diff --git a/agent_service/agents/master_agent.py b/agent_service/agents/master_agent.py
index 8af8eec..eba946d 100644
--- a/agent_service/agents/master_agent.py
+++ b/agent_service/agents/master_agent.py
@@ -147,23 +147,37 @@ class MasterAgent:
         msgs = [{'role': 'system', 'content': system}, *history,
                 {'role': 'user', 'content': message}]
         resp = await self._llm.submit(msgs, caller='master')
+        raw = (resp.content or '').strip()
+        # Strip markdown fences like ```json ... ```
+        if raw.startswith('```'):
+            raw = raw.strip('`')
+            if raw.lower().startswith('json'):
+                raw = raw[4:]
+            raw = raw.strip()
+        # Pull out the first {...} block if there's surrounding prose.
+        first = raw.find('{')
+        last = raw.rfind('}')
+        if first != -1 and last != -1 and last > first:
+            raw = raw[first:last + 1]
         try:
-            raw = resp.content.strip()
-            if raw.startswith(chr(96)*3):
-                raw = raw.split(chr(10), 1)[1].rsplit(chr(10), 1)[0]
             data = json.loads(raw)
+            if not isinstance(data, dict):
+                raise ValueError(f'expected JSON object, got {type(data).__name__}')
             return IntentResult(
-                needs_clarification=data.get('needs_clarification', False),
+                needs_clarification=bool(data.get('needs_clarification', False)),
                 clarification_question=data.get('clarification_question'),
-                is_continuation=data.get('is_continuation', False),
-                agents=data.get('agents', []),
-                intent_summary=data.get('intent_summary', ''),
-                params=data.get('params', {}),
-                context_hints=data.get('context_hints', []))
-        except (json.JSONDecodeError, KeyError) as exc:
-            logger.warning('Intent classification parse failed: %s', exc)
-            return IntentResult(needs_clarification=True,
-                                clarification_question='Could you clarify what you need?')
+                is_continuation=bool(data.get('is_continuation', False)),
+                agents=data.get('agents') or [],
+                intent_summary=data.get('intent_summary', '') or '',
+                params=data.get('params') or {},
+                context_hints=data.get('context_hints') or [])
+        except Exception as exc:
+            logger.warning('Intent classification parse failed (%s); raw=%r',
+                           exc, (resp.content or '')[:300])
+            # Treat unparseable LLM output as 'no specialist agent applies' so
+            # the direct-answer fallback can handle it instead of looping on
+            # clarification.
+            return IntentResult(needs_clarification=False, agents=[])
 
     async def _check_access(self, user_id, agents) -> AccessResult:
         denied = []