feat: script command routing + LLM slot acquisition in base_agent

- Add _handle_script_command() to bypass LLM and call script skill directly
- Add 'script' to _handle_system_command routing
- Add _llm_slot_acquire() / _llm_slot_release() MQTT semaphore for Ollama coordination

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

agents_core/base_agent.py

@@ -559,12 +559,37 @@ class BaseAgent(ABC):
             all_agents = [a.agent_id for a in self.registry.all_agents()]
             return f"Agents en ligne : {online}\nAgents connus : {all_agents}"
 
+        if cmd == "script":
+            return self._handle_script_command(args, source_msg)
+
         if cmd == "update":
             return self._do_self_update(source_msg)
 
         # Commandes custom de l'agent
         return self.handle_custom_command(cmd, args, source_msg)
 
+    # ──────────────────────────────────────────────
+    # Commande script directe (bypass LLM)
+    # ──────────────────────────────────────────────
+
+    def _handle_script_command(self, args: str, source_msg: Optional[Message] = None) -> str:
+        """
+        Exécute une commande script sans passer par le LLM.
+        args : 'exec <nom> [args]' | 'list' | 'show <nom>' | 'run | <contenu>'
+        """
+        context = AgentContext(self)
+        result  = self.skills.run("script", args, context)
+        if result is None:
+            result = f"[{self.agent_id}] Skill 'script' non disponible."
+
+        # Renvoi MQTT si la commande vient d'un autre agent
+        if source_msg and source_msg.reply_to:
+            try:
+                self.mqtt.reply(source_msg, result)
+            except Exception as e:
+                logger.debug(f"[Script] Impossible d'envoyer la réponse : {e}")
+        return result
+
     # ──────────────────────────────────────────────
     # Mise à jour self
     # ──────────────────────────────────────────────