Prompt Regression Test Checklist
Check that edits did not break old behavior.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
Create a regression process for a prompt that changes over time. The prompt and its job: [PROMPT_AND_JOB]. Who edits it and how often: [EDITORS_AND_FREQUENCY]. The test cases I have: [EXISTING_TESTS]. Build a checklist with boxes: [ ] Every prompt change has a version number, date, author, and reason in a changelog [ ] A baseline output set exists for the test cases, stored with the version and tool settings [ ] After each edit, rerun all test cases and compare with the baseline [ ] Compare the format, key facts, tone, and length; flag any differences beyond a tolerance I define [ ] Review failures first, then differences that look like improvements (they can hide new risks) [ ] Re-run high-risk cases several times to check variation [ ] Check adversarial and out-of-scope cases again [ ] Record the model or tool version; provider updates can change behavior even if the prompt is unchanged, so rerun periodically [ ] A rollback plan: keep the previous version ready [ ] Approval step for prompts used in customer-facing or high-stakes work [ ] Monthly sample of real production outputs Provide a table template for results (case ID, pass or fail, notes), a rule for when a change needs a full retest, an example changelog entry, and a script outline for automating the runs if the tool has an interface. Note that some difference is normal and exact repeat is not guaranteed.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.