torbjorn/my-pal-mcp-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixed contamination in consensus https://github.com/BeehiveInnovations/zen-mcp-server/issues/162

Browse Source 
Fixed broken test
This commit is contained in:
Fahad
2025-08-08 10:04:40 +05:00
parent 2bbb858d93
commit 2fa2d5a408
2 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
simulator_tests/test_o3_model_selection.py
View File

@@ -154,7 +154,10 @@ def multiply(x, y):
openai_model_usage = len(openai_model_logs) >= 1 # Should see at least 1 model usage log openai_model_usage = len(openai_model_logs) >= 1 # Should see at least 1 model usage log
openai_responses_received = len(openai_response_logs) >= 1 # Should see at least 1 response openai_responses_received = len(openai_response_logs) >= 1 # Should see at least 1 response
some_chat_calls_to_openai = len(chat_openai_logs) >= 1 # Should see at least 1 chat call some_chat_calls_to_openai = len(chat_openai_logs) >= 1 # Should see at least 1 chat call
some_workflow_calls_to_openai = len(codereview_openai_logs) >= 1 or len([line for line in logs.split("\n") if "openai" in line and "codereview" in line]) > 0 # Should see evidence of workflow tool usage some_workflow_calls_to_openai = (
len(codereview_openai_logs) >= 1
or len([line for line in logs.split("\n") if "openai" in line and "codereview" in line]) > 0
) # Should see evidence of workflow tool usage
self.logger.info(f" OpenAI API call logs: {len(openai_api_logs)}") self.logger.info(f" OpenAI API call logs: {len(openai_api_logs)}")
self.logger.info(f" OpenAI model usage logs: {len(openai_model_logs)}") self.logger.info(f" OpenAI model usage logs: {len(openai_model_logs)}")
@@ -179,7 +182,10 @@ def multiply(x, y):
("OpenAI model usage logged", openai_model_usage), ("OpenAI model usage logged", openai_model_usage),
("OpenAI responses received", openai_responses_received), ("OpenAI responses received", openai_responses_received),
("Chat tool used OpenAI", some_chat_calls_to_openai), ("Chat tool used OpenAI", some_chat_calls_to_openai),
("Workflow tool attempted", some_workflow_calls_to_openai or response3 is not None), # More flexible check (
"Workflow tool attempted",
some_workflow_calls_to_openai or response3 is not None,
), # More flexible check
] ]
passed_criteria = sum(1 for _, passed in success_criteria if passed) passed_criteria = sum(1 for _, passed in success_criteria if passed)

4
tools/consensus.py
View File

@@ -537,11 +537,13 @@ of the evidence, even when it strongly points in one direction.""",
provider = self.get_model_provider(model_name) provider = self.get_model_provider(model_name)
# Prepare the prompt with any relevant files # Prepare the prompt with any relevant files
# Use continuation_id=None for blinded consensus - each model should only see
# original prompt + files, not conversation history or other model responses
prompt = self.initial_prompt prompt = self.initial_prompt
if request.relevant_files: if request.relevant_files:
file_content, _ = self._prepare_file_content_for_prompt( file_content, _ = self._prepare_file_content_for_prompt(
request.relevant_files, request.relevant_files,
request.continuation_id, None, # Use None instead of request.continuation_id for blinded consensus
"Context files", "Context files",
) )
if file_content: if file_content: