diff --git a/src/interface.jl b/src/interface.jl
index 62623bb..f550cbd 100755
--- a/src/interface.jl
+++ b/src/interface.jl
@@ -4,7 +4,8 @@ module interface
 export agentReact, agentReflex,
         addNewMessage, clearMessage, removeLatestMsg, conversation, directconversation, 
         writeEvaluationGuideline, grading, analyze, selfReflext, 
-        formulateUserresponse, extractinfo, updateEnvState, chat_mistral_openorca
+        formulateUserresponse, extractinfo, updateEnvState, chat_mistral_openorca, 
+        recap
 
 using JSON3, DataStructures, Dates, UUIDs, HTTP
 using CommUtils, GeneralUtils
@@ -128,7 +129,7 @@ function chat_mistral_openorca(a::agentReflex)
     $(a.roles[a.role])
     Your earlier talk with the user:
     $(a.earlierConversation)
-    <|/s|>
+    </s>
     $conversation
     <|assistant|>
   """
@@ -202,7 +203,7 @@ function planner_mistral_openorca(a::agentReflex)
     Your job is to do the following:
     Plan: first you should always think about your conversation with the user and your earlier work thoroughly then extract and devise a complete, task by task plan to achieve your objective (pay attention to correct numeral calculation and commonsense).
     P.S.1 each task of the plan should be a single action.
-    <|/s|>
+    </s>
     $conversation
     <|assistant|>
     Plan: 
@@ -256,7 +257,7 @@ function updatePlan(a::agentReflex)
     Plan: 1. Ask the user for their food type.
     Obs: It will be Thai dishes.
     Updated plan: 1. Ask the user for their food type (Thai dishes).
-    </s|>
+    </s>
     Updated plan: 
   """
 
@@ -304,7 +305,9 @@ function actor_mistral_openorca(a::agentReflex, taskrecap="")
 
   shorttermMemory = dictToString(a.memory[:shortterm], skiplist=["user:"])
 
-  # conversation = messagesToString_nomark(a.messages, addressAIas="I")
+  # conversation = conversationSummary(a)
+  # println("")
+  # @show conversationSum = conversation
   
   # context = 
   # """
@@ -322,9 +325,9 @@ function actor_mistral_openorca(a::agentReflex, taskrecap="")
       <Your plan>
       $(a.memory[:shortterm]["Plan 1:"])
       </Your plan>
-      <Your earlier work's recap>
+      <Your earlier work's recap in JSON format>
       $taskrecap
-      </Your earlier work's recap>
+      </Your earlier work's recap in JSON format>
       <Your job> 
       Use the following format:
       Thought: based on the plan and the recap of the plan, what to do? (pay attention to correct numeral calculation and commonsense).
@@ -332,16 +335,16 @@ function actor_mistral_openorca(a::agentReflex, taskrecap="")
       Actinput: your input to the action based on your thought (pay attention to the tool's input)
       Obs: observed result of the action
 
-      P.S.1 ask user one by one question.
+      P.S.1 ask the user one by one question.
       </Your job>
-    <|/s|>
+    </s>
     <|assistant|>
     Thought: 
   """
   prompt = replace(prompt, "{toolnames}" => toolnames)
 
   println("")
-  @show prompt_actor = prompt
+  @show actor_prompt = prompt
 
   response = nothing
   chunkedtext = nothing
@@ -370,7 +373,7 @@ function actor_mistral_openorca(a::agentReflex, taskrecap="")
     headers = detectCharacters(response, headerToDetect)
 
     println("")
-    @show response_actor = response
+    @show actor_response = response
 
     headerToDetect = ["Plan $(a.attempt):",
                       "Thought $latestTask:", 
@@ -925,7 +928,7 @@ function formulateUserresponse(a)
     $work
     
     From your talk with the user and your work, formulate a response for the user.
-    <|/s|>
+    </s>
     <|assistant|>
     response:
   """
@@ -1088,14 +1091,14 @@ function checkTaskCompletion(a)
       </Example 1>
 
       Let's think step by step.
-    </s|>
+    </s>
     <|assistant|> After 
   """
   response = nothing
   _response = nothing
   _response = sendReceivePrompt(a, prompt, max_tokens=512)
   @show checkTaskCompletion_raw = _response
-  _response = split(_response, "</s|>")[1]
+  _response = split(_response, "</")[1]
   _response = split(_response, "\n\n")[1]
   # response = "I " * split(_response, "{")[1]  # sometime response have more than 1 {answer: done}
 
@@ -1110,9 +1113,46 @@ function checkTaskCompletion(a)
 end
 
 function recap(a)
-  @show a.memory[:shortterm]["Plan 1:"]
   # stimulus = a.memory[:shortterm]["user:"]
-  work = dictToString(a.memory[:shortterm])
+  getonlykeys = ["Actinput", "Obs"]
+  worknoplan = similar(a.memory[:shortterm])
+  for (k, v) in a.memory[:shortterm]
+    count = 0
+    for i in getonlykeys
+      if occursin(i, k)
+        count += 1
+      end
+    end
+    if count != 0
+      worknoplan[k] = v 
+    end
+  end
+
+  work = dictToString(worknoplan)
+
+  # prompt =
+  # """
+  #   <|system|>
+  #     <Symbol meaning>
+  #     Plan: a plan
+  #     Thought: your thought
+  #     Act: the action you took
+  #     Actinput: the input to the action
+  #     Obs: the result of the action
+  #     </Symbol meaning>
+
+  #     <Your earlier work>
+  #     $work
+  #     </Your earlier work>
+  #     <Your job>
+  #     Recap: list all your observed results in detail
+  #     </Your job>
+
+  #     Let's think step by step.
+  #   </s>
+  #   <|assistant|>
+  #   Recap: 
+  # """
 
   prompt =
   """
@@ -1129,16 +1169,19 @@ function recap(a)
       $work
       </Your earlier work>
       <Your job>
-      Recap: list all your observed results in detail
+      Extract info: extract info in details from ALL Actinput and corresponding observed result into JSON format
       </Your job>
 
