- Refinement of the chat client to have better visible clues for user vs chatbot messages

- Introduction of interview_phase and normal phase in TRAICIE_SELECTION_SPECIALIST to make interaction with bot more human.
- More and random humanised messages to TRAICIE_SELECTION_SPECIALIST

config/tasks/globals/ADVANCED_RAG_TASK/1.0.0.yaml

@@ -0,0 +1,43 @@
+version: "1.0.0"
+name: "Advanced RAG Task"
+task_description: >
+  Answer the following question (in between triple £):
+  
+  £££{question}£££
+  
+  Base your answer on the following context (in between triple $): 
+
+  $$${context}$$$
+
+  Take into account the following history of the conversation (in between triple €):
+
+  €€€{history}€€€
+
+  The HUMAN parts indicate the interactions by the end user, the AI parts are your interactions.
+  
+  Best Practices are:
+  
+  - Answer the provided question as precisely and directly as you can, combining elements of the provided context.
+  - Always focus your answer on the actual question.
+  - Limit repetition in your answers to an absolute minimum, unless absolutely necessary. 
+  - Always be friendly and helpful for the end user.
+  
+  Tune your answers to the following:
+  
+  - You use the following Tone of Voice for your answer: {tone_of_voice}, i.e. {tone_of_voice_context}
+  - You use the following Language Level for your answer: {language_level}, i.e. {language_level_context}
+  
+  Use the following language in your communication: {language}
+  
+  If the question cannot be answered using the given context, answer "I have insufficient information to answer this 
+  question." and give the appropriate indication.
+  
+  {custom_description}
+
+expected_output: >
+
+metadata:
+  author: "Josako"
+  date_added: "2025-07-30"
+  description: "A Task that performs RAG and checks for human answers"
+  changes: "Initial version"

config/tasks/globals/RAG_TASK/1.1.0.yaml

@@ -1,19 +1,32 @@
 version: "1.0.0"
 name: "RAG Task"
 task_description: >
-  Answer the question based on the following context, and taking into account the history of the discussion. Try not to
-  repeat answers already given in the recent history,  unless confirmation is required or repetition is essential to 
-  give a coherent answer.
+  Answer the following question (in between triple £):
+  
+  £££{question}£££
+  
+  Base your answer on the following context (in between triple $): 
+  
+  $$${context}$$$
+  
+  Take into account the following history of the conversation (in between triple €):
+  
+  €€€{history}€€€
+  
+  The HUMAN parts indicate the interactions by the end user, the AI parts are your interactions.
+  
+  Best Practices are:
+  
+  - Answer the provided question as precisely and directly as you can, combining elements of the provided context.
+  - Always focus your answer on the actual HUMAN question.
+  - Try not to repeat your answers (preceded by AI), unless absolutely necessary. 
+  - Focus your answer on the question at hand.
+  - Always be friendly and helpful for the end user.
+
   {custom_description}
-  Use the following {language} in your communication, and cite the sources used at the end of the full conversation.
+  Use the following {language} in your communication.
   If the question cannot be answered using the given context, answer "I have insufficient information to answer this 
-  question."
-  Context (in between triple $):
-    $$${context}$$$
-  History (in between triple €):
-    €€€{history}€€€
-  Question (in between triple £):
-    £££{question}£££
+  question." and give the appropriate indication.
 expected_output: >
   Your answer.
 metadata:

config/tasks/traicie/TRAICIE_AFFIRMATIVE_ANSWER_CHECK_TASK/1.0.0.yaml

@@ -0,0 +1,29 @@
+version: "1.0.0"
+name: "Traicie Affirmative Answer Check"
+task_description: >
+  You are provided with the following end user answer (in between triple £):
+  
+  £££{question}£££
+  
+  This is the history of the conversation (in between triple €):
+
+  €€€{history}€€€
+  
+  (In this history, user interactions are preceded by 'HUMAN', and your interactions with 'AI'.)
+  
+  Check if the user has given an affirmative answer or not.
+  Please note that this answer can be very short:
+  - Affirmative answers: e.g. Yes, OK, Sure, Of Course
+  - Negative answers: e.g. No, not really, No, I'd rather not.
+  
+  Please consider that the answer will be given in {language}!
+  
+  {custom_description}
+
+expected_output: >
+  Your determination if the answer was affirmative (true) or negative (false)
+metadata:
+  author: "Josako"
+  date_added: "2025-07-30"
+  description: "A Task to check if the answer to a question is affirmative"
+  changes: "Initial version"

config/tasks/traicie/TRAICIE_COMPETENCIES_INTERVIEW_DEFINITION/1.0.0.yaml

@@ -1,30 +0,0 @@
-version: "1.0.0"
-name: "KO Criteria Interview Definition"
-task_description: >
-  In context of a vacancy in your company {tenant_name}, you are provided with a set of competencies. (both description 
-  and title). The competencies are in between triple backquotes. You need to prepare for the interviews, 
-  and are to provide for each of these ko criteria:
-  
-  - A question to ask the recruitment candidate describing the context of the competency. Use your experience to not 
-  just ask a closed question, but a question from which you can indirectly derive a positive or negative qualification of
-  the competency based on the answer of the candidate.
-  
-  Apply the following tone of voice in both questions and answers: {tone_of_voice}
-  Apply the following language level in both questions and answers: {language_level}
-  Respect the language of the competencies, and return all output in the same language.
-  
-  ```{competencies}```
-  
-  {custom_description}
-
-expected_output: >
-  For each of the ko criteria, you provide:
-  - the exact title in the original language
-  - the question
-  - a set of answers, with for each answer an indication if it is the correct answer, or a false response.
-  {custom_expected_output}
-metadata:
-  author: "Josako"
-  date_added: "2025-06-15"
-  description: "A Task to define interview Q&A from given KO Criteria"
-  changes: "Initial Version"

config/tasks/traicie/TRAICIE_DETERMINE_INTERVIEW_MODE_TASK/1.0.0.yaml

@@ -0,0 +1,23 @@
+version: "1.0.0"
+name: "Traicie Determine Interview Mode"
+task_description: >
+  you are provided with the following user input (in between triple backquotes):
+  
+  ```{question}```
+
+  If this user input contains one or more questions, your answer is simply 'RAG'. In all other cases, your answer is
+  'CHECK'. 
+  
+  Best practices to be applied:
+  - A question doesn't always have an ending question mark. It can be a query for more information, such as 'I'd like 
+  to understand ...', 'I'd like to know more about...'. Or it is possible the user didn't enter a question mark. Take 
+  into account the user might be working on a mobile device like a phone, making typing not as obvious.
+  - If there is a question mark, then normally you are provided with a question of course.
+
+expected_output: >
+  Your Answer.
+metadata:
+  author: "Josako"
+  date_added: "2025-07-30"
+  description: "A Task to determine the interview mode based on the last user input"
+  changes: "Initial version"