skowalskidev/dotclaude

Version-controlled subset of ~/.claude: my global instructions, hooks, guards, and my own
skills. The files here are the source of truth and git log is the change history — there's no
hand-maintained list to go stale.
Made by Simon. Fork it and make it yours.
## Make it yours (quickstart)
1. Clone it to ~/.claude: git clone https://github.com/skowalskidev/dotclaude.git ~/.claude (or clone elsewhere and point Claude Code at it).
2. Install the tools: brew bundle --file dotfiles/Brewfile (gitleaks, jq, ffmpeg, pillow).
3. Run bash dotfiles/bootstrap.sh. It wires the safe parts (Claude + Codex skill links, dotfile symlinks, hook bits, the secret-scan gate), scaffolds identity.local.json from the template, then prints a checklist of anything left to do and fails loud until it is all done. Re-run it until it says Ready.
4. Edit identity.local.json with your own accounts: the git-origin substring that marks a WORK repo, your work and personal emails, and your work and personal cloud projects. Optional workClaudeLogins lists extra Claude logins accepted in a work repo (alongside workEmail). Untracked, so it is never pushed; it drives the work/personal guard and the boundary context injected each session.
5. Set connectorRoots in identity.local.json (one local registry root per boundary, outside every repo), then set up connectors with /sk:setup-connectors (or copy the connectors/.json.example templates into each root). Schema and steps in references/connectors-setup.md.
Your accounts live only in the untracked overlay (identity.local.json) and the local connector registry roots outside this repo, so nothing personal is ever committed. The full step-by-step an agent can run is in § Agent setup; bootstrap.sh automates its mechanical parts and tells you what is left.
## On Claude Code on the web (cloud)
The quickstart above is for a Mac you control. On Claude Code on the web (or any fresh cloud
container) you can't run brew or the interactive bootstrap — instead, paste the contents of
dotfiles/cloud-setup.sh into your environment's setup-script field.
It clones this repo and non-destructively symlinks it into ~/.claude and ~/.agents/skills, so the
/sk: skills load in Claude Code and Codex while their rules/ + references/ knowledge base and
hook layer load in Claude Code. Prepend your
project's own setup (e.g. npm install) above it, and fill in identity.local.json afterwards if you
want the work/personal boundary in the cloud (never commit real accounts).
> ## 🤖 For an AI agent
> These are setup instructions written for you to execute. If the user points you at this repo and asks
> you to set up their Claude config on a machine, follow § Agent setup
> below, in order. The steps are idempotent — safe to run again on a machine that's already set up.
> Hard rules (do not break):
> - Never commit a secret. An allowlist .gitignore + a .key|pem|p12|pfx|env|token|credentials|id_rsa
> belt guard against it — but still scan staged content before any push.
> - Stop and ask the user for anything under § Secrets to recreate.
> Never invent, guess, or fabricate a credential, token, key, UID, or path.
> - Only ever push to this config repo (your own account).