feat(agent): add learn action flow and grounding guards

agent_v0/agent_v1/main.py

@@ -121,10 +121,7 @@ class AgentV1:
         # Wiring ChatWindow → Executor pour Plan B (pause_message → bulle interactive)
         # Permet à l'executor d'afficher une bulle paused dans la fenêtre Léa V1
         # quand le serveur signale replay_paused=True via /replay/next.
-        try:
-            self._executor._chat_window_ref = self._chat_window
-        except Exception:
-            logger.debug("Wiring chat_window→executor échoué (non bloquant)", exc_info=True)
+        self._wire_chat_window_to_executor()
 
         # Boucles permanentes (pas besoin de session active)
         self.running = True
@@ -154,6 +151,15 @@ class AgentV1:
             shared_state=self._state,
         )
 
+    def _wire_chat_window_to_executor(self) -> None:
+        """Relie l'executor courant à la ChatWindow pour les pauses supervisees."""
+        if self._executor is None or self._chat_window is None:
+            return
+        try:
+            self._executor._chat_window_ref = self._chat_window
+        except Exception:
+            logger.debug("Wiring chat_window->executor echoue (non bloquant)", exc_info=True)
+
     def _delayed_cleanup(self):
         """Nettoyage en arrière-plan après 30s pour ne pas bloquer le démarrage."""
         time.sleep(30)
@@ -224,6 +230,7 @@ class AgentV1:
 
         # Initialiser l'executeur partage
         self._executor = ActionExecutorV1()
+        self._wire_chat_window_to_executor()
 
         self.shot_counter = 0
         self.running = True