       Let's think step by step.
-    </s|>
+    </s>
     <|assistant|>
-    Recap: 
+    Extracted info: 
   """
+
   response = sendReceivePrompt(a, prompt, max_tokens=512, temperature=0.0)
-  response = split(response, "</s|>")[1]
+  response = split(response, "</")[1]
+  response = split(response, "<|")[1]
+  response = split(response, "\n\n")[1]
   
   return response
 end
diff --git a/src/type.jl b/src/type.jl
index 2caea59..f476785 100644
--- a/src/type.jl
+++ b/src/type.jl
@@ -173,7 +173,8 @@ function agentReflex(
       Actinput: your input to the action based on your thought (pay attention to the tool's input)
       Obs: observed result of the action
 
-      P.S.1 ask user one by one question.
+      P.S.1 ask the user one by one question.
+      P.S.2 ask the user what you want to know if you didn't ask yet.
       </Your job>
       """,
     :actorOriginal=> 
diff --git a/src/utils.jl b/src/utils.jl
index 3aef1c5..e4f626c 100644
--- a/src/utils.jl
+++ b/src/utils.jl
@@ -1,6 +1,6 @@
 module utils
 
-export makeSummary, sendReceivePrompt, chunktext, extractStepFromPlan, checkTotalTaskInPlan,
+export sendReceivePrompt, chunktext, extractStepFromPlan, checkTotalTaskInPlan,
         detectCharacters, findDetectedCharacter, extract_number, toolNameBeingCalled,
         isUsePlans, conversationSummary, checkReasonableness, replaceHeaders,
         addShortMem!, splittext, dictToString, removeHeaders, keepOnlyKeys, experience,
@@ -12,49 +12,6 @@ using ..type
 
 #------------------------------------------------------------------------------------------------100
 
-function makeSummary(a::T1, input::T2) where {T1<:agent, T2<:AbstractString}
-  summary = "Nothing."
-  prompt = 
-  """
-    <|im_start|>system
-    Input text:
-    $input
-
-    Your job is to determine now whether you can make a summary of the input text by choosing one of following choices:
-    If you cannot make a summary say, "{No}".
-    If you can make a summary say, "{Yes}".
-    <|im_end|>
-  """
-  prompt = replace(prompt, "{input}" => input)
-  result = sendReceivePrompt(a, prompt)
-  result = GeneralUtils.getStringBetweenCharacters(result, "{", "}")
-  if result == "Yes"  # seperate summary part 
-    prompt = 
-    """
-      <|im_start|>system
-      Input text:
-      $input
-      Your job is to make a concise summary of the input text.
-      <|im_end|>
-    """
-    result = sendReceivePrompt(a, prompt)
-    if result[1:1] == "\n"
-      summary = result[2:end]
-    else
-      summary = result
-    end
-  end
-  input_summary = input
-  @show input_summary
-  @show summary
-
-  return summary
-end
-
-
-
-
-
 """
   Send a msg to registered mqtt topic within mqttClient.
   
@@ -496,43 +453,22 @@ end
   ```
 """
 function conversationSummary(a::T) where {T<:agent}
+  conversation = messagesToString_nomark(a.messages, addressAIas="I")
   prompt = 
   """
-    <|im_start|>system
-    You talked with a user earlier.
-    Now you make a detailed bullet summary of the conversation from your perspective.
+    <|system|>
+    Your conversation with the user:
+    $conversation
+
+    Your job is to paraphrase a conversation from your perspective.
     You must refers to yourself by "I" in the summary.
-
-    Here are the conversation:
-    {conversation}
-    <|im_end|>
-
+    <|/s|>
+    <|assistant|>
+    Paraphrase:  
   """
-  conversation = ""
-  summary = ""
-  if length(a.messages)!= 0
-    for msg in a.messages[1:end-1]
-      role = msg[:role]
-      content = msg[:content]
-
-      if role == "user"
-        conversation *= "$role: $content\n"
-      elseif role == "assistant"
-        conversation *= "I: $content\n"
-      else
-        error("undefied condition role = $role $(@__LINE__)")
-      end
-    end
-
-    prompt = replace(prompt, "{conversation}" => conversation)
-    result = sendReceivePrompt(a, prompt)
-    summary = result === nothing ? "N/A" : result
-    summary = split(summary, "<|im_end|>")[1]
-    if summary[1:1] == "\n"
-      summary = summary[2:end]
-    end
-  end
-  @show summary
+  result = sendReceivePrompt(a, prompt)
+  summary = split(result, "<|/s|>")[1]
+  summary =
   return summary
 end