Prompt version diff
Prompts rarely change all at once — they drift through dozens of small edits, and any one of them can move behaviour. When you are A/B testing or chasing a regression, the first question is always what exactly changed? This tool answers it with a clean, word-level diff between two prompt versions.
How it works
The diff tokenises both versions into words and computes the longest common
subsequence — the standard algorithm behind tools like git diff. Words present
only in the old version are marked as deletions (struck through, red); words only
in the new version are additions (green); shared words stay plain. Whitespace is
normalised so that reflowing a paragraph doesn’t register as a thousand changes.
Alongside the highlighted view, a short summary reports the word count of each version and the net change, so you can see at a glance whether an edit tightened or expanded the prompt. All computation happens locally in your browser.
Tips and notes
When debugging a behaviour change, diff the exact two prompt strings your app sent — not your mental model of them. Small things like a removed “only”, a swapped verb, or a deleted example are easy to miss by eye but jump out in the diff. Pair this with the system-prompt linter: diff to see what changed, lint to check whether the change introduced a new structural problem. Keep your prompt versions in source control too — the diff here is for quick comparison, not permanent history.