Prompting can sometimes be.....frustrating? It's too vague, too verbose or just not what you meant. While building an AI capability (A skill) prompting can change the outputs you receive.
Writing prompts that work for you isn't magic - it's structure.
At Knowledge 2025 we tackled the art of crafting good prompts with a hands-on workshop where you can use actionable frameworks to build any kind of skill you want in skill kit.
With this session participants learned how to
- Breakdown vague prompt asks into sharp and actionable prompts
- Use different structures to craft different types of prompts to build a skill
- Use frameworks that can producible repeatable patterns
- 📌 Bonus: I also share a real example and show you how these prompt frameworks can alter the outputs you receive
Why prompting is a skill (Not a hack)
- Prompt engineering sounds complex, but at its core, it's just structured communication.
- We prompt every day—whether it's writing emails, asking search engines, or delegating tasks.
- While prompting or interacting with LLMs you need context, clarity and specificity to get the outcomes you desire.
How prompt engineering enables better skill creation
- A skill is a predefined area of expertise that AI has been tailored for. It can execute that specific need.
- A skill can be - email recommendations, KB generation, Resolution notes generation etc
- Prompt engineering helps to tailor the LLMs response to that specific need.
How do you take a prompt from vague to a more desirable output
What does a vague prompt look like?
Help me add my expenses for Knowledge 2025
How would we improve this prompt?
Framework 1 : Make it specific, provide context, create constrains
Design an expense tracker for my Knowledge 2025 business trip. The trip includes flights, hotel stays, meals, conference registration, and other miscellaneous costs. Travel budget is $400, Accommodation budget is $500, Meals $200, Registration $250, Miscellaneous $100. The tracker should categorize each expense, calculate total spend per category, and flag categories that go over budget.
Framework 2 : GUIDE / Goal ,User, Instruction, Details, Example
I’m an employee, I’ll give you a list of conference expenses. Categorize each expense into Travel, Accommodation, Meals, Registration, or Miscellaneous. Show if I stayed within budget for each category. Travel budget is $400, Accommodation is $500, Meals is $200, Registration is $250, and Miscellaneous is $100. Format each expense as: [Item] - [$Amount] - [Category] ([Budget Status]).Also, highlight if I went over budget, for example, Flight - $450 - Travel (Over budget by $50).
Framework 3: RASCEF / Role, Action, Steps, Context, Example, Format
Track and categorize your expenses for a recent event or trip, and identify whether any category has exceeded the allocated budget. You are managing the expenses for a business conference you attended recently. You have the following categories: Travel, Accommodation, Meals, Registration, Miscellaneous.
Here's what you have to review
1. Check list of expenses and their amounts.
2. Check if each category exceeds the budgeted amount.
3. Output the result for each category: if the expense is over budget, mark it with 🚨 and the amount exceeded; if within budget, mark it with ✅.
The goal is to track the total amount spent in each category and ensure none of the categories go over budget. Each expense needs to be categorized accordingly, and the result needs to be displayed in the format:
<Item Name> - $<Amount> - <Category> (<Status>)
Flight - $450 - Travel (🚨 Over budget by $50)
Hotel - $550 - Accommodation (🚨 Over budget by $50)
Dinner - $75 - Meals (✅ Within budget)
Conference fee - $200 - Registration (✅ Within budget)
Coffee - $15 - Miscellaneous (✅ Within budget)
These prompts are all crafted for the same use-case, however they've been worded differently and hence yield different results when they've been tested.
Do you need to craft different version of a prompt for the same use-case?
Crafting different versions of a prompt for a single skill is essential because goals, context, and desired outputs often vary. Whether you're tracking expenses for a business trip, summarizing monthly spending, or preparing data for a report, each use case demands a tailored prompt. Varying the level of detail, output format, and user perspective helps the AI deliver more accurate, useful results. Prompt iteration isn't redundancy—it’s refinement.
What do you once you have different versions of a prompt?
Evaluate.
Within the Skill Kit, after crafting a custom skill, you can utilize the Evaluations tab to assess its effectiveness. This feature allows you to run your skill against a dataset of any size, providing insights into how well your prompt performs across different scenarios. The evaluation focuses on two key metrics:
-
Faithfulness: Determines if the AI's output remains true to the source material.
-
Correctness: Assesses whether the output accurately responds to the input instructions.
Why should you evaluate?
- Accuracy:
Is the AI's response is factually correct, relevant, and aligned with the intended goal of the prompt?
- Consistency:
Does it behave similarly with new inputs?Evaluates how reliably the AI responds to similar prompts.
- Reduced hallucinations:
When AI gives information that is wrong, but it presents that information confidently as if it’s true.
- Predictability:
Trustworthiness of the AI's response. Can expect the AI to provide reliable and valid results?
Take a look at the session and prompt engineer your skills in ServiceNow using any of the frameworks above. We also have a handout with frameworks you can use. .Join the conversation and share how you prompt, what frameworks you use!
