Paper 1 · Question 57

A: Root CLAUDE.md loads for every session. Including all three rule sets wastes context tokens when editing only one documentation type. The model must also correctly identify which section applies, adding an unnecessary point of failure.

B: Path-scoped rules in .claude/rules/ with glob patterns load only when editing files in the matching directory. Each documentation type gets its own rules loaded automatically and exclusively, conserving context tokens and eliminating the risk of applying incorrect standards.

C: Directory-level CLAUDE.md files load based on the working directory hierarchy, but Claude Code sessions typically operate from the repository root. These directory-level files would only load when the agent is directly working within that specific directory, which is less reliable than path-scoped rules that match by file pattern regardless of working directory.

D: This scenario explicitly requires writers to invoke the right skill manually, which depends on human memory and discipline — the wrong pick silently applies the wrong standards. Even if the skills added a paths frontmatter to auto-activate, they'd still load as a task-style workflow rather than as always-in-context guidance. Path-scoped rules in .claude/rules/ are the correct choice: they load into context automatically when Claude reads a matching file, with no human step in the loop.