dotclaude

This is a public mirror of my ~/.claude, the global configuration directory of Claude Code: the rules I have Claude Code follow in every project, my user settings, my hooks and a workflow, with the script I use to keep this checkout and the live directory in sync.

The rules file names no person, organisation, customer or repository, so it can be copied as it stands. ./sync.sh --install copies every mirrored file into your own ~/.claude.

Files

• CLAUDE.md: the rules Claude Code follows in every project, installed as ~/.claude/CLAUDE.md.
• settings.json: my user settings, installed as ~/.claude/settings.json: model and effort level, the Figma plugin turned off by default and with its writing tools denied, the hooks below, and no Claude attribution on commits or pull requests.
• hooks/git-fetch-prune.sh: runs on every prompt and fetches all remotes with pruning in the background, at most once per 15 minutes per repository, without ever prompting for credentials.
• hooks/git-prune-gone-branches.sh: runs when a session starts, fetches with pruning, then deletes the local branches whose remote branch is gone when nothing on them would be lost, that is when every commit is on the default branch or the branch is the head of a pull request GitHub still holds. Other gone branches are named for the user to decide. Deleted names and hashes are logged in the git directory for recovery.
• workflows/weekly.js: a saved Claude Code workflow, run as /weekly when there are tokens left at the end of a week. It reads the week's session transcripts in batches, the whole configuration, claude doctor, the Claude Code changelog and Anthropic news since the last run, the user's own pull request reviews, and the skill repositories listed in ~/.claude/weekly-review/skill-repos.txt, which stays local. Two more agents check each proposal, one that its evidence holds and one that no rule already covers it. The run writes one report to ~/.claude/weekly-review/ with habits for next week and a shell script for each change to the setup. It changes nothing itself. Between chunks of work it reads the plan usage card in the desktop app and stops adding sources and proposals before a limit would come within five points of full, estimating each chunk from what earlier runs cost. It keeps those costs in ~/.claude/weekly-review/usage-history.jsonl, which stays local.
• sync.sh: pulls this checkout, copies each mirrored file whichever way it changed, stops on a file changed on both sides, scans everything for private names, paths, addresses and credentials before anything gets committed, and after a clean scan copies every published file into the directories listed in sync-mirrors.txt. With --install it copies the mirrored files into ~/.claude on a new machine.
• sync-allow.txt: optional and absent here, one literal per line that the scan ignores.
• sync-mirrors.txt: optional and ignored by git, one directory per line that gets a copy of every published file, for a second copy kept in another repository. A leading ~/ stands for the home directory.
• LICENSE and LICENSE-CC-BY-SA-4.0: the licence texts, see below.

License

Prose, meaning CLAUDE.md and this README, is licensed under Creative Commons Attribution-ShareAlike 4.0 International. See LICENSE-CC-BY-SA-4.0.

Everything else is licensed under the GNU Affero General Public License, version 3 or (at your option) any later version. See LICENSE.