fix(marketplace): address code review findings

- Fix str.format() crash when user input contains curly braces
- Fix Layer 2 passing str to find_best_agent (expects list[str])
- Fix AlignmentGuard fail-open on LLM audit failure (now fail-closed)
- Fix _config_reload_lock not initialized in create_app()
- Fix evolve_soul redundant reflector.reflect() call (reuse existing reflection)
- Fix test mocks using AsyncMock for sync find_best_agent method
- Remove unused _COMPLEXITY_CLASSIFY_PROMPT constant

src/agentkit/chat/skill_routing.py

@@ -186,16 +186,6 @@ _CHAT_MODE_RE = re.compile(
     re.IGNORECASE,
 )
 
-_COMPLEXITY_CLASSIFY_PROMPT = (
-    "Assess the complexity of the following user request on a scale of 0.0 to 1.0.\n"
-    "0.0 = trivial greeting / simple chat\n"
-    "0.3 = single-skill task (e.g. search, translate)\n"
-    "0.7 = multi-step or cross-domain task (e.g. market research + competitor analysis)\n"
-    "1.0 = highly complex, multi-agent collaboration needed\n\n"
-    'User request: "{content}"\n\n'
-    'Respond ONLY with a JSON object: {{"complexity": <float>}}'
-)
-
 
 class CostAwareRouter:
     """三层成本感知路由器。
@@ -251,7 +241,13 @@ class CostAwareRouter:
         if self._llm_gateway is None:
             return 0.5
 
-        prompt = _COMPLEXITY_CLASSIFY_PROMPT.format(content=content)
+        prompt = (
+            'You are a complexity classifier. Rate the complexity of the user request on a scale of 0.0 to 1.0.\n'
+            '0.0 = trivial greeting, 0.3 = simple question, 0.5 = moderate task, '
+            '0.7 = complex multi-step task, 1.0 = very complex research task.\n\n'
+            f'User request: "{content}"\n\n'
+            'Respond ONLY with a JSON object: {"complexity": <float>}'
+        )
         try:
             response = await self._llm_gateway.chat(
                 messages=[{"role": "user", "content": prompt}],
@@ -283,7 +279,10 @@ class CostAwareRouter:
         """Layer 2: 高复杂度任务通过 org_context.find_best_agent 路由。"""
         if self._org_context is not None and hasattr(self._org_context, "find_best_agent"):
             try:
-                best_agent = await self._org_context.find_best_agent(content)
+                # Extract capability-like keywords from content for matching
+                # find_best_agent expects list[str] of required capabilities
+                content_words = [w for w in content.split() if len(w) > 2][:5]
+                best_agent = self._org_context.find_best_agent(required_capabilities=content_words)
                 if best_agent is not None:
                     agent_name = best_agent if isinstance(best_agent, str) else getattr(best_agent, "name", str(best_agent))
                     result = SkillRoutingResult(

src/agentkit/evolution/lifecycle.py

@@ -149,7 +149,7 @@ class EvolutionMixin:
 
         # Step 2: Soul 进化检查
         if memory_store is not None:
-            await self.evolve_soul(task, result, memory_store)
+            await self.evolve_soul(task, result, memory_store, reflection=reflection)
 
         # Step 3: 如果有改进建议，触发 Prompt 优化
         if not reflection.suggestions:
@@ -378,6 +378,7 @@ class EvolutionMixin:
         task: TaskMessage,
         result: TaskResult,
         memory_store: MemoryStore | None = None,
+        reflection: Reflection | None = None,
     ) -> bool:
         """Check if soul should be updated based on accumulated reflections.
 
@@ -389,10 +390,10 @@ class EvolutionMixin:
         if memory_store is None:
             return False
 
-        if self._reflector is None:
+        if reflection is None:
-            return False
+            if self._reflector is None:
-
+                return False
-        reflection = await self._reflector.reflect(task, result)
+            reflection = await self._reflector.reflect(task, result)
 
         # 只关注低质量且有建议的反思
         if reflection.quality_score >= 0.5:

src/agentkit/quality/alignment.py

@@ -159,7 +159,11 @@ class AlignmentGuard:
                 )
         except Exception as e:
             logger.warning(f"LLM audit failed: {e}")
-            return AlignmentCheckResult(passed=True, checked_by="rule")
+            return AlignmentCheckResult(
+                passed=False,
+                violations=[f"LLM audit unavailable: {e}"],
+                checked_by="rule",
+            )
 
     def record_interaction(self, session_id: str) -> CascadeAlert | None:
         """记录一次 agent 间交互，超过阈值返回 CascadeAlert"""

src/agentkit/server/app.py

@@ -487,6 +487,7 @@ def create_app(
     app.state.runner = BackgroundRunner(task_store=app.state.task_store)
     app.state.server_config = server_config
     app.state.api_key = effective_api_key
+    app.state._config_reload_lock = asyncio.Lock()
 
     # Initialize session manager for Chat mode
     from agentkit.session.manager import SessionManager

tests/integration/test_marketplace_e2e.py

@@ -111,7 +111,7 @@ class TestCapabilityMatching:
         ))
 
         # Mock find_best_agent to return the research agent
-        org_context.find_best_agent = AsyncMock(
+        org_context.find_best_agent = MagicMock(
             return_value=org_context.get_agent_profile("research_agent")
         )
 
@@ -515,7 +515,7 @@ class TestFullPipeline:
             skills=["market_analysis"],
             current_load=0,
         ))
-        org_context.find_best_agent = AsyncMock(
+        org_context.find_best_agent = MagicMock(
             return_value=org_context.get_agent_profile("analyst")
         )
 

tests/unit/test_cost_aware_router.py

@@ -267,7 +267,7 @@ class TestLayer2CapabilityMatching:
         """'做市场调研+竞品分析' 复杂度 > 0.7，触发能力匹配"""
         gateway = _make_llm_gateway(json.dumps({"complexity": 0.85}))
         org_context = MagicMock()
-        org_context.find_best_agent = AsyncMock(return_value="market-researcher")
+        org_context.find_best_agent = MagicMock(return_value="market-researcher")
 
         router = CostAwareRouter(llm_gateway=gateway, model="default", org_context=org_context)
         result = await router.route(
@@ -289,7 +289,7 @@ class TestLayer2CapabilityMatching:
         agent_obj = MagicMock()
         agent_obj.name = "analyst-agent"
         org_context = MagicMock()
-        org_context.find_best_agent = AsyncMock(return_value=agent_obj)
+        org_context.find_best_agent = MagicMock(return_value=agent_obj)
 
         router = CostAwareRouter(llm_gateway=gateway, model="default", org_context=org_context)
         result = await router.route(