Scoping the skill

  • Release version: Australia
  • Updated March 12, 2026
  • 1 minute to read
  • Scoping the skill before you build it helps determine the requirements needed for the skill and the expected outcomes.

    Before you begin

    Before you start to build a custom skill with Now Assist Skill Kit, you should first review the base system skills that are available in the Now Assist Admin console. Use these pre-existing skills whenever possible.

    User prerequisites

    Now Assist Skill Kit is meant to be used by a developer or someone with experience using generative AI. Because you must write the initial prompt template, you should:
    1. Be knowledgeable about prompt engineering, including:
      • Having familiarity with the development and testing of machine learning systems.
      • Having informed expectations about the behavior and capabilities of a prompted large language model (LLM). You should understand:
        • The fundamentally probabilistic nature of LLMs.
        • That performance can vary greatly across different LLMs and different target tasks.
      • Having experience or training with writing and evaluating prompts for LLMs.
    2. Have an in-depth understanding of the use case that they’re trying to solve and the persona that they’re building the skill for.

    Design

    Before you begin building a custom skill, you must think about the overall design and document the requirements. Consider the following:
    1. What precisely do you want the skill to do?
      • What will be the inputs to the model?
      • What should be the outputs of the model? (Both content and format)
      • Who will be using the skill?
      • How can you characterize success?
    2. What don't you want the model to do?
      • It’s useful to think about the possible risks and downsides of using generative AI outputs so that you can try to guard against them during development and testing.
      • It’s worth listing any specific model behaviors (LLM outputs) that would be detrimental or harmful in your specific use case.