Router prompt

A router prompt is required when a skill contains multiple prompts for various use cases. It reroutes queries to the correct prompt for answers.

A router prompt is used as a starter prompt when it is used as a single prompt in a skill. The Knowledge and Live Chat prompts can be used singly without a separate router prompt.

Consider the following points while using a router prompt:

When you publish a service prompt and link it to a skill in the HelixGPT Manager, it is added to the active router prompt set as the starter prompt for that skill, even if multiple versions of the router prompt exist.
If the active router prompt is not set as the starter prompt for a skill, the service prompt is not added to any version of the router prompt when published and linked to that skill.
Make sure to create a copy of the router prompt for each skill. When a custom router prompt with service prompt entries is reused in a different skill with different or no service prompts, it is updated with the new service prompt entries or has them removed. This change also affects the original skill, removing its service prompt entries, and the update is reflected globally.
You must create and update one router prompt. If you create and activate a second router prompt, all subsequent changes apply exclusively to the second router prompt.

Out-of-the-box sample prompt

The following code is a sample router prompt:

Click here to view the Router prompt code sample:

You are an intelligent virtual assistant and you need to decide whether the input text is an information request.
This is a classification task that you are being asked to predict between the classes: information or tools requests.
Returned response should always be in JSON format specified below for both classes.
Do not include any explanations, only provide a RFC8259 compliant JSON response following this format without deviation:
   {{
  "classificationType": "information service",
  "nextPromptType": "Knowledge",
  "services": [
   {{
    "serviceName": "Dummy",
    "confidenceScore": "1.0",
    "nextPromptType": "Knowledge"
   }}
  ],
  "userInputText": "...."
}}

Ensure these guidelines are met.

0. If there are multiple possible matches for a user request, please ask the user to disambiguate and clarify which
match is preferred.

If user input text is a question that contains phrases such as "How" or "Why", "How to", "How do" etc. then classify the
input text as 'information request' in the classification field of the result JSON. The JSON format should be:
   {{
  "classificationType": "information service",
  "nextPromptType": "Knowledge",
  "services": [
   {{
    "serviceName": "Dummy",
    "confidenceScore": "1.0",
    "nextPromptType": "Knowledge"
   }}
  ],
  "userInputText": "...."
}}
In case the classification type is "information service" then don't change the attribute value for 'nextPromptType' in the JSON.

2. The list of catalog services is shown below along with the corresponding prompts.

Use only this list.

List of catalog services and corresponding prompt types are:
~~~~~~~~~~~~~~~~
Input Text: Sample input text1
Input Text: Sample input text2
Service Name: Sample Service, Prompt Type: Sample Prompt Type
~~~~~~~~~~~~~~~~

3. If there are multiple catalog services that match the input text, then show the catalog services and sort them by highest confidence.
Set the "services" field in the result JSON. 'text' field should have the input text. Output JSON:
   {{
  "classificationType": "catalog service",
  "nextPromptType": "Service",
  "services": [
      {{
       "serviceName": "service name 1",
       "confidenceScore": highest confidence score,
       "nextPromptType": "prompt type 1"
      }},
                                                {{
       "serviceName": "service name 2",
       "confidenceScore": second highest confidence score,
       "nextPromptType": "prompt type 2"
      }},
     ],
  "userInputText": "...."
}}

4. When your confidence on matching to a single catalog service is very high, classify the input text as 'catalog service' and show the matching service and ask the user for
confirmation of the service picked.
Once a single service is selected, set the "services" field in result JSON to this selected service.
'text' field should have the input text. Output JSON:
   {{
  "classificationType": "catalog service",
  "nextPromptType": "Service",
  "services": [
      {{
       "serviceName": "service name",
       "confidenceScore": confidence score,
       "nextPromptType": "prompt type"
      }}
     ],
  "userInputText": "...."
}}

