Fix recall endpoint: Add search_term, input validation, and proper contexts array return

- Add search_term parameter with regex validation (alphanumeric + punctuation)
- Add tag validation to prevent SQL injection
- Change return format from {context: string} to {total, contexts: array}
- Use ConversationContextResponse schema for proper serialization
- Improves security and provides structured data for clients

Related: Context Recall System fixes (COMPLETE_SYSTEM_SUMMARY.md)

api/routers/conversation_contexts.py

@@ -76,8 +76,18 @@ def list_conversation_contexts(
     status_code=status.HTTP_200_OK,
 )
 def recall_context(
+    search_term: Optional[str] = Query(
+        None,
+        max_length=200,
+        pattern=r'^[a-zA-Z0-9\s\-_.,!?()]+$',
+        description="Full-text search term (alphanumeric, spaces, and basic punctuation only)"
+    ),
     project_id: Optional[UUID] = Query(None, description="Filter by project ID"),
-    tags: Optional[List[str]] = Query(None, description="Filter by tags (OR logic)"),
+    tags: Optional[List[str]] = Query(
+        None,
+        description="Filter by tags (OR logic)",
+        max_items=20
+    ),
     limit: int = Query(
         default=10,
         ge=1,
@@ -96,20 +106,33 @@ def recall_context(
     """
     Retrieve relevant contexts formatted for Claude prompt injection.
 
-    This endpoint returns a token-efficient markdown string ready for
+    This endpoint returns contexts matching the search criteria with
-    injection into Claude's prompt. It's the main context recall API.
+    properly formatted JSON response containing the contexts array.
 
     Query Parameters:
+    - search_term: Full-text search across title and summary (uses FULLTEXT index)
     - project_id: Filter contexts by project
     - tags: Filter contexts by tags (any match)
     - limit: Maximum number of contexts to retrieve
     - min_relevance_score: Minimum relevance score threshold
 
-    Returns a formatted string ready for prompt injection.
+    Returns JSON with contexts array and metadata.
     """
+    # Validate tags to prevent SQL injection
+    if tags:
+        import re
+        tag_pattern = re.compile(r'^[a-zA-Z0-9\-_]+$')
+        for tag in tags:
+            if not tag_pattern.match(tag):
+                raise HTTPException(
+                    status_code=status.HTTP_400_BAD_REQUEST,
+                    detail=f"Invalid tag format: '{tag}'. Tags must be alphanumeric with hyphens or underscores only."
+                )
+
     try:
-        formatted_context = conversation_context_service.get_recall_context(
+        contexts, total = conversation_context_service.get_recall_context(
             db=db,
+            search_term=search_term,
             project_id=project_id,
             tags=tags,
             limit=limit,
@@ -117,11 +140,13 @@ def recall_context(
         )
 
         return {
-            "context": formatted_context,
+            "total": total,
+            "limit": limit,
+            "search_term": search_term,
             "project_id": str(project_id) if project_id else None,
             "tags": tags,
-            "limit": limit,
+            "min_relevance_score": min_relevance_score,
-            "min_relevance_score": min_relevance_score
+            "contexts": [ConversationContextResponse.model_validate(ctx) for ctx in contexts]
         }
 
     except Exception as e: