Stop guessing what your CLAUDE.md does. Test it before you commit it.
A sourced library of CLAUDE.md dos and don'ts, plus a browser tester: paste two drafts and a task prompt, run both on your own API key, compare results.
Sourced dos and don'ts
Every rule links back to the thread, changelog, or issue it came from.
Side by side testing
Paste two CLAUDE.md drafts and a task prompt, then run both against your own API key.
Runs in your browser
Your API key and prompts stay in the tab. There is no backend and nothing to sign up for.
How it works
Paste two CLAUDE.md drafts
Drop in your current file and the version you want to test.
Add a task prompt
Type the task your team actually runs, the one the rules are meant to affect.
Compare the outputs
Run both drafts against your own API key and see exactly what changed.
Before / After
- Copy CLAUDE.md tips from Reddit threads that contradict each other
- Rules go stale the moment a model update changes behavior
- Ship a change and find out days later it broke something
- Pull dos and don'ts that link to a source and a date
- Check which rules still hold after a model update
- Test a draft against a real task before you commit it
FAQ
Do I need to create an account?
No, Rulebench runs in your browser and only needs your own Anthropic API key to run the tests. What you paste is never sent to a server we control.
Where do the pattern library rules come from?
Each entry links to its source, a GitHub issue, a changelog note, or a documented test case, so you can check whether it still applies to your model version.
Can I test any CLAUDE.md, not just a template?
Yes. Paste any two drafts and a task prompt and Rulebench runs both through the same model so the comparison is fair.