refactor: add history records as separate input

Author: dori

src/mcp_as_a_judge/constants.py

@@ -13,6 +13,5 @@
 
 # Database Configuration
 DATABASE_URL = "sqlite://:memory:"
-MAX_CONTEXT_RECORDS = 20
-CONTEXT_ENRICHMENT_COUNT = 10
+MAX_SESSION_RECORDS = 20  # Maximum records to keep per session (FIFO)
 RECORD_RETENTION_DAYS = 1

src/mcp_as_a_judge/db/conversation_history_service.py

@@ -52,7 +52,7 @@ async def load_context_for_enrichment(
         # Load recent conversations for this session
         recent_records = await self.db.get_session_conversations(
             session_id=session_id,
-            limit=self.config.database.context_enrichment_count,  # load last X records
+            limit=self.config.database.max_session_records,  # load last X records (same as save limit)
         )
 
         logger.info(f"📚 Retrieved {len(recent_records)} conversation records from DB")
@@ -109,33 +109,35 @@ async def get_conversation_history(
 
         return context_records
 
-    def format_conversation_history_as_context(
+    def format_conversation_history_as_json_array(
         self, conversation_history: list[ConversationRecord]
-    ) -> str:
+    ) -> list[dict]:
         """
-        Convert conversation history list to formatted string for context field.
+        Convert conversation history list to JSON array for prompt injection.
 
         Args:
             conversation_history: List of conversation records
 
         Returns:
-            Formatted string representation of conversation history
+            List of dictionaries with conversation history data including timestamps
         """
         if not conversation_history:
-            return ""
+            return []
 
         logger.info(
-            f"📝 Formatting {len(conversation_history)} conversation records as context string"
+            f"📝 Formatting {len(conversation_history)} conversation records as JSON array"
         )
 
-        context_parts = []
+        json_array = []
         for record in conversation_history:
-            context_parts.append(f"Tool: {record.source}")
-            context_parts.append(f"Input: {record.input}")
-            context_parts.append(f"Output: {record.output}")
-            context_parts.append("")  # Empty line between records
-
-        formatted_context = "\n".join(context_parts).strip()
-        logger.info(f"📝 Generated context string: {len(formatted_context)} characters")
-
-        return formatted_context
+            json_array.append(
+                {
+                    "source": record.source,
+                    "input": record.input,
+                    "output": record.output,
+                    "timestamp": record.timestamp.isoformat()
+                }
+            )
+
+        logger.info(f"📝 Generated JSON array with {len(json_array)} records")
+        return json_array

src/mcp_as_a_judge/db/db_config.py

@@ -6,9 +6,8 @@
 """
 
 from mcp_as_a_judge.constants import (
-    CONTEXT_ENRICHMENT_COUNT,
     DATABASE_URL,
-    MAX_CONTEXT_RECORDS,
+    MAX_SESSION_RECORDS,
     RECORD_RETENTION_DAYS,
 )
 
@@ -61,8 +60,7 @@ class DatabaseConfig:
 
     def __init__(self) -> None:
         self.url = DATABASE_URL
-        self.max_context_records = MAX_CONTEXT_RECORDS
-        self.context_enrichment_count = CONTEXT_ENRICHMENT_COUNT
+        self.max_session_records = MAX_SESSION_RECORDS
         self.record_retention_days = RECORD_RETENTION_DAYS
 
 

src/mcp_as_a_judge/db/factory.py

@@ -54,14 +54,14 @@ def create_provider(cls, config: Config) -> ConversationHistoryDB:
         if provider_name in ["in_memory", "sqlite"]:
             # SQLite-based providers (both SQLModel and legacy)
             return provider_class(
-                max_context_records=config.database.max_context_records,
+                max_session_records=config.database.max_session_records,
                 url=config.database.url,
             )
         else:
             # For future network database providers (PostgreSQL, MySQL, etc.)
             return provider_class(
                 url=config.database.url,
-                max_context_records=config.database.max_context_records,
+                max_session_records=config.database.max_session_records,
             )
 
     @classmethod

src/mcp_as_a_judge/db/interface.py

@@ -29,20 +29,6 @@ class ConversationRecord(SQLModel, table=True):
 class ConversationHistoryDB(ABC):
     """Abstract interface for conversation history database operations."""
 
-    @abstractmethod
-    def __init__(
-        self, max_context_records: int = 20, retention_days: int = 1, url: str = ""
-    ) -> None:
-        """
-        Initialize the database provider.
-
-        Args:
-            max_context_records: Maximum number of conversation records to keep per session
-            retention_days: Number of days to keep conversation records before deletion
-            url: Database connection URL
-        """
-        pass
-
     @abstractmethod
     async def save_conversation(
         self, session_id: str, source: str, input_data: str, output: str

src/mcp_as_a_judge/db/providers/sqlite_provider.py

@@ -34,7 +34,7 @@ class SQLiteProvider(ConversationHistoryDB):
     - Session-based conversation retrieval
     """
 
-    def __init__(self, max_context_records: int = 20, url: str = "") -> None:
+    def __init__(self, max_session_records: int = 20, url: str = "") -> None:
         """Initialize the SQLModel SQLite database with LRU and time-based cleanup."""
         # Parse URL to get SQLite connection string
         connection_string = self._parse_sqlite_url(url)
@@ -48,7 +48,7 @@ def __init__(self, max_context_records: int = 20, url: str = "") -> None:
             else {},
         )
 
