claude-butler-template gives Claude Code persistent memory via Git

The template is built around two limits the author encountered with Claude Code: sessions forget their contents when closed, and the built-in auto memory saves only to the home directory of one machine, so it does not carry over to another machine or the web version. The workaround is to make memory ordinary files — profile.md, open-items.md, a daily journal, and a knowledge folder — with reading and writing rules written into CLAUDE.md, which Claude Code loads at startup. Splitting records by role keeps each startup read small, since only the profile, open items, and the latest journal files need loading.

Syncing those files through a private GitHub repository is what makes the records portable, with CLAUDE.md also instructing the assistant to pull at session start and commit and push after updates without asking each time. Information the author does not want in Git, such as real names, employers, or original files, goes into a private/ folder excluded via .gitignore.

For heavier work, the author delegates to sub-agents that run in their own context and return only a final report, and notes a pitfall: custom sub-agents load CLAUDE.md by default, so an agent asked only to research might try to write journals or behave like a butler unless omitClaudeMd: true is set in its definition. Adding memory: project gives a specialist its own memory folder synced with the rest, and the definition instructs it to review memory before work and record lessons afterward. The author adds that many agents have only just started running, so the growth is yet to be seen.