@import in CLAUDE.md saves zero tokens; to slim it down, run /doctor prompt-audit

AGENTS.md and CLAUDE.md load in full at the start of every session and ride along on every turn. Longer costs more and gets followed less. The official guideline is under 200 lines per file. Codex reads at most 32 KiB combined and drops the rest.

From a widely shared post this week, here is what checks out against the docs.

• @path imports only make the file look tidier. Imported files load in full at launch, so the token count is unchanged.
• Codex reads them once, at session start (global, then project root down to the current folder). Claude Code loads a subfolder's AGENTS.md when it reads a file there. Rules that matter for one folder belong in that folder.
• Every installed skill's name and description takes up context on every turn. /skill-doctor shows what each costs and how often it runs, so you can turn off the ones you don't use (v2.1.252 or later).

Don't prune by hand. This finds instructions written for older models, references to files that don't exist, and files that contradict each other, and changes nothing until you approve (v2.1.283 or later).

/doctor prompt-audit

If a repo has both files, Claude Code reads only CLAUDE.md by default (from v2.1.277 it reads AGENTS.md directly only when there is no CLAUDE.md). Keep shared rules in AGENTS.md and let CLAUDE.md point at it with one line.

@AGENTS.md