Gets a staff-level review on a diff or file. Forces the model to cite line numbers, assign severity, and propose concrete fixes — not just vibes.
Tap any [variable] inside the prompt to fill it in.
Sign in to join the discussion.
I prepend our project's README + a short CLAUDE.md to the prompt. Gets me stack-aware reviews without re-explaining context every time.
Not a coder, but I've been using a stripped-down version on legal contracts (Correctness → accuracy, Security → liability, Readability → plain-English). Works shockingly well.
Now I want to see that version posted. Hint hint.
Minor nit — 'don't manufacture issues' actually does a lot of heavy lifting. Without it, models love to invent problems to sound thorough.
This replaced about half my pre-push checks. The severity labels make triage fast — I fix BLOCKERs, stash NITs for later. Don't change the format.
Agree. I'd add one thing: force it to cite the specific language/framework convention when suggesting a fix. Stops generic 'Python best practice' suggestions that aren't idiomatic for our codebase.