fix(gateway): implement platform-aware PID termination

gateway/run.py

@@ -7582,7 +7582,7 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
     # setups (each profile using a distinct HERMES_HOME) will naturally
     # allow concurrent instances without tripping this guard.
     import time as _time
-    from gateway.status import get_running_pid, remove_pid_file
+    from gateway.status import get_running_pid, remove_pid_file, terminate_pid
     existing_pid = get_running_pid()
     if existing_pid is not None and existing_pid != os.getpid():
         if replace:
@@ -7591,10 +7591,10 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
                 existing_pid,
             )
             try:
-                os.kill(existing_pid, signal.SIGTERM)
+                terminate_pid(existing_pid, force=False)
             except ProcessLookupError:
                 pass  # Already gone
-            except PermissionError:
+            except (PermissionError, OSError):
                 logger.error(
                     "Permission denied killing PID %d. Cannot replace.",
                     existing_pid,
@@ -7614,9 +7614,9 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
                     existing_pid,
                 )
                 try:
-                    os.kill(existing_pid, signal.SIGKILL)
+                    terminate_pid(existing_pid, force=True)
                     _time.sleep(0.5)
-                except (ProcessLookupError, PermissionError):
+                except (ProcessLookupError, PermissionError, OSError):
                     pass
             remove_pid_file()
             # Also release all scoped locks left by the old process.

gateway/status.py

@@ -14,6 +14,8 @@ concurrently under distinct configurations).
 import hashlib
 import json
 import os
+import signal
+import subprocess
 import sys
 from datetime import datetime, timezone
 from pathlib import Path
@@ -23,6 +25,7 @@ from typing import Any, Optional
 _GATEWAY_KIND = "hermes-gateway"
 _RUNTIME_STATUS_FILE = "gateway_state.json"
 _LOCKS_DIRNAME = "gateway-locks"
+_IS_WINDOWS = sys.platform == "win32"
 
 
 def _get_pid_path() -> Path:
@@ -49,6 +52,33 @@ def _utc_now_iso() -> str:
     return datetime.now(timezone.utc).isoformat()
 
 
+def terminate_pid(pid: int, *, force: bool = False) -> None:
+    """Terminate a PID with platform-appropriate force semantics.
+
+    POSIX uses SIGTERM/SIGKILL. Windows uses taskkill /T /F for true force-kill
+    because os.kill(..., SIGTERM) is not equivalent to a tree-killing hard stop.
+    """
+    if force and _IS_WINDOWS:
+        try:
+            result = subprocess.run(
+                ["taskkill", "/PID", str(pid), "/T", "/F"],
+                capture_output=True,
+                text=True,
+                timeout=10,
+            )
+        except FileNotFoundError:
+            os.kill(pid, signal.SIGTERM)
+            return
+
+        if result.returncode != 0:
+            details = (result.stderr or result.stdout or "").strip()
+            raise OSError(details or f"taskkill failed for PID {pid}")
+        return
+
+    sig = signal.SIGTERM if not force else getattr(signal, "SIGKILL", signal.SIGTERM)
+    os.kill(pid, sig)
+
+
 def _scope_hash(identity: str) -> str:
     return hashlib.sha256(identity.encode("utf-8")).hexdigest()[:16]