Regression Test Suite for a Production Prompt

Build a regression suite for this prompt before it is changed.

Current prompt: <prompt> [Paste] </prompt>

What it should do: [description] Known past failures: [if any]

Generate 15 test cases:

• 4 baseline cases — typical input, clearly expected output.
• 4 edge cases — empty input, very long input, missing fields, mixed Hebrew-English text.
• 3 failure cases — input that should produce a refusal or a defined error message.
• 2 injection cases — input containing instructions attempting to override the prompt ("ignore previous instructions").
• 2 uncertainty cases — input where the correct answer is "insufficient information".

For each case: { "id": "", "input": "", "assertion": "an automatically runnable check — not 'the answer is good'", "must_not_contain": ["strings whose presence is a failure"], "why": "which regression this case catches" }

Assertion rules:

• Prefer deterministic checks: JSON validity, field presence, value range, substring presence or absence.
• For subjective judgement, write an explicit binary criterion for a judge model.
• Every assertion must fail if the prompt were removed entirely. An assertion that passes without the prompt tests nothing.

Finally: which metric measures overall success and what the acceptance threshold is.