add validation check

Signed-off-by: Mingshi Liu <[email protected]>

Author: mingshl

ml-algorithms/src/main/java/org/opensearch/ml/engine/algorithms/agent/MLChatAgentRunner.java

@@ -543,7 +543,7 @@ private void runReAct(
                         return;
                     }
                     // Emit PRE_LLM hook event
-                    if (hookRegistry != null) {
+                    if (hookRegistry != null && !interactions.isEmpty()) {
                         List<MLToolSpec> currentToolSpecs = new ArrayList<>(toolSpecMap.values());
                         ContextManagerContext contextAfterEvent = AgentContextUtil
                             .emitPreLLMHook(tmpParameters, interactions, currentToolSpecs, memory, hookRegistry);
@@ -568,9 +568,10 @@ private void runReAct(
         // Emit PRE_LLM hook event for initial LLM call
         List<MLToolSpec> initialToolSpecs = new ArrayList<>(toolSpecMap.values());
         tmpParameters.put("_llm_model_id", llm.getModelId());
-        if (hookRegistry != null) {
+        if (hookRegistry != null && !interactions.isEmpty()) {
             ContextManagerContext contextAfterEvent = AgentContextUtil
                 .emitPreLLMHook(tmpParameters, interactions, initialToolSpecs, memory, hookRegistry);
+
             if (tmpParameters.get(INTERACTIONS) != null || tmpParameters.get(INTERACTIONS) != "") {
                 tmpParameters.put(INTERACTIONS, StringUtils.toJson(contextAfterEvent.getParameters().get(INTERACTIONS)));
             }

ml-algorithms/src/main/java/org/opensearch/ml/engine/algorithms/agent/MLPlanExecuteAndReflectAgentRunner.java

@@ -396,16 +396,16 @@ private void executePlanningLoop(
         // completedSteps to context management.
         // TODO should refactor the completed steps as message array format, similar to chat agent.
 
-        Map<String, String> requestParams = new HashMap<>(allParams);
-
+        allParams.put("_llm_model_id", llm.getModelId());
         if (hookRegistry != null && !completedSteps.isEmpty()) {
-            requestParams.put("_llm_model_id", llm.getModelId());
-            requestParams.put(INTERACTIONS, ", " + String.join(", ", completedSteps));
+            allParams.put(INTERACTIONS, ", " + String.join(", ", completedSteps));
+            Map<String, String> requestParams = new HashMap<>(allParams);
             try {
                 AgentContextUtil.emitPreLLMHook(requestParams, completedSteps, null, memory, hookRegistry);
+
                 if (requestParams.get(INTERACTIONS) != null || requestParams.get(INTERACTIONS) != "") {
-                    requestParams.put(COMPLETED_STEPS_FIELD, StringUtils.toJson(requestParams.get(INTERACTIONS)));
-                    requestParams.put(INTERACTIONS, "");
+                    allParams.put(COMPLETED_STEPS_FIELD, StringUtils.toJson(requestParams.get(INTERACTIONS)));
+                    allParams.put(INTERACTIONS, "");
                 }
             } catch (Exception e) {
                 log.error("Failed to emit pre-LLM hook", e);
@@ -418,7 +418,7 @@ private void executePlanningLoop(
             RemoteInferenceMLInput
                 .builder()
                 .algorithm(FunctionName.REMOTE)
-                .inputDataset(RemoteInferenceInputDataSet.builder().parameters(requestParams).build())
+                .inputDataset(RemoteInferenceInputDataSet.builder().parameters(allParams).build())
                 .build(),
             null,
             allParams.get(TENANT_ID_FIELD)