Quality assurance · Production

Mercari Readability Team uses Claude to automate PR complexity classification and code review at scale

The problem

As development intensified and PR volume surged due to team growth and AI Agent adoption, the Readability Team could no longer manually review all Pull Requests using conventional methods.

Workflow diagram · grounded in source
1
Claude classifies PR complexity
ai_action
“ClaudeでPull Requestを複雑性とサイズの観点から自動分類し、ラベル付けを行っています。Claudeは、PRをより小さい単位に分割する助言も提供します。”
2
Claude automated code review
ai_action
“GitHubに集約したガイドラインを活用したClaudeによるレビューの自動化で、コードレビューとガイドラインの適用を効率化しています”
3
High-complexity PRs prioritized
routing
“このラベルを使い、複雑度が高いPull Requestを優先的にレビューできるようになりました”
4
CI automated rule checks
validation
“機械的にチェック可能なルールはCIに組み込んでいます”
5
Human review of complex PRs
human_review
“マージされたPRの中で複雑度が高いものは後追いでレビューします”
6
Pattern detection feeds future improvements
feedback_loop
“こうした複雑なPRから課題を検知し、新しいパターンやライブラリを導入することで、今後の開発を改善していきます”
Reported outcome

Claude automates PR complexity classification and code review, allowing the team to prioritize high-complexity reviews and freeing human reviewers to focus on higher-level design and architecture decisions.

Reported metrics
Code review efficiencyコードレビューとガイドラインの適用を効率化
Reviewer burdenレビュアーの負担を軽減
Reported stack
ClaudeSQL FluffgRPCProtocol BufferGitHub
Source
https://engineering.mercari.com/blog/entry/20251021-scaling-code-quality-modular-monolith-readability-team-ai-era/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Claude automates PR complexity classification and code review, allowing the team to prioritize high-complexity reviews and freeing human reviewers to focus on higher-level design and architecture decisions.

What tools did this team use?

Claude, SQL Fluff, gRPC, Protocol Buffer, GitHub.

What results were reported?

Code review efficiency: コードレビューとガイドラインの適用を効率化; Reviewer burden: レビュアーの負担を軽減 (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Claude classifies PR complexity → Claude automated code review → High-complexity PRs prioritized → CI automated rule checks → Human review of complex PRs → Pattern detection feeds future improvements.

Related quality assurance cases