CLAUDE.md는 규칙과 참조 링크만 담고, 상세 내용은 별도 파일로 분리하십시오.
API 스펙·DB 스키마·아키텍처 문서를 매 세션 로드하는 것은 토큰 낭비이며, 참조로 필요할 때만 불러오는 것이 효율적입니다.
CLAUDE.md should hold rules and references only — push API specs, DB schemas, and architecture details into separate files. Loading them every session is a tax; reference them on demand instead.
CLAUDE.md는 규칙과 참조 링크만 담고, 상세 내용은 별도 파일로 분리하십시오.
API 스펙·DB 스키마·아키텍처 문서를 매 세션 로드하는 것은 토큰 낭비이며, 참조로 필요할 때만 불러오는 것이 효율적입니다.
CLAUDE.md가 매 세션 자동 로드된다는 사실에서 출발한 운영 규칙입니다. API 엔드포인트 50개, DB 테이블 40개를 다 담고 있으면 매번 수천 토큰이 낭비됩니다 — 정작 그 세션에 필요한 정보는 5%도 안 되는데. 해법은 CLAUDE.md에 참조(`docs/api.md`, `docs/db-schema.md`)만 두고, 클로드가 필요한 순간에만 그 파일을 로드하게 만드는 것입니다.Also in EnglishCLAUDE.md should hold rules and references only — push API specs, DB schemas, and architecture details into separate files. Loading them every session is a tax; reference them on demand instead.
A discipline that follows from one fact: CLAUDE.md auto-loads every session. Cramming 50 API endpoints + 40 DB tables into it burns thousands of tokens each time, when less than 5% of that content is relevant to the current task. The fix: CLAUDE.md carries references (`docs/api.md`, `docs/db-schema.md`) only, and Claude loads each detail file on demand when the task actually needs it.