fix: improve auto-edge reliability — error visibility, similarity-based strength, heuristic docs

- Surface auto-edge errors in store response (auto_edge_error key) instead
  of silently swallowing exceptions, making failures diagnosable
- Pass similarity score as edge strength so higher-confidence matches get
  stronger edges instead of always defaulting to 0.8
- Document why same-type pairs and REQUIRES/FOLLOWS are excluded from the
  heuristic table vs edge-proposer.py

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

mcp_server.py

@@ -22,6 +22,9 @@ SYNC_SCRIPT = Path(__file__).parent / "scripts" / "memory-git-sync.sh"
 
 # Auto-edge heuristics: (new_type, match_type) -> (rel_type, direction)
 # direction "ab" = new->match, "ba" = match->new
+# Subset of edge-proposer.py's TYPE_HEURISTICS — excludes:
+#   - Same-type pairs (decision/decision, fix/fix, solution/solution): too noisy for auto-edges
+#   - REQUIRES/FOLLOWS pairs: need stronger signal than title recall provides
 AUTO_EDGE_HEURISTICS = {
     ("fix", "problem"): ("SOLVES", "ab"),
     ("solution", "problem"): ("SOLVES", "ab"),
@@ -555,7 +558,8 @@ def _auto_create_edges(
 
     Uses recall to find similar memories and heuristic type-pairs to choose
     relationship types. Returns list of created edge info dicts.
-    Never raises — failures are silently swallowed so store always succeeds.
+    Never raises — returns partial results or empty list with error info so
+    store always succeeds. Returns (edges_list, error_string_or_None).
     """
     try:
         # Build recall query from title + tags for better precision (#4)
@@ -613,11 +617,15 @@ def _auto_create_edges(
             from_title = title if from_id == memory_id else match_title
             to_title = match_title if to_id == match_id else title
             desc = f"Auto-edge: {from_title} → {to_title}"
+            # Use similarity as edge strength when available (default 0.8)
+            sim = result.get("similarity")
+            strength = round(sim, 2) if sim is not None else 0.8
             edge_id = client.relate(
                 from_id=from_id,
                 to_id=to_id,
                 rel_type=rel_type,
                 description=desc,
+                strength=strength,
             )
 
             if edge_id:  # Empty string means duplicate
@@ -630,9 +638,9 @@ def _auto_create_edges(
                     }
                 )
 
-        return created_edges
+        return created_edges, None
-    except Exception:
+    except Exception as e:
-        return []
+        return [], f"{type(e).__name__}: {e}"
 
 
 def handle_tool_call(tool_name: str, arguments: Dict[str, Any]) -> Dict[str, Any]:
@@ -669,16 +677,19 @@ def handle_tool_call(tool_name: str, arguments: Dict[str, Any]) -> Dict[str, Any
                     tags=tags,
                 )
                 episode_logged = True
-            auto_edges = _auto_create_edges(client, memory_id, title, mem_type, tags)
+            auto_edges, auto_edge_error = _auto_create_edges(
-            _trigger_git_sync()
+                client, memory_id, title, mem_type, tags
-            return ok(
-                {
-                    "success": True,
-                    "memory_id": memory_id,
-                    "episode_logged": episode_logged,
-                    "auto_edges": auto_edges,
-                }
             )
+            _trigger_git_sync()
+            result = {
+                "success": True,
+                "memory_id": memory_id,
+                "episode_logged": episode_logged,
+                "auto_edges": auto_edges,
+            }
+            if auto_edge_error:
+                result["auto_edge_error"] = auto_edge_error
+            return ok(result)
 
         elif tool_name == "memory_recall":
             results = client.recall(