Caveman Review

It strips code review down to location, problem, fix. One line per finding. You get `L42: 🔴 bug: user can be null after .find(). Add guard before .email.` instead of "I noticed that you might want to consider..." It keeps severity prefixes when issues are mixed (bug/risk/nit/question), drops all hedging and restating, and knows when to break its own rules for security issues or onboarding contexts. Invoke it with "/review" or "/caveman-review" and paste the output straight into GitHub. It won't write the fix or approve the PR, just give you the actionable signal without the noise.