Your monorepo has a root CLAUDE.md that has grown to over 400 lines, covering Python conventions, TypeScript conventions, infrastructure rules, and testing standards. Developers report that Claude sometimes applies the wrong conventions to the wrong files, and the file is difficult to maintain. What is the best approach to reorganize this configuration?

解析

• Ais correct, as verified against the official Claude Code memory docs (code.claude.com/docs/en/memory), which document exactly this mechanism: place topic-focused markdown files in .claude/rules/, and use YAML frontmatter with a paths field of glob patterns so a rule "only load[s] into context when Claude works with matching files." This directly cures both reported symptoms — wrong conventions applied to wrong files (path scoping means the Python rules never load while editing a .ts file) and hard-to-maintain monoliths (modular per-topic files). The distractors each fail on a documented point:
• B(per-package CLAUDE.md, delete root) uses real nested-CLAUDE.md scoping but is only directory-scoped, cannot separate Python vs TS within one package, and deleting the root discards shared/cross-cutting rules.
• Cmisdescribes /memory (it lists/edits memory files and toggles auto memory — it does not "prioritize a section per task") and still requires manual intervention every session.
• Dis explicitly called out by the docs — "Splitting into @path imports helps organization but doesn't reduce context, since imported files load at launch," so unconditional imports keep every convention in context and the wrong-convention problem persists.