AI Blocks › Set up AI to work for you
Change a prompt only when the tests say it is better
Give it a rough prompt. Get a clear version, scores on your tests and safe rules for changing it.
Tune a prompt with real tests, then keep versions is a free AI workflow for the job "Set up AI to work for you". It guides your AI through several skills in order, passing each step's output to the next.
Free · No account · Made by AI Blocks
What it does
A prompt that does real work gets edited again and again. Someone fixes one bad answer and breaks three good ones. Nobody notices for weeks, because no one kept the old version or a set of cases to check against. The steps fix that. First it rebuilds the rough prompt into a clear one. You then write 10 to 20 test cases with the answers you would accept, including cases where it failed before. The next step scores the rebuilt prompt and each change against those cases, one change at a time. You pick the best version and save it as version 1. The last step sets up names for versions, a check before each change, a way to try a change on part of the work and a rollback.
You give it
- The rough prompt
- What a good answer looks like and who reads it
- Examples of answers that went wrong (optional)
- Where the prompt lives and who edits it (optional)
You get
- A clear version of your prompt
- A score for each version against your test cases
- A winning version, saved with its tests
- Rules for naming versions, checking changes, trying them on part of the work and rolling back
Use it when
- A prompt gives good answers one day and weak ones the next.
- Several people edit the same prompt and you cannot tell which edit hurt.
- A prompt runs in a product or a weekly routine, and a bad answer costs you.
Steps
- Prompt improver: Give it a rough prompt. Get it rebuilt into a clear, structured prompt using a fitting framework.
- Prompt tester and tuner: Give it a prompt and test cases. Get scores for each version and a better prompt backed by results.
- Prompt change manager: Keep prompts in production. Get versioning, tests of changes, a registry and safe rollbacks.
How to use it
- Download the handoff pack. It's a zip with HANDOFF.md and every skill file this workflow uses.
- Give the folder to Claude (in Claude Code, or as files in a Claude Project) and say "Follow HANDOFF.md".
- Paste the rough prompt and say what a good answer is.
- When it pauses, write 10 to 20 test cases with the answer you would accept. Include two or three where it failed before.
- After the scores, save the best version as version 1 in a file, with the test cases beside it.