-        self._max_context_records = max_context_records
+        self._max_session_records = max_session_records
 
         # Initialize cleanup service for time-based cleanup
         self._cleanup_service = ConversationCleanupService(engine=self.engine)
@@ -58,7 +58,7 @@ def __init__(self, max_context_records: int = 20, url: str = "") -> None:
 
         logger.info(
             f"🗄️ SQLModel SQLite provider initialized: {connection_string}, "
-            f"max_records={max_context_records}, retention_days={self._cleanup_service.retention_days}"
+            f"max_records={max_session_records}, retention_days={self._cleanup_service.retention_days}"
         )
 
     def _parse_sqlite_url(self, url: str) -> str:
@@ -88,8 +88,8 @@ def _cleanup_old_records(self) -> int:
 
     def _cleanup_old_messages(self, session_id: str) -> int:
         """
-        Remove old messages from a session using LRU (Least Recently Used) strategy.
-        Keeps only the most recent max_context_records messages per session.
+        Remove old messages from a session using FIFO strategy.
+        Keeps only the most recent max_session_records messages per session.
         """
         with Session(self.engine) as session:
             # Count current messages in session
@@ -100,16 +100,16 @@ def _cleanup_old_messages(self, session_id: str) -> int:
             current_count = len(current_records)
 
             logger.info(
-                f"🧹 LRU cleanup check for session {session_id}: {current_count} records "
-                f"(max: {self._max_context_records})"
+                f"🧹 FIFO cleanup check for session {session_id}: {current_count} records "
+                f"(max: {self._max_session_records})"
             )
 
-            if current_count <= self._max_context_records:
+            if current_count <= self._max_session_records:
                 logger.info("   No cleanup needed - within limits")
                 return 0
 
-            # Get oldest records to remove
-            records_to_remove = current_count - self._max_context_records
+            # Get oldest records to remove (FIFO)
+            records_to_remove = current_count - self._max_session_records
             oldest_stmt = (
                 select(ConversationRecord)
                 .where(ConversationRecord.session_id == session_id)

src/mcp_as_a_judge/logging_config.py

@@ -131,8 +131,7 @@ def log_startup_message(config: Any) -> None:
         "🚀 MCP Judge server starting with conversation history logging enabled"
     )
     logger.info(
-        f"📊 Configuration: max_context_records={config.database.max_context_records}, "
-        f"context_enrichment_count={config.database.context_enrichment_count}"
+        f"📊 Configuration: max_session_records={config.database.max_session_records}"
     )
 
 

src/mcp_as_a_judge/models.py

@@ -118,14 +118,18 @@ class JudgeCodingPlanUserVars(BaseModel):
     user_requirements: str = Field(
         description="The user's requirements for the coding task"
     )
-    context: str = Field(description="Context including conversation history")
+    context: str = Field(description="Additional context about the task")
     plan: str = Field(description="The coding plan to be evaluated")
     design: str = Field(description="The design documentation")
     research: str = Field(description="Research findings and analysis")
     research_urls: list[str] = Field(
         default_factory=list,
         description="URLs from MANDATORY online research - minimum 3 URLs required",
     )
+    conversation_history: list = Field(
+        default_factory=list,
+        description="Previous conversation history as JSON array with timestamps",
+    )
 
 
 class JudgeCodeChangeSystemVars(BaseModel):
@@ -145,7 +149,11 @@ class JudgeCodeChangeUserVars(BaseModel):
     file_path: str = Field(description="Path to the file being changed")
     change_description: str = Field(description="Description of what the change does")
     code_change: str = Field(description="The actual code content being reviewed")
-    context: str = Field(description="Context including conversation history")
+    context: str = Field(description="Additional context about the code change")
+    conversation_history: list = Field(
+        default_factory=list,
+        description="Previous conversation history as JSON array with timestamps",
+    )
 
 
 class ResearchValidationSystemVars(BaseModel):
@@ -167,7 +175,11 @@ class ResearchValidationUserVars(BaseModel):
         default_factory=list,
         description="URLs from MANDATORY online research - minimum 3 URLs required",
     )
-    context: str = Field(description="Context including conversation history")
+    context: str = Field(description="Additional context about the research validation")
+    conversation_history: list = Field(
+        default_factory=list,
+        description="Previous conversation history as JSON array with timestamps",
+    )
 
 
 class WorkflowGuidanceSystemVars(BaseModel):
@@ -182,7 +194,11 @@ class WorkflowGuidanceUserVars(BaseModel):
     """Variables for build_workflow user prompt."""
 
     task_description: str = Field(description="Description of the development task")
-    context: str = Field(description="Context including conversation history")
+    context: str = Field(description="Additional context about the task")
+    conversation_history: list = Field(
+        default_factory=list,
+        description="Previous conversation history as JSON array with timestamps",
+    )
 
 
 class ValidationErrorSystemVars(BaseModel):

src/mcp_as_a_judge/prompts/user/build_workflow.md

@@ -6,4 +6,7 @@ Please analyze this development task and provide workflow guidance:
 ## Context
 {{ context }}
 
+## Previous Conversation History as JSON array 
+{{ conversation_history }}
+
 Determine which MCP as a Judge tools should be used next and provide clear guidance.

src/mcp_as_a_judge/prompts/user/judge_code_change.md

@@ -14,6 +14,9 @@ Please review the following code:
 
 ## Code Content
 
+## Previous Conversation History as JSON array 
+{{ conversation_history }}
+
 ```
 {{ code_change }}
 ```