Type: url
Original: https://blog.shibayu36.org/entry/2026/03/23/173000
Status: generated
Language: ja
Summary
The author describes a /self-review slash command for Claude Code that, once an implementation is done, runs three reviewer agents with different characteristics in parallel — a general-purpose reviewer, one that delegates to Codex CLI’s codex review command, and one focused on readability and over-engineering — and then has a /fix-review-comments skill critically evaluate each finding, fix only the ones it judges valid, and give a reason for each one it rejects. The approach, adapted from another developer’s idea of judging whether a review finding is appropriate before acting on it, is presented as a way to raise the quality of Claude Code’s output before the author reads the code himself.