Update task_temp.py

task_temp.py

@@ -2,8 +2,10 @@ rag_temp = '''First retrieve some example questions using the tool "QuestionRetr
 QuestionRetrieverTool(subject="{subject}", topic="{topic}", grade="{grade}")
 Use the returned questions as inspiration.
 
-Then, generate a question and answer pair that matches the target cognitive demand.
-You generate {subject} question candidates for {grade} on "{topic}".
+Then, generate a question and answer pair for the given target cognitive demand while taking into consideration the examples that were retrived.
+
+You generate {subject} question for {grade} on "{topic}".
+
 After you propose a candidate, you MUST immediately call:
 classify_and_score(
     question=<just the question text>,
@@ -12,14 +14,14 @@ classify_and_score(
     agg="max"
 )
 Use the "QuestionRetrieverTool" only once at the start.
-Keep answers concise.
+Keep answers well structured and detailed.
 Use the returned dict:
     - print ONLY compact JSON {{"question": "...", "answer": "...", "reasoning": "..."}}
 Additionally, when you call classify_and_score, pass the exact question text you propose.
 If you output JSON, ensure it is valid JSON (no trailing commas, use double quotes).
 '''
 # RAG + Classification + Generation
-rag_cls_temp = '''First retrieve some example questions using the tool "QuestionRetrieverTool".
+rag_cls_temp = '''First retrieve example questions using the tool "QuestionRetrieverTool".
 QuestionRetrieverTool(subject="{subject}", topic="{topic}", grade="{grade}")
 Use the returned questions to inspire your own candidate.
 
@@ -36,10 +38,10 @@ classify_and_score(
 
 Use the returned dict:
 - If ok == True: print ONLY compact JSON {{"question": "...", "answer": "...", "reasoning": "..."}} and finish.
-- If ok == False: briefly explain the needed shift, revise the question based on the feedback from the results of classify_and_score from the previous question, and call classify_and_score again.
-Repeat up to {attempts} attempts.
+- If ok == False: briefly explain the needed shift, revise the question based on the feedback from the results of classify_and_score from the 1 preceding question to the current generated candidate and call classify_and_score again.
+Repeat max up to {attempts} attempts.
 Use the "QuestionRetrieverTool" only once at the start.
-Keep answers concise.
+Keep answers well structured and detailed.
 Additionally, when you call classify_and_score, pass the exact question text you propose.
 If you output JSON, ensure it is valid JSON (no trailing commas, use double quotes).
 '''
@@ -54,7 +56,7 @@ Use the returned dict:
 - If ok == True: print ONLY compact JSON {{"question": "...", "answer": "...", "reasoning": "..."}} and finish.
 - If ok == False: briefly explain the needed shift, revise the question, and call classify_and_score again.
 Repeat up to {attempts} attempts.
-
+Keep answers well structured and detailed.
 If you output JSON, ensure it is valid JSON (no trailing commas, use double quotes).
 '''
 # Generation only
@@ -71,7 +73,7 @@ Provide the question and answer in the following JSON format:
   "reasoning": "..."
 }}
 Ensure that the question is clear, concise, and appropriate for the specified grade level and subject matter.
-Keep answers concise.
+Keep answers well structured and detailed.
 If you output JSON, ensure it is valid JSON (no trailing commas, use double quotes).
 '''