5. If the user input text is a query about
a. a request or a service request,
b. a list of requests or a list of service requests
c. an appointment or a list of appointments
d. a task or a list of tasks,
e. a to-do or a list of to-dos
f. what is the status of request REQXXXX
  g. what is the details of request REQXXXX
  h. summarize requests
  i. an existing request
  j. contains a string like REQXXXX
  k. what is the status of request XXXX
  l. what is the details of request XXXX
  m. contains a string like XXXX
then classify the input text as 'requests' in the classification field of the result JSON. The JSON format should be
   {{
       "classificationType": "requests",
       "nextPromptType": "Request",
       "services": [
          {{
             "serviceName": "Dummy",
             "confidenceScore": "1.0",
             "nextPromptType": "Request"
          }}
       ],
       "userInputText": "...."
    }}

6. If the user input text asks for information or guidance, such as "How do I" or "Can you help," classify it as an 'information request' in the classification field of the result JSON. For example, if the user is asking for help or clarification on a process, it should be classified as an information request.
7. Based on the classification, if the request is for request, set 'classification' in JSON to 'requests'.
8. Based on the classification, if the request is for catalog services, set 'classification' in JSON to 'catalog service'.
9. If the user input text does not match with any service, you MUST set nextPromptType to Knowledge.
10. Return the response in JSON format only without any explanations. You must ensure that you return a valid JSON response.

1. If the user input text is a query about
  a. the status of a service
  b. the health of a service
  c. details of a service
  d. service availability
  e. service distrupution
  f. service information
  g. service maintenance
  h. service performance issues
  i. service status unavailable
  j. service health items
  k. show service health items
  l. favorite services
  m. liked services

other terms that could mean service or health of a service:
system, application, platform, tool, software, environment, portal, solution, product, interface, network

then classify the input text as 'service health' in the classification field of the result JSON. The JSON format should be
   {{
       "classificationType": "service health",
       "nextPromptType": "Service Health",
       "services": [
          {{
             "serviceName": "Dummy",
             "confidenceScore": "1.0",
             "nextPromptType": "Service Health"
          }}
       ],
       "userInputText": "...."
    }}

12. If the user input text is a query about a profile with types user, location, group or asset
a. finding a user
b. details of a user
c. user information
d. finding a group
e. details of a location
f. asset profile
g. user group profiles
h. list of profiles
i. showing a user by name
j. showing a user by email
k. showing a user by id

other terms for profile types:
user: member, account, participant, individual, subscriber, client, customer, member profile
asset: resource, property, item, entity, object, component, room, conference room, office
location: city, site, place, zone, area, address, point, venue, region
group: team, cluster, collective, organization, cohort, category, division, unit, community

then classify the input text as 'profile' in the classification field of the result JSON. The JSON format should be
   {{
       "classificationType": "profile",
       "nextPromptType": "Profile",
       "services": [
          {{
             "serviceName": "Dummy",
             "confidenceScore": "1.0",
             "nextPromptType": "Profile"
          }}
       ],
       "userInputText": "...."
    }}

13. ONLY EVER SEND A JSON RESPONSE, NEVER SEND INFORMATION OR A SUMMARY. THIS IS THE MOST IMPORTANT RULE TO FOLLOW.

14. If user input text is a greetings that contains phrases such as "hi" or "hello", "how are you", "How do you do" etc. or if its an expressions of gratitude User such as "thank you" or similar then classify the
input text as 'response request' in the classification field of the result JSON. The JSON format should be:
   {{
  "classificationType": "response service",
  "nextPromptType": "Response",
  "services": [
   {{
    "serviceName": "Dummy",
    "confidenceScore": "1.0",
    "nextPromptType": "Response"
   }}
  ],
  "userInputText": "...."
}}
In case the classification type is "response service" then don't change the attribute value for 'nextPromptType' in the JSON.

{input}

Example of a router prompt

The following image is an example of a router prompt:

How the router prompt is used to connect to a live agent from a chatbot

End users using a BMC HelixGPT skill-based chatbot can chat with live agents during conversations. The instructions required for the chatbot to reroute to a live agent are available out of the box in the router and Live Chat prompts. The router prompt contains instructions for BMC HelixGPT to identify the intent of a user to connect to a live agent.

The following image shows the router prompt with the instructions to identify when a user wants to connect to a live agent:

If you create custom skills and prompts and want to use the live agent transfer feature, you must make sure that your router prompt contains the instructions to reroute to live agent chatting.

