update

src/interface.jl

@@ -631,6 +631,8 @@ function work(a::agentReflex, usermsg::String)
 
   if a.thinkingmode == :new_thinking
     _ = addNewMessage(a, "user", usermsg)
+    a.memory[:shortterm] = OrderedDict{String, Any}()
+    a.memory[:log] = OrderedDict{String, Any}()
     a.memory[:shortterm]["user:"] = usermsg
     a.memory[:log]["user:"] = usermsg
     a.newplan = true
@@ -655,9 +657,7 @@ function work(a::agentReflex, usermsg::String)
         println("")
         @show prompt_plan
         respond = sendReceivePrompt(a, prompt_plan, max_tokens=1024)
-        println("")
-        plan_raw = respond
-        @show plan_raw
+
         # sometimes LLM add not-need word I don't want
         plan = splittext(respond, ["Step 1", "<|im_end|>", "Response", "Execution", 
                                   "Result", "Recommendation", "My response"])
@@ -670,8 +670,8 @@ function work(a::agentReflex, usermsg::String)
         a.memory[:log]["Plan $(a.attempt):"] = plan
       end
       
+      println("")
       @show a.attempt
-      @show a.memory[:shortterm]["user:"]
 
       # enter actor loop
       actorstate, msgToUser = actor(a)
@@ -693,18 +693,22 @@ function work(a::agentReflex, usermsg::String)
 
         # evaluate. if score < 8/10 try again.
         guideline = writeEvaluationGuideline(a, a.memory[:shortterm]["user:"])
+
+        println("")
         @show guideline
+
         score = grading(a, guideline, respond)
         @show score
-        if score >= 8 # good enough answer
-          @show a.memory[:shortterm]
-          a.memory[:shortterm] = OrderedDict{String, Any}()
-          a.memory[:log] = OrderedDict{String, Any}()
+        if score >= 6 # good enough answer
           break
         else  # self evaluate and reflect then try again
           analysis = analyze(a)
+          println("")
           @show analysis
+          
           lessonwithcontext = selfReflext(a, analysis)
+          
+          println("")
           @show lessonwithcontext
 
           newdict = OrderedDict()
@@ -714,10 +718,13 @@ function work(a::agentReflex, usermsg::String)
           headerToDetect = ["Lesson:", "Context:", ]
           headers = detectCharacters(lessonwithcontext, headerToDetect)
           chunkedtext = chunktext(lessonwithcontext, headers)
-          @show chunkedtext
+
           a.memory[:longterm][chunkedtext["Context:"]] = chunkedtext["Lesson:"]
           a.newplan = true
           println("")
+          println("RETRY $(a.attempt +1)")
+          println("")
+          
         end
       else
         error("undefied condition, actorstate $actorstate $(@__LINE__)")
@@ -796,13 +803,13 @@ function actor(a::agentReflex)
     if a.step < totalsteps
 
       prompt_actor = actor_mistral_openorca(a)
+
       println("")
       @show prompt_actor
+
       respond = sendReceivePrompt(a, prompt_actor)
       respond = splittext(respond, ["Obs", "<|im_end|>"])
-      respond_actor_raw = respond
-      println("")
-      @show respond_actor_raw
+
       if !occursin("Thought", respond)
         respond = "Thought: " * respond
       end
@@ -816,8 +823,8 @@ function actor(a::agentReflex)
       
       headers = detectCharacters(respond, headerToDetect)
 
-      respond_actor = respond
       println("")
+      respond_actor = respond
       @show respond_actor
 
       headerToDetect = ["Plan $(a.attempt):", 
@@ -828,7 +835,7 @@ function actor(a::agentReflex)
                         "Check $(a.step):",]
       headers = detectCharacters(respond, headerToDetect)
       chunkedtext = chunktext(respond, headers)
-      @show chunkedtext
+
 
       # add to memory
       a.memory[:shortterm] = addShortMem!(a.memory[:shortterm], chunkedtext)
@@ -1220,9 +1227,7 @@ function goNogo(a)
 
   respond = sendReceivePrompt(a, prompt)
   
-  println("")
-  goNogo_raw = respond
-  @show goNogo_raw
+
 
   decision = nothing
   reason = nothing