AI Evaluation Prompts

You are testing the installed ServiceNow CLI as a customer would use it. Work in a fresh durable project workspace for this prompt unless I explicitly say it depends on a previous prompt. Keep reviewable artifacts in repo paths such as specs/, manifests/, samples/, reports/, and evidence/; use .tmp only for disposable scratch. Discover how to initialize and use the product from the installed CLI, generated project files, and normal help output. Do not inspect the product source repository or ask for hidden internal instructions. Build reviewable local artifacts, make live changes only when you have enough confidence, prove the requested business behavior, and leave a concise report under reports/customer-runs/<prompt-id>/ with the evidence and any unresolved gaps.

Build an employee equipment request item in ServiceNow. The requester should choose laptop type, peripherals, needed-by date, delivery location, and business justification. Expensive requests should go through approval before the IT hardware team gets fulfillment work. Give me evidence that a realistic request can be submitted and that the right follow-up work is created.

Create a Service Catalog request for access to our finance analytics tool. It should collect manager, cost center, access level, and justification. The request should go to the manager first, then to the finance data owner, and only then create work for the access team. Show that the request path behaves correctly.

Use the finance analytics access request from the previous prompt. The finance team changed the approval wording and wants admin-level access to go through a separate data owner before fulfillment. Update the existing work in a controlled way without deleting and rebuilding the request item. Show what changed and prove the updated route works.

Assume one of the request artifacts from the previous work was edited directly on the ServiceNow instance by an admin who changed customer-facing copy. Help me find what changed, show me the local value versus the live value, and either restore the intended source value or explain how to keep the live edit in source control.

Create a small import process for office locations. The incoming rows have a location code, name, region, manager email, and active flag. It should update an existing office when the location code already exists. Also account for the fact that the source feed recently renamed the incoming code column, and show that a rerun updates existing offices instead of creating duplicates.

We have an outbound vendor status integration that moved to a new base URL and now uses a different header name for its API key. Help me update the ServiceNow setup in a controlled way, keep secrets out of source files, and show that a reusable action or flow can still check one vendor by ID.

Create a custom support queue table and a support coordinator role. Coordinators should be able to read and update queue records. Normal users should not be able to change those records. Include proof that the allowed and denied access paths behave as intended.

Build an automation for new critical incidents. When a critical incident is created, it should create follow-up work for the major incident team and record what it did somewhere an admin can review later. Give me evidence that the automation only affects critical incidents.

I already have a small ServiceNow customization in my test instance, but I do not know exactly how it was built. Help me bring it under source control, explain what it does, and make one safe improvement that I can review before it changes live behavior.

Create a new hire workspace bundle request. The requester should provide employee type, start date, department, shipping address, and whether a phone is needed. The process should coordinate the right follow-up work for facilities and IT without turning into a one-off script that is hard to review.