Claude Code prompt-regression-tester for prompt version comparison

Compare old and new prompts against identical fixtures for format, required information, omissions, and regressions.

Install in Claude Code

Use Claude Code with the old and new prompts, replayable inputs, and observable output constraints. Repeat runs when measuring stability.

Review `.claude/skills/prompt-regression-tester/SKILL.md` and its requirements at the pinned GitHub commit.
Create `.claude/skills/prompt-regression-tester/` in your project and save the reviewed file as `.claude/skills/prompt-regression-tester/SKILL.md`.
Start Claude Code from that project's root, then select `/prompt-regression-tester` in the slash menu.

Example task

Case prompt-21 says “Move the garden tour to Friday”. The field day must be Friday. The old output is {day: 'Friday'} and the new output is {day: 'Thursday'}.

What to check

Report {verdict: 'regression', expectedDay: 'Friday', actualDay: 'Thursday'} for the case. Extract the new date from the input, keep the output format, and rerun the fixture.

Original SKILL.md on GitHub · Updated 2026-11-14

Frequently asked question

Does one comparison establish prompt stability?

One fixture can show the observed difference. Reuse the same cases across repeated runs to assess consistency of format and behavior.