Template library · 20 suites
20 prompt-eval templates with ready-made tests
On this page
The short answer
- 20 templates, each with ready-made cases and a one-click Litmiq link. Eleven were already published; nine are new as of 19 Nov 2026.
- Every suite is illustrative: fictional brands and made-up facts, no model outputs or scores.
- The earlier prompt eval templates list and the original templates page stay published.
Templates
- Template
Support bot
Refunds, shipping, unknowns and tone for a help-desk prompt.
Open the template → - Template
RAG assistant
Groundedness cases with a snapshot of retrieved context.
Open the template → - Template
Invoice extraction
Pull fields from fictional invoices into a fixed shape.
Open the template → - Template
Safety pack
Injection, over-refusal and disallowed requests.
Open the template → - Template
Ticket classifier
Assign a short, closed set of IT ticket labels.
Open the template → - Template
Meeting notes
Short summaries with action items, not invented decisions.
Open the template → - Template
Tool calling
When to call a tool, and when to ask a clarifying question.
Open the template → - Template
Instruction following
Format, length and constraint checks.
Open the template → - Template
Brand voice
Tone and wording rules for a fictional brand.
Open the template → - Template
Multilingual support
The same policy answers across languages.
Open the template → - Template
Structured output
Valid programme records in a fixed schema.
Open the template → - New template
SQL generation
Six cases for a fictional warehouse: simple filters, joins, a refused destructive statement, an unknown table, a date window, and a limit on row counts.
Open the template → - New template
Email replies
Six inbound emails for a fictional boat-rental shop: a booking change, a refund policy, an unknown fee, an angry tone, a prompt-injection in the email body, and a request the shop cannot do.
Open the template → - New template
Code review
Six review requests: a real bug, a style-only change, a secrets leak, an unknown file, a request to rubber-stamp, and a performance note with no invented benchmarks.
Open the template → - New template
PII redaction
Six fictional snippets: an email, a phone number, a card number, a clean sentence, a mixed paragraph, and a request to keep the email visible.
Open the template → - New template
Product copy
Six copy tasks: a complete fact sheet, a missing allergen, a banned health claim, a length cap, a competitor comparison, and an empty fact sheet.
Open the template → - New template
Translation
Six short strings: a simple sentence, a product name that must stay in English, a missing target language, a number and date, an instruction to add marketing, and an empty source.
Open the template → - New template
Review sentiment
Six fictional reviews: clearly positive, clearly negative, mixed, empty, a prompt-injection, and a review in another language.
Open the template → - New template
SEO meta tags
Six pages: a complete brief, a missing product name, a request to stuff keywords, a character overflow, a competitor claim, and a request to invent a customer quote.
Open the template → - New template
Release notes
Six note-writing tasks: a clean list of commits, a missing version, an internal-only change, an invented customer quote, a length cap, and a security fix that must not over-share.
Open the template →
How they work
Each template page lists the system prompt, the scoring criteria and the cases. The Run in Litmiq button opens /app?suite=/assets/suites/….json. The app fetches that file from this site, shows a preview, and loads it only after you confirm. Your API key stays in your browser and is sent only to the provider you choose.
Suites are licensed CC BY 4.0. Credit Litmiq and link back to the template page.
FAQ
Are these official vendor templates?
No. They are illustrative Litmiq suites with fictional products and made-up facts. They are free to adapt under CC BY 4.0, with credit. No scores are included.
How is this different from /prompt-eval-templates?
The earlier library has 11 use-case pages. This library lists all 20: those 11, plus 9 new suites published on 19 Nov 2026. Both libraries stay up.
Do the tests include model outputs?
No. Each suite has a system prompt, scoring criteria and expected behavior for every case. You supply your own API key in the Litmiq app to produce outputs.
Can I use these in CI?
You can export a Promptfoo YAML from the app after you load a suite. A full Litmiq CLI is a separate product stream and is not claimed on this page.
Check a prompt first
If you are not ready for a full suite, paste a system prompt into the no-signup checker.
- Free, no signup
- Runs in your browser
- Your OpenAI key, sent only to OpenAI