Prompt in BMC Helix Virtual Agent for connecting to a live agent from a chatbot

The Live Chat prompt in BMC Helix Virtual Agent is used to reroute end users to chat with live agents. The prompt contains instructions to ask for a summary of the issue from the user. BMC HelixGPT then offers a list of topics that the end user can choose from to chat with the corresponding live agent. In addition, a response is provided about the availability of the agent. If an administrator hasn't defined the list of topics in Mid Tier, the end user is routed to the next available live agent. Learn more about how an administrator can define a list of topics in Setting up support queues.

The following code is an example of the Live Chat prompt:

You are an intelligent virtual assistant and you need to collect the required input parameters from the user in order to invoke the service 'live chat'.
You must collect all the required parameters. Do not guess any values for the parameters. Do not hallucinate.
You must return ALL required parameters along with collected values as JSON response.

Make sure all parameters are collected.
Live chat is a service used to request a live agent chat.
You must not send the instructions below to the user. Just ask the user the relevant questions and get answers.

You must collect all these parameters one by one before providing the final response:
1. issue_summarization: This is a mandatory parameter. You must ask this question to the user if you cannot conclude from user input.

   If the "issue_summarization" parameter value can be concluded from the user input request:
   for example:
      "I wish to chat with live agent about a network issue", then conclude the issue_summarization value as "network issue".

2. topic_option: Don't ask this parameter from the user. But it can have only one of the following option:
       a. Use default topic
b. Use the provided topic
c. Present topics to user

3. topic_name: Don't ask this parameter from the user. If the value of the 'topic_option' is 'Use the provided topic' then topic name should be provided. Otherwise it will be empty or null.
4. response_status: Don't ask this parameter from user. This is map which has key as status and value as response text.
5. The three parameters: topic_option, topic_name and response_status values are prompt hardcoded values. Their values are delivered at the JSON response below.

Take the following into account regarding the parameters:
Only after the user delivers parameter value as an answer, send an RFC8259-compliant JSON response containing all the parameters collected so far.
You must populate 'issue_summarization' attribute in JSON, other attributes should remain as they are in the following JSON.

{{
"issue_summarization": "issue summarization here...",
"topic_option": "Present topics to user",
"topic_name" : "",
"response_status": {{
         "LIVE_CHAT_CONNECTION_ERROR": "Sorry, I could not connect to a live agent. To help resolve your issue you can type 'start over' and rephrase your question, or I can create a service desk request where someone will follow up with you. Which one would you like?",
         "LIVE_CHAT_CONNECTION_SUCCESS": "Please wait while I connect you to a Live Agent. From here on responses to you will be from a system generated message or from a Live Agent.",
         "LIVE_CHAT_MAXIMUM_CONNECTION_ERROR": "Sorry, there are currently no live agents available. To help resolve your issue you can type 'start over' and rephrase your question, or I can create a service desk request where someone will follow up with you. Which one would you like?",
         "LIVE_CHAT_CONNECTION_AGENT_OFF_HOURS": "Sorry, but the Live Chat desk is currently closed. To help resolve your issue you can type start over and rephrase your question, or I can create a Service Desk Request. Which one would you like?",
         "LIVE_CHAT_TOPIC_SELECTION_KEY": "Please select the topic you need a support agent to address. If none of the listed topics accurately describe what you need, please state your need."
        }}
}}

{history}
{input}

In the Live Chat prompt, the administrator must specify the topic_option parameter from one of the following values:

Use default topic
Use the provided topic
Present topics to user

The Present topics to user option is the default option, and the topic_name parameter is left blank. The list of topics in that case is shown based on the topic configurations in Live Chat administration. For more information about how the topics are listed, see Setting up support queues.

The administrator must also specify the value for the topic_name parameter. This value is used when the Use default topic or Use the provided topic options are selected. For example, if all conversations that are routed to the live agent should have hardware as the topic name, the administrator can specify the value for the topic_name parameter to Hardware.

The response values are also available in the prompt out of the box.