SanjivMeher
Mega Patron
Mega Patron

This Community Week Livestream will begin shortly! (11).png

 

Welcome to our Community Week MVP Insights series! Join ServiceNow MVP Sanjiv Meher as he walks through how you can unlock Now Assist custom skills.


 

Small Skill, Big Impact: Unlocking Now Assist Custom Skills


In today’s AI era, it’s critical to leverage artificial intelligence to optimize our work. ServiceNow provides a strong foundation—with Now Assist and Agentic AI—to simplify our customers’ experience. Today I’ll spotlight a small but powerful capability in the Now Assist Skill Kit: custom skills. Out-of-the-box skills cover many common scenarios, but when you need organization-specific behavior, custom skills let you design and deploy exactly what your users need.

 


My Use Case


Consider an example: I aim to simplify the daily tasks of my cybersecurity analysts. They are currently overwhelmed with hundreds of phishing requests, dedicating hours to analyze each one and determine its legitimacy.


My goal is to implement AI to analyze every incoming phishing email request. This will identify whether the email is legitimate or a phishing attempt, saving my analysts hundreds of hours. This will allow them to focus on critical security incidents instead.

 


Creating a Custom Skill


The first step is to create a custom skill. To do that navigate to Now Assist Skill Kit -> Home

 

jordancbaron_0-1762545922415.png

 

In the Custom Skills Tab, Click on Create skill button.

 

Note: You can also create custom skills by copying existing out-of-the-box (OOB) skills and modifying them to suit your organization's specific requirements.

 

jordancbaron_1-1762545979333.png

 

To get a skill set up, you'll need a few key things.

 

Inputs: This section is used to provide input to your prompt, such as a record, strings, array etc. For example, you might input a phishing request here.

 

Tools: Before proceeding to the prompts, you can perform additional processing in this section. Options include using a script, a flow designer, an AI search etc.

 

Prompts: Select the model and craft your prompt here. ServiceNow supports providers like Now LLM, OpenAI, and Gemini. Clear, well-scoped prompts produce better results.

 

Outputs: This parameter is where you will receive the results of your process.

 

 

Skill Inputs and Outputs

 

Provide a skill name and a brief description about the skill.

 

jordancbaron_2-1762546066716.png

 

You can add input using the Add button in the Skill Inputs and Outputs section.

 

jordancbaron_3-1762546097557.png

 

In my case, I want to pass the phishing request as an Input. So I will select the data type as Record.

 

jordancbaron_4-1762546137541.png

 

For Output, we don't need to add any additional output. The default output can be used.

 

 

Tool Editor

 

You can add tools which can be used for any pre-processing before the prompt is executed.

 

jordancbaron_5-1762546187416.png

 

There are several tools, but in my case, I am going to use a Flow Action.

 

I'd like to integrate an existing flow action. This action takes a phishing request sys_id as input, then extracts the corresponding phishing email, returning both the email body and header as output.

 

Here I have an existing Flow Action “Get Phishing Email” which I can use in the Tool.

 

jordancbaron_6-1762546237217.png

 

jordancbaron_7-1762546265703.png

 

 

Prompt Generation

 

The next step involves setting up a prompt. You have the option to either create a prompt from scratch or utilize an AI-generated prompt, which can be modified later as needed.

 

jordancbaron_8-1762546312476.png

 

You can add form inputs that were received as output from the previous state's flow action to your prompt. In my case I will use the Email Body and Email Header which I received as output from my Flow Action.

 

jordancbaron_9-1762546347746.png

 

 

Testing your prompts

 

After the prompt is prepared, you can save it and run a test to confirm the output is correct using the "Run test" option.

 

jordancbaron_10-1762546393056.png

 

Review the prompt results, then refine the prompt to improve them. Once everything looks good, you can finalize the prompt. You can also instruct the prompt to return the response in JSON so that you can easily parse the response and extract the verdict and use it to take decisions.

 

jordancbaron_11-1762546427502.png

 

 

Deployment Settings

 

The Deployment Setting allows you to specify the product to which the skill is assigned and where it will be utilized.

 

First select the workflow and product where this skill will be listed.

 

jordancbaron_12-1762546478443.png

 

Next, choose where this skill will be utilized, such as in the Now Assist Panel, a UI action, or a Flow Designer.

 

jordancbaron_13-1762546509758.png

 

We will use this within a flow designer, as our goal is to automatically trigger it upon the creation of a Phishing request.

 

Once all the above steps are completed, you need to publish it by using the Publish button on the skill.

 

jordancbaron_14-1762546550619.png

 

 

Activate Skill

 

To activate the skill navigate to Now Assist Admin->Skills and from Technology select the product you added your skill to. This steps mayn’t be required, but if you don’t find your skill listed in the flow, the skill may not be active.

 

jordancbaron_15-1762546615523.png

 

Use the Activate Skill button to activate your custom skill.

 

jordancbaron_16-1762546644871.png

 

 

Use the skill in a Flow

 

You can then use the skill in your Flow.

 

To use the skill in the flow, use the “Execute Skill” flow action which is an OOB action. Select all the inputs and then in the Skill Config, select the custom skill you created. In the Dynamic Inputs, we will pass the phishing Request.

 

jordancbaron_17-1762546692315.png

 

 

 

Capture the result in worknotes

 

You can then check if the Execute skill status was a success, and then proceed to update the worknotes. The response is generated in a JSON format. So you can use fd_data.<execute_skill>.output.response to extract to response and then update the record with details.

 

jordancbaron_18-1762546769820.png

 

I can even use the result to notify the cyber security team immediately to take action. The output from this can also be used in AI agents.

 

 

Conclusion

 

Already thinking about where to use it? Focus on high-value scenarios and this compact feature can be a game changer. If we target the right use cases, this powerful yet simple feature can create big wins—saving your customers meaningful time.

 


Questions for Sanjiv? Leave them in the comments! Mark this post as helpful if you found it as such!

 

Connect with Sanjiv here!