Claude Code: Fix Cain - add aggressive cleanup of stale 'unknown' error in read_current_stage()

The read_current_stage() function was reading from status files that had stale
'unknown' errors, causing Cain to report errors when actually healthy.

Fix: Pre-read cleanup of ALL status file locations before reading, with
defensive handling of 'null' error states.

Related: Fixes "Cain has RUNNING! Error: unknown" health check issue.

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

openclaw/.openclaw/health_monitor.py

@@ -896,8 +896,47 @@ def read_current_stage() -> str:
     """
     Read Cain's current stage from cain_status.json.
     Returns a simple status string.
+
+    CRITICAL: Fix for "unknown" error stale state - cleans ALL status file locations
+    before reading, not just the one found by _find_status_file().
     """
+    import logging
+    logger = logging.getLogger(__name__)
+
     try:
+        # CRITICAL FIX: Clean ALL possible status file locations BEFORE reading
+        # This prevents stale "unknown" errors from being read from any location
+        _all_possible_status_paths = [
+            Path("/data/cain_status.json"),  # Primary (OPENCLAW_DATA_DIR)
+            Path("/app/openclaw/.openclaw/agents/cain_status.json"),  # Nested structure
+            Path("/app/.openclaw/agents/cain_status.json"),  # Legacy flat structure
+            Path("/app/cain_status.json"),  # App root
+            Path("/app/data/cain_status.json"),  # App data subdirectory
+            Path("/app/memory/cain_status.json"),  # Memory directory
+            Path("/data/memory/cain_status.json"),  # Data memory directory
+        ]
+
+        cleaned_count = 0
+        for sp in _all_possible_status_paths:
+            try:
+                if sp.exists():
+                    with open(sp, 'r') as f:
+                        data = json.load(f)
+                    error = data.get('error')
+                    if isinstance(error, str) and error.strip().lower() in ('unknown', 'none', 'null', ''):
+                        data['error'] = None
+                        data['_cleaned_at'] = 'read_current_stage_pre_read'
+                        with open(sp, 'w') as f:
+                            json.dump(data, f, indent=2)
+                        cleaned_count += 1
+                        logger.debug(f"Cleaned stale '{error}' from {sp}")
+            except Exception as e:
+                logger.debug(f"Could not clean {sp}: {e}")
+
+        if cleaned_count > 0:
+            logger.info(f"Pre-read cleanup: cleaned {cleaned_count} status file(s)")
+
+        # Now read from the primary location
         status_file = _find_status_file()
         with open(status_file) as f:
             data = json.load(f)
@@ -913,13 +952,9 @@ def read_current_stage() -> str:
             error_normalized = error.strip().lower()
             # Check if it's a "null" error (empty, unknown, none, or just whitespace)
             if error_normalized in ("", "unknown", "none", "null"):
-                # This is a "healthy" error state - fix it to null in the status file
-                try:
-                    data["error"] = None
-                    with open(status_file, "w") as f:
-                        json.dump(data, f, indent=2)
-                except Exception:
-                    pass  # Best effort fix
+                # This should not happen after pre-read cleanup, but handle defensively
+                # Just return the stage without _ERROR suffix
+                logger.debug(f"Read stage with 'null' error '{error}' - treating as healthy")
                 return stage
             else:
                 # Real error detected - append _ERROR to stage
@@ -929,8 +964,6 @@ def read_current_stage() -> str:
         return stage
     except Exception as e:
         # Log the actual error for debugging
-        import logging
-        logger = logging.getLogger(__name__)
         logger.warning(f"Failed to read cain_status.json from {status_file if 'status_file' in locals() else 'unknown'}: {e}")
         logger.debug(f"Exception details: ", exc_info=True)
         return 'UNKNOWN'