Type: url
Original: https://zenn.dev/pepabo/articles/claude-code-review-loop-zero-findings
Status: generated
Language: ja
Summary
A GMO Pepabo engineer describes an AI code review built in Claude Code from six specialised reviewers run in parallel, a validation agent that filters their findings, and an auto-fix loop meant to repeat until no findings remained, with safety valves such as minimal fixes, a 20-file scope limit and a five-round cap. The loop never reached zero findings; the author reports that AI review took over checklist-style review while design and intent remained human work, and that merge throughput rose after adoption.