What Content diff does
A dependable text-diff tool with an optional review on top. The script finds the differences, the AI explains what they mean, and the two are never mixed. That split is the whole point: a model reading two documents and noticing what changed will miss things, and you will not know which things.
What you give it
Two versions of any text: files on disk, or pasted snippets labelled ORIGINAL and UPDATED.
What you get back
- An exact diff: side-by-side or unified, word, character or smart precision
- A numbered change index with counts taken from the script, never estimated
- A separate section reviewing which changes alter meaning and deserve attention
- An accept/reject merge written to a new file, never over either source
Use cases
Checking a client's edits to approved copy
Find out exactly which words moved between the version that was signed off and the version that came back.
Comparing HTML, JSON, YAML or schema snippets
Structured text compares cleanly, and ignore rules can hide timestamps or whitespace noise without touching the stored text.
Confidential documents
Nothing is sent to an online diff site or a search engine. Client, patient and contract content stays on your machine.
Contracts and policies
The change index tells you what changed; the review section tells you which of those changes shift the meaning.
How to use Content diff
Install the productivity toolkit
/plugin install productivity-toolkit@claude-skills
Give it both versions
Point at two files or paste both texts, labelled original and updated. Say "compare these", "what changed", or run /content-diff.
Pick options if you need them
Ignore whitespace, ignore case, ignore blank lines, ignore a pattern such as a timestamp, hide unchanged lines, or switch between word, character and smart precision.
Follow up
Next change, previous change, swap the sides, hide unchanged, or merge. Each follow-up re-runs the script rather than editing an old result by hand.
Benefits
Deterministic, not remembered
Differences come from a Python script every time. If the script cannot run, the skill says the result is not deterministic instead of quietly guessing.
Your text is never rewritten
Neither input is corrected, trimmed or re-wrapped, and neither source file is overwritten. Ignore rules affect the comparison only.
Interpretation is labelled
The exact diff and the AI analysis are kept in separate sections, and nothing is called an error just because it changed.
Stays local
No external service sees the content unless you explicitly allow it.
Install content-diff
This skill ships in the productivity-toolkit plugin. Add the marketplace once, then install the plugin in Claude Code.
/plugin marketplace add rodneylungayan/claude-skills
/plugin install productivity-toolkit@claude-skillsAfter that the skill loads on its own whenever your request matches it. You can also invoke it by name with /content-diff.
- Needs
- Python 3.8+ (standard library only). Without it the skill falls back to git diff and tells you the result is not deterministic.
Questions about content-diff
- Why run a script instead of letting the model read both versions?
- Because a model reading two documents will miss changes, and you have no way to tell which ones. The script finds every difference and reports exact counts; the model is only allowed to explain them.
- Does it send my documents anywhere?
- No. It will not send the content to a web search, an online diff site or any other API unless you explicitly allow it.