fix(realtime): handle non-code files and filter spurious events

codebase_rag/constants.py

@@ -848,11 +848,19 @@ class TreeSitterModule(StrEnum):
 class EventType(StrEnum):
     MODIFIED = "modified"
     CREATED = "created"
+    DELETED = "deleted"
 
 
 CYPHER_DELETE_MODULE = "MATCH (m:Module {path: $path})-[*0..]->(c) DETACH DELETE m, c"
+CYPHER_DELETE_FILE = "MATCH (f:File {path: $path}) DETACH DELETE f"
+CYPHER_DELETE_FOLDER = "MATCH (f:Folder {path: $path}) DETACH DELETE f"
 CYPHER_DELETE_CALLS = "MATCH ()-[r:CALLS]->() DELETE r"
 
+# (H) Queries for orphan pruning — returns all paths stored in the graph
+CYPHER_ALL_FILE_PATHS = "MATCH (f:File) RETURN f.path AS path"
+CYPHER_ALL_MODULE_PATHS = "MATCH (m:Module) RETURN m.path AS path"
+CYPHER_ALL_FOLDER_PATHS = "MATCH (f:Folder) RETURN f.path AS path"
+
 REALTIME_LOGGER_FORMAT = (
     "<green>{time:YYYY-MM-DD HH:mm:ss.SSS}</green> | "
     "<level>{level: <8}</level> | "

codebase_rag/graph_updater.py

@@ -327,6 +327,8 @@ def run(self, force: bool = False) -> None:
         logger.info(ls.ANALYSIS_COMPLETE)
         self.ingestor.flush_all()
 
+        self._prune_orphan_nodes()
+
         self._generate_semantic_embeddings()
 
     def remove_file_from_state(self, file_path: Path) -> None:
@@ -439,6 +441,13 @@ def _process_files(self, force: bool = False) -> None:
             for deleted_key in deleted_keys:
                 deleted_path = self.repo_path / deleted_key
                 self.remove_file_from_state(deleted_path)
+                if isinstance(self.ingestor, QueryProtocol):
+                    self.ingestor.execute_write(
+                        cs.CYPHER_DELETE_MODULE, {cs.KEY_PATH: deleted_key}
+                    )
+                    self.ingestor.execute_write(
+                        cs.CYPHER_DELETE_FILE, {cs.KEY_PATH: deleted_key}
+                    )
 
         if skipped_count > 0:
             logger.info(ls.INCREMENTAL_SKIPPED, count=skipped_count)
@@ -475,6 +484,43 @@ def _process_function_calls(self) -> None:
                 file_path, root_node, language, self.queries
             )
 
+    def _prune_orphan_nodes(self) -> None:
+        """Remove graph nodes whose files/folders no longer exist on disk."""
+        if not isinstance(self.ingestor, QueryProtocol):
+            return
+
+        logger.info(ls.PRUNE_START)
+        total_pruned = 0
+
+        prune_specs: list[tuple[str, str, str]] = [
+            (cs.CYPHER_ALL_FILE_PATHS, cs.CYPHER_DELETE_FILE, "File"),
+            (cs.CYPHER_ALL_MODULE_PATHS, cs.CYPHER_DELETE_MODULE, "Module"),
+            (cs.CYPHER_ALL_FOLDER_PATHS, cs.CYPHER_DELETE_FOLDER, "Folder"),
+        ]
+
+        for query_all, delete_query, label in prune_specs:
+            rows = self.ingestor.fetch_all(query_all)
+            orphans = [
+                r["path"]
+                for r in rows
+                if r.get("path")
+                and not (self.repo_path / r["path"]).exists()
+            ]
+
+            if orphans:
+                logger.info(ls.PRUNE_FOUND, count=len(orphans), label=label)
+                for orphan_path in orphans:
+                    logger.debug(ls.PRUNE_DELETING, label=label, path=orphan_path)
+                    self.ingestor.execute_write(
+                        delete_query, {cs.KEY_PATH: orphan_path}
+                    )
+                total_pruned += len(orphans)
+
+        if total_pruned:
+            logger.info(ls.PRUNE_COMPLETE, count=total_pruned)
+        else:
+            logger.info(ls.PRUNE_SKIP)
+
     def _generate_semantic_embeddings(self) -> None:
         if not has_semantic_dependencies():
             logger.info(ls.SEMANTIC_NOT_AVAILABLE)

codebase_rag/logs.py

@@ -653,6 +653,13 @@
 INCREMENTAL_CHANGED = "Re-indexing {count} changed files"
 INCREMENTAL_DELETED = "Removed state for {count} deleted files"
 INCREMENTAL_FORCE = "Force mode enabled, bypassing hash cache"
+
+# (H) Orphan pruning logs
+PRUNE_START = "--- Pruning orphan nodes from graph ---"
+PRUNE_FOUND = "Found {count} orphan {label} nodes to remove"
+PRUNE_DELETING = "Pruning orphan {label}: {path}"
+PRUNE_COMPLETE = "Pruning complete. Removed {count} orphan nodes."
+PRUNE_SKIP = "No orphan nodes found. Graph is clean."
 FILE_HASH_UNCHANGED = "File unchanged (hash match): {path}"
 FILE_HASH_CHANGED = "File changed (hash mismatch): {path}"
 FILE_HASH_NEW = "New file detected: {path}"