||||
f"[CONSENSUS_METADATA] {self.get_name()}: Using consensus-specific metadata instead of single-model metadata"
|
||||
)
|
||||
|
||||
def store_initial_issue(self, step_description: str):
|
||||
"""Store initial prompt for model consultations."""
|
||||
self.initial_prompt = step_description
|
||||
|
||||
Reference in New Issue
Block a user