Introduction to Writing Prompts
In order to get the most out of AI responses it is important to provide detailed and well-constructed prompts to achieve high-quality outcomes. Knowing whether you are working with a generic AI chatbot or an AI built for a specific purpose determines the level of detail to include in your prompts. Specific-purpose AIs are configured with a system message and/or uploaded documentation that users then do not need to provide context for within their prompts.
Terminology
system message: The prompt an AI chatbot designed for a specific purpose is built with that guides it's responses to users. Often it includes what role the chatbot is serving and the context in which users will be asking questions.
Generic AI Prompt Structure
Generic AI chatbots require more details to achieve good initial responses because they do not have a system message or specific task they have been designed for. Starting with a well structured prompt will provide a response that either provides what you are looking for or is easy to build off of.
Engineering prompts to feed into a generic AI can be broken down into the rough structure of:
"I want you to act as <role title> to do <task>. I am using <background info> and need the <output> to include <specifics>."
To develop your prompt, replace the following placeholders with the appropriate information:
-
<Role title> The position you want the AI to fulfill. Examples: data analyst, marketing director, technical writer, developer...etc.
-
<task> The task you want the AI to perform.
-
<background info> Provide context to the AI for what you are trying to accomplish. For example, if you want the AI to add functionality to an exisiting script you could provide the contents of a script you are currently using. Another use case would be if you are trying to have it provide possible solutions to a log error you might provide the system details, such as the OS, version and other applications involved.
-
<output> The result or output that you expect the AI to produce from your task request.
-
<specifics> Provide any restrictions or inclusions you want the AI to consider in it's response. For example, asking for the output of a script to be in python or requesting that the AI only provide summarized results from an analyzed table starting from the year 2021.
This is also the structure that many system messages start with before supplying supporting examples and any additional context the AI would need.
Purpose-specific AI Prompt Structure
System messages that purpose-specific AIs are built with generally contain the <Role title> and some <background info> that provides context for the AIs purpose. They may also include some specifics on how the <output> should be given. Users can then shorten their prompts to only include the task and any specifics about the background info or output they feel need to be added.
Users do not often have insight into what the system message for an AI bot is and generally the AI is not able to provide details about it's configuration for security reasons. Users are often provided the general purpose of the bot so they know what type of questions to ask it and what sources it has available.
Prompt structure for role-specific AIs built with a system message:
"Do <task>. I am using <background info> and need the <output> to include <specifics>."
Take an AI bot that has the following system message: "You are a human resources manager that answers questions from ABC Company employees about ABC company internal policies. In your responses you only provide factual information from the ABC Company HR employee handbook. You are kind, friendly and respectful in your responses."
Employees are then told: "Use the following AI Chatbot to ask questions about ABC Company internal policies. It will provide answers from information found within the company HR handbook. URL: [email protected]"
From the info, employees know that the chatbot has the <Role title> and <background info> of the HR handbook covered within the configuration. A sample employee prompt would be: "Give me a list of stat holidays the company is closed for this year. Include whether the closure day is a day "in leiu" of the actual stat holiday."
Prompting Past Initial AI Responses
Initial AI responses may get close to desired output and need subsequent prompts to refine it. Situations where you want to refine the initial response include wanting it to be in a different:
- format (ex. list, spreadsheet)
- coding language (ex. markdown, rich text, HTML)
- language (ex. French, Spanish)
- tone (ex. friendlier, encouraging)
As an example, look at the sample prompt an employee asked an HR AI bot: "Give me a list of stat holidays the company is closed for this year. Include whether the closure day is a day "in leiu" of the actual stat holiday." . The AI then replies with a horixontal list of dates with the words "in lieu" written next to any closure date that is not the actual stat holday date. This was not the desired format and the employee finds this difficult to read. They further prompt the AI: "Give me this list in vertical format with each date on a new line and make "in lieu" dates in bold instead of writing the words "in lieu".". Now the AI provides the employee with a vertical list of dates that they can easily read and identify the "in lieu" dates in bold quickly.
AI Use cases
Prompt ideas and examples for Bravura Cloud tasks and integrations where Bravura Security Fabric has been integrated with Bravura Cloud as the Identity Provider. In these use cases, the prompts are being entered into an AI bot that has been configured with uploaded Bravura Security Fabric and Bravura Cloud documentation.
How-to Prompts
-
Use Case: How do I target XYZ?
Prompt example: I want you to act as a systems integrator. I am using Bravura Cloud to bring in data from Bravura Security Fabric. How do I target an Active Directory system in Bravura Security Fabric and then pull the new data into Bravura Cloud?
-
Use Case: How do I list XYZ from my target?
Prompt example: I want you to act as a systems integrator. I am using Bravura Cloud to bring in data from Bravura Security Fabric. How do I list group memberships from my Linux target in Bravura Security Fabric and then pull the new group data into Bravura Cloud?
-
Use Case: How do I create a compliance rule that does XYZ?
Prompt example: I want you to act as a product administrator. Write me a compliance rule for Bravura Cloud that checks if all users have an assigned manager.
Investigative and Clarifying Prompts
-
Use Case: Explain what the following code is doing.
Prompt example: I want you to act as a technical support analyst for Bravura Cloud. Explain what the following rego expression from a Bravura Cloud compliance rule is doing: <compliance rule>. Include the explanation for each line of code inline.
-
Use Case: Summarize the following instructions into an action item list.
Prompt example: I want you to act as a project manager. Read the following email I received from a technical support agent for Bravura Cloud about XYZ. Summarize the action items that I need to complete from their suggestions.
-
Use Case: Where can I find information on XYZ?
Prompt Where can I find information about importing data from Bravura Security 12.4 into Bravura Cloud?
Troubleshooting Prompts
When asking AI to troubleshoot issue for you, the responses from the AI often include several areas and methods to use to identify where the problem lies. This provides an excellent starting point for solving issues.
-
Use Case: I am receiving the following error: XYZ. How do I fix it?
Prompt example: I want you to act as a technical support analyst. I am using Bravura Cloud to bring in Active Directory data from Bravura Security Fabric. My target ID for the Active Directory target is AD_CORPORATE. I am receiving the following error when testing the AD_CORPORATE target credentials: "Failed (Failed: Operation results missing for index [0].)". How do I fix it?
-
Use Case: I am trying to do XYZ, but it is not working. Why?
Prompt example: I want you to act as a technical support analyst. I am using Bravura Cloud to bring in Active Directory data from Bravura Security Fabric. When I try to view Active Directory users and groups in Bravura Cloud, I am not seeing any users or groups. Why?
-
Use Case: From the issue description and error logs I have provided, write me an email that I can send to Bravura Security support so that they can advise how to fix the problem. In the email I want you to include XYZ.
Prompt example From the issue description and error logs I have provided, write me an email that I can send to Bravura Security support so that they can advise how to fix the problem. In the email I want you to include the date and time the issue occured, what the issue is, the changes I have made to try and resolve the problem and the results of those changes. After providing the email to me, tell me which files and screenshots I should attach to the email to give the support agent appropriate context.