Now Assist Skill Kit Overview
In Chapter 1, I introduced the Skill Kit as your toolbox for building custom capabilities. Now we're getting into the specifics. How does it actually work? What can you build with it? What are the key concepts you need to understand?
The Skill Kit was released in the Xanadu release and has been evolving rapidly since. It's designed for scenarios where out of the box skills don't quite fit your needs, whether that's different formatting requirements, organisation specific use cases, or integration with external language models.
What You Can Build
The Skill Kit lets you create custom generative AI skills that integrate directly into ServiceNow workflows. These aren't isolated experiments. They become first class capabilities that users can trigger just like any out of the box skill.
Common use cases include augmenting existing workflows with AI functionality, creating organisation specific skills that out of the box options don't cover, and leveraging external language models when NowLLM doesn't meet your requirements.
I've seen teams build skills that categorise knowledge articles automatically based on their content. Others analyse expense reports against company policies. Some generate customer communications in their organisation's specific tone and style. The possibilities depend on your creativity and your organisation's needs.
The Building Blocks
Every custom skill has core components that work together.
Skill inputs define what data flows into your skill. These are the contextual variables that get injected into your prompt. You might pull the incident short description, work notes, priority, or any other field relevant to your use case. The Skill Kit supports multiple data types including Record, String, Numeric, Boolean, arrays, and JSON objects.
Prompt templates contain the actual instructions sent to the language model. This is where your prompt engineering skills matter most. You write the instructions, reference your inputs using variables, and define what output you expect.
Usage conditions determine when different prompts execute. A single skill can have multiple prompt templates, with conditions controlling which one runs based on the context. If no conditions match, the default prompt executes.
Model settings specify which language model provider handles the request and how it behaves. You select the provider (NowLLM, Azure OpenAI, or others) and configure parameters like temperature. For enterprise use cases, keep temperature low (around 0.2) to ensure predictable, factual outputs.
Tools for Context Enrichment
Skills often need more than just record fields. The Skill Kit provides tools to gather additional context before the prompt executes.
Retrievers use the RAG framework to search your instance and surface relevant content. If you're building a policy compliance skill, a retriever can dynamically find the applicable policies based on the input record's characteristics.
Subflows let you gather data through Flow Designer processes. Create a modular subflow that collects information, then use its output in your prompt.
Scripts give you flexibility when standard tools don't fit. Write custom logic to gather or transform data before it reaches the prompt.
Flow Actions connect to your existing workflow building blocks, letting you reuse logic you've already built.
These tools can run in parallel or in series depending on your needs. You can even chain skills together, using one skill's output as another's input.
Deployment Options
Once you've built and tested a skill, you need to get it to users. The Skill Kit provides multiple deployment paths.
UI Actions trigger skills from buttons on forms. Click a button, the skill executes, and the output appears. Simple and direct.
Now Assist Panel makes skills available through the conversational interface that fulfillers access from the banner. The panel uses your skill's title and description to decide when it's relevant, so name them clearly.
Virtual Agent surfaces skills to requestors through Employee Centre or Service Portals. This is how end users interact with your custom AI capabilities.
Flow Actions let you embed skills within larger workflow automations. The skill becomes a step in a broader process.
AI Agents can use custom skills as tools when resolving scenarios. Your skill becomes a building block that agents draw on when they need its specific capability.
UI Builder allows direct integration into custom workspaces and interfaces. The skill output appears exactly where and how you design it.
The Development Workflow
Building a skill follows a predictable pattern.
Start by creating a new skill and configuring model settings. Navigate to All > Now Assist Skill Kit > Home to begin. Name your skill clearly because that name helps users and the Now Assist Panel understand what it does.
Add skill inputs for the data your prompt needs. Each input requires a description that tells the language model what the field represents and how to interpret it. Better descriptions mean better context for the AI.
Write your prompt template. Reference inputs using variables. Be clear, specific, and follow prompt engineering best practices. I'll cover those in detail in Part II of this guide.
Test extensively. The Skill Kit provides testing capabilities that show you the grounded prompt (what actually goes to the LLM after all tools and inputs are assembled) and individual tool outputs. Use these to debug issues.
Publish when you're satisfied. Then activate through the Now Assist Admin console and configure deployment options.
Key Configuration Considerations
A few settings deserve extra attention.
Mandatory inputs ensure the skill fails gracefully if required data is missing. Better to get a clear error than a hallucinated response based on missing context.
Truncation handles large text fields that might exceed token limits. Enable truncation for fields like work notes that can grow very long. Without it, oversized inputs cause execution failures.
Field descriptions matter more than you might expect. The language model uses these descriptions to understand context. Invest time in writing clear, accurate descriptions for each input field.
Where This Fits in Your Journey
The Skill Kit is where prompt engineering theory becomes practice. Everything you learn about crafting effective prompts gets applied here.
In Part II, I'll cover prompt engineering fundamentals in depth. By the time you've worked through those chapters, you'll understand exactly how to design prompts that work well within the Skill Kit framework.
For now, understand that the Skill Kit is your primary workspace for custom AI development in ServiceNow. Master it, and you can extend Now Assist to handle virtually any use case your organisation needs.
Right, one more architectural component to cover. Let's look at how Now Assist integrates with the broader ServiceNow platform.
Last updated