Type: url
Original: https://developers.cyberagent.co.jp/blog/archives/62639/
Status: generated
Summary
A CyberAgent frontend engineer describes a GitHub Actions workflow that periodically collects pull-request review comments, has an AI distill them into coding-guideline candidates, and opens a pull request where humans decide which to adopt. The workflow runs in two phases — extract, which builds the proposal pull request, and sync, which writes the accepted items back into the guideline document using metadata embedded in the pull request body — and is packaged as a reusable workflow. The author keeps a human in the loop deliberately, on the grounds that guidelines become team norms and that review comments often depend on context an AI cannot weigh.