This is basically a protocol wrapper that stops you from saying "great point!" to code review feedback and forces actual verification before claiming anything works. It has three distinct modes: receiving feedback (ask questions, push back if needed, skip the pleasantries), requesting reviews (dispatch a code-reviewer subagent after each task with git SHAs), and verification gates (run the actual command, read the output, then make your claim). The iron law framing is a bit dramatic but honestly needed if you're working with subagents that love to hallucinate test results. Most useful in subagent-driven workflows where false completion claims will cascade into bigger problems.
npx skills add https://github.com/mrgoonie/claudekit-skills --skill code-review