Tool 03 — live
PROMPT TESTING KIT Free · lint, diff & hallucination review
Four tools for the part of prompt engineering that is actually work: reviewing answers for claims worth checking, linting prompts for what they forgot, building test matrices and diffing versions. All of it by hand, none of it guesswork.
Hallucination review
Find the claims worth verifying
Paste an AI answer and it highlights the spans that are most often wrong — figures, dates, named entities, citations and absolute claims. It turns "read this carefully" into a specific list of things to check.
It flags shapes, not falsehoods. A highlighted number is a number, not an error. Verification is still yours.
THE AI ANSWER
REVIEW CHECKLIST
→ FLAGGED CLAIMS
Paste an answer and press Highlight claims.
Robustness checklist
A linter for prompts
Fourteen static checks for the things prompts usually forget — output format, length limits, audience, what to do when the model does not know. Every check is a rule over the text, so it is instant and explains itself.
A pass means the check found the words it looks for, not that the prompt is good. Treat it as a reminder list, not a verdict.
YOUR PROMPT
→ RESULTS
Paste a prompt and press Run checks.
Test case generator & matrix
One template, every combination
Write a prompt with {{variables}}, give each variable a
list of values, and get every combination as a filled-in prompt ready
to paste into a model. Export the lot as CSV for a spreadsheet or
JSON for a test harness.
TEMPLATE
VARIABLES
Variables you use in the template appear here automatically.
→ TEST MATRIX
Fill the template and press Build matrix.
A/B and version comparator
See exactly what changed
A word-level diff of two prompts, with the numbers that usually matter — length, tokens, instruction count, sections. Save versions as you iterate and diff any two of them later.
It shows you the difference. It cannot tell you which is better — that needs running both, which is what the Scorecard tab is for.
VERSION A
VERSION B
→ DIFFERENCES
Paste two prompts and press Compare.
SAVED VERSIONS
Stored in this browser only. Nothing is uploaded.
WHAT THIS KIT CANNOT DO
It cannot run your prompt against a model. Every tool here produces, organises, compares or scores — the running and the judging stay with you and whatever model you use. That is the trade for being instant, deterministic, free, and never sending your prompts anywhere.