OpenDiffs

Structured code review from the terminal, powered by your existing AI agents. No new accounts, no API keys, just your Claude or OpenAI subscription.

npm install -g opendiffs

Requires Node.js 18+ and Claude Code or Codex.

Review your staged changes or any individual file before you commit. Your agent reads the diffs, explores your codebase for context, and gives you a structured review with a Diffs Score.

How it works

Stage your changes, then run opendiffs:

opendiffs

From there, review all your staged changes or pick any individual file.

Your agent reads the diffs and checks your codebase. You get back a markdown review scored 1-10 with findings.

Note: your agents are doing real work — this will use your plans.

Reviews

Reviews are saved as markdown in .opendiffs/reviews/, organized by branch. Browse them anytime:

opendiffs --reviews

.opendiffs/ is automatically added to .gitignore.

Settings

opendiffs --settings

Setting	Options	Default
Providers	claude, codex, or both in parallel	claude
Save reviews	always, staged-only, never	always
Max reviews	any number	50
Review prompt	default or custom .opendiffs/prompt.md	default

Custom prompt

The default prompt tells your agent what to look for and how to score. To change it, go to opendiffs --settings, select "Review prompt", then select "Create .opendiffs/prompt.md". Once that file exists, your agent will use it instead of the default. Any edits you make to that file will be picked up on the next review.

CLI

opendiffs                    Interactive — pick what to review
opendiffs --staged           Review staged changes directly
opendiffs <file>             Review a specific file
opendiffs --provider codex   Pick your agent (or claude,codex for both)
opendiffs --settings         Configure providers, prompt, reviews
opendiffs --reviews          Browse saved reviews

License

MIT