Free LLM test-case templates
6 ready-to-run prompt test suites (54 cases in total). Each one opens in the Litmiq app with its system prompt, scoring criteria and every test case filled in, so you can run it, then swap in your own facts.
- Free, no signup
- Loading a template only downloads a public file from litmiq.com
- Runs use your own OpenAI key, sent from your browser only to OpenAI
Single-turn prompt-injection suite with a canary (Lumen Ledger help assistant, fictional)
12 illustrative single-turn injection cases with benign controls and a canary-string method for spotting system prompt leaks.
JSON extraction suite (Pinecrest Supply order emails, fictional)
Test extraction prompts in three layers (parse, shape, values) with 10 illustrative cases and an offline checker script.
Expected-behavior writing practice suite (Maple Street Physio booking assistant, fictional)
A Must / Must not / Okay template for expected behavior that a person or an LLM judge can check, with 8 rewrites and a 6-case suite.
Customer support starter suite (Northwind Notes, fictional)
16 illustrative single-turn cases for a support system prompt, grouped by how support bots go wrong, plus a scoring rubric.
Prompt regression starter (Acme support bot, illustrative)
The 5 illustrative support-bot cases from the regression guide, with the Acme system prompt.
Judge sanity checks (support bot with a 30-day return policy, illustrative)
5 cases that check whether the judge's scores move the way they should, with the guide's example criteria.
How do these templates work?
Every template is an illustrative example built around a fictional product, written for one of our guides. It contains a system prompt, scoring criteria and single-turn test cases, each with an expected behavior. Opening a template replaces the setup in the app; if you've edited your setup, the app asks first and offers an undo afterwards. Nothing has been run for you: there are no outputs or scores in the files.
Before you run one, count the calls: each case runs once per selected model, and each output gets up to one judge call, all billed to your OpenAI key. Ten cases on two models is 20 model calls plus up to 20 judge calls.
Where to go next
Read how to write expected behavior before writing your own cases, the prompt regression testing guide for the re-run loop, or see all guides.