fix: preserve eval tasks for repeated node visits and fix lint

_pending_eval_tasks was keyed by node key, so repeated visits (e.g. cycles
or tool loops) would silently overwrite earlier eval tasks. Changed to
Dict[str, List[Task]] with setdefault/append so all invocations are tracked.
flush() now iterates the full list per node.

Also wraps the long __evaluate_agent call in create_agent to satisfy E501.

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

Author: jsonbailey

packages/ai-providers/server-ai-langchain/src/ldai_langchain/langgraph_agent_graph_runner.py

@@ -84,7 +84,7 @@ def __init__(
         self._compiled: Any = None
         self._fn_name_to_config_key: Dict[str, str] = {}
         self._node_keys: Set[str] = set()
-        self._pending_eval_tasks: Dict[str, asyncio.Task] = {}
+        self._pending_eval_tasks: Dict[str, List[asyncio.Task]] = {}
 
     def _ensure_compiled(self) -> None:
         """Build and cache the compiled graph if not already done."""
@@ -188,7 +188,8 @@ async def invoke(state: WorkflowState) -> dict:
                         output_text = (
                             response.content if hasattr(response, 'content') else str(response)
                         )
-                        self._pending_eval_tasks[nk] = node_obj.get_config().evaluator.evaluate(input_text, output_text)
+                        task = node_obj.get_config().evaluator.evaluate(input_text, output_text)
+                        self._pending_eval_tasks.setdefault(nk, []).append(task)
 
                     return {'messages': [response]}
 

packages/ai-providers/server-ai-langchain/src/ldai_langchain/langgraph_callback_handler.py

@@ -195,8 +195,9 @@ async def flush(self, graph: AgentGraphDefinition, eval_tasks=None) -> None:
         Call this once after the graph run completes.
 
         :param graph: The AgentGraphDefinition whose nodes hold the LD config trackers.
-        :param eval_tasks: Optional dict mapping node key to an awaitable that returns
-            judge evaluation results.
+        :param eval_tasks: Optional dict mapping node key to a list of awaitables that
+            return judge evaluation results. Multiple tasks arise when a node is visited
+            more than once (e.g. in a graph with cycles).
         """
         node_trackers: Dict[str, Any] = {}
         for node_key in self._path:
@@ -226,11 +227,8 @@ async def flush(self, graph: AgentGraphDefinition, eval_tasks=None) -> None:
             if not eval_tasks:
                 continue
 
-            eval_task = eval_tasks.get(node_key)
-            if not eval_task:
-                continue
-
-            results = await eval_task
-            for r in results:
-                if r.success:
-                    config_tracker.track_judge_result(r)
+            for eval_task in eval_tasks.get(node_key, []):
+                results = await eval_task
+                for r in results:
+                    if r.success:
+                        config_tracker.track_judge_result(r)

packages/sdk/server-ai/src/ldai/client.py

@@ -459,7 +459,10 @@ async def create_agent(
         """
         self._client.track(_TRACK_USAGE_CREATE_AGENT, context, key, 1)
         log.debug(f"Creating managed agent for key: {key}")
-        config = self.__evaluate_agent(key, context, default or _DISABLED_AGENT_DEFAULT, variables, default_ai_provider=default_ai_provider)
+        config = self.__evaluate_agent(
+            key, context, default or _DISABLED_AGENT_DEFAULT, variables,
+            default_ai_provider=default_ai_provider,
+        )
 
         if not config.enabled:
             return None