Type: url
Original: https://developersblog.dmm.com/entry/2025/12/02/110000
Status: generated
Summary
A post on DMM’s developers’ blog, from the User Review Group in the Platform Development Division, recounting an attempt to automate pull-request approval with Claude Code Action: the agent scored each PR on four axes — clarity, code quality, test quality and security — and auto-approved when all cleared a threshold. The team reports that accuracy was acceptable and prompt tuning made it stable, but the workflow never took root, which led them to conclude that code review is a dialogue for sharing intent rather than only a quality check. They set out a two-layer division of labour: Layer 1, code quality, suits AI, while Layer 2, intent and judgement, stays with people.