Imported from kemalozansayin/ai-boilerplate (claude/skills/sync/SKILL.md). Install upstream with npx skills add kemalozansayin/ai-boilerplate --skill sync. Copyright stays with the author.

Sync — User-Scope Config ↔ ai-boilerplate

Keeps this machine's Claude/Codex user-scope config in sync with https://github.com/kemalozansayin/ai-boilerplate (branch: master), so the whole system — rules, skills, agents, hooks, settings, and the Codex AGENTS.md mirror — can be carried to another machine.

Local clone: D:\Projects\ai-boilerplate on this machine (adjust to wherever this machine keeps its projects if different — sibling to the other project directories).

Push mode is restricted to the repo owner (kemalozansayin) — see the ownership check that opens push mode below. If this isn't your own fork of kemalozansayin/ai-boilerplate, don't run push mode against this origin; fork the repo and update the remote URL/local clone path above first. Setup mode (cloning + copying into ~/.claude/ and ~/.codex/) works for anyone, regardless of who owns the repo.

Repo layout mirrors the live locations 1:1, just without the leading dot so they read normally in a GitHub file tree:

ai-boilerplate/
├── claude/
│   ├── settings.json        ← templated, see "Portable settings.json" below
│   ├── rules/*.md
│   ├── skills/**/*          ← deploy/, new-project/ (+ references/), sync/
│   ├── agents/*.md
│   └── hooks/*.mjs
└── codex/
    └── AGENTS.md

Two modes. If the request doesn't make the mode obvious, ask which one — this is a live, per-run decision, not something to assume from a previous invocation.

Portable settings.json

The live ~/.claude/settings.json registers hooks with absolute paths (C:/Users/Lenovo/.claude/hooks/lint-gate.mjs, …) — specific to this machine and this user account, and would silently break on another one. The repo's copy is templated instead: every occurrence of the live home-relative Claude root is replaced with the literal token {{CLAUDE_HOME}}, e.g.:

"args": ["{{CLAUDE_HOME}}/hooks/lint-gate.mjs"]

Push mode resolves the real path back out to {{CLAUDE_HOME}} before writing the repo copy. Setup mode does the reverse: resolves {{CLAUDE_HOME}} → the new machine's real ~/.claude path (forward slashes, matching the existing convention) before writing the live file. Never commit the resolved (machine-specific) form to the repo.

Push mode — local → repo → GitHub

Push mode is owner-restricted by design, not just by GitHub permissions: this repo's write access belongs to kemalozansayin alone (public repo, no collaborators — GitHub already refuses a push from anyone else), but push mode checks this itself first too, so a non-owner gets a clear explanation instead of a confusing git/auth failure partway through.

0. Confirm ownership before touching anything: gh api user --jq .login and compare against kemalozansayin. If it doesn't match, or gh isn't authenticated, stop — report that this repo's push mode is restricted to its owner, and that anyone else wanting to sync their own config should fork the repo and repoint the remote URL / local clone path (see Local clone above and the README's Make It Yours) rather than push here. Do not attempt the git push.
1. Copy the current state into the local clone, overwriting what's there (the live ~/.claude / ~/.codex/AGENTS.md is always the source of truth; the clone is a mirror):
   • ~/.claude/rules/*.md → claude/rules/
   • ~/.claude/skills/** → claude/skills/**
   • ~/.claude/agents/*.md → claude/agents/
   • ~/.claude/hooks/*.mjs → claude/hooks/
   • ~/.claude/settings.json → claude/settings.json, templated per above
   • ~/.codex/AGENTS.md → codex/AGENTS.md
   • Never copy ~/.claude/settings.local.json, ~/.claude/.credentials.json, or anything under ~/.claude/projects/, ~/.claude/cache/, ~/.claude/shell-snapshots/, ~/.claude/tasks/, ~/.claude/plans/, ~/.claude/backups/, ~/.claude/ide/ — none of that is portable config, and some of it (credentials, session transcripts) must never leave the machine.
2. git status --short in the clone and review what actually changed.
3. Stage, then propose a commit in the house format (icon | [PREFIX] | Message, see commit-standards) — usually 🔧 | [INFRA] | ... for a routine config sync, or a more specific prefix if the underlying change was e.g. a new rule/skill/agent worth calling out. Wait for the user's explicit approval of that commit — this flow has no standing exception to Commit Approval.
4. Once approved, commit and git push origin master.
5. Report what was synced.

Setup mode — bring a (new) machine up to date

Invoking this mode is the go-ahead to install the two CLIs below and copy these files into place — a per-invocation authorization, not a standing permission for unrelated installs.

1. Preflight: check for git, node/npm, claude, codex, gh, pnpm. Report the full picture before changing anything.
2. Install what's missing and CLI-installable:
   • Claude Code: npm install -g @anthropic-ai/claude-code (Node 18+).
   • Codex CLI: npm install -g @openai/codex (Node 22+).
   • pnpm missing but corepack available (ships with modern Node): corepack enable.
   • git, node itself, and gh are OS-level installs — report if missing and let the user install those themselves; don't attempt to.
3. Get the repo: clone https://github.com/kemalozansayin/ai-boilerplate.git (branch master) to the local projects root if not already present there; git pull if it is.
4. Before overwriting anything, check whether ~/.claude/{rules,skills,agents,hooks} or ~/.codex/AGENTS.md already have content on this machine. A truly fresh machine has none of this — but if something is already there, surface what would be overwritten and confirm with the user first rather than silently clobbering possibly-unrelated existing config.
5. Copy into place:
   • claude/rules/*.md → ~/.claude/rules/
   • claude/skills/** → ~/.claude/skills/**
   • claude/agents/*.md → ~/.claude/agents/
   • claude/hooks/*.mjs → ~/.claude/hooks/
   • claude/settings.json → ~/.claude/settings.json, resolving {{CLAUDE_HOME}} to this machine's real path first (see Portable settings.json). Leave any existing ~/.claude/settings.local.json alone — it's machine-local and not part of this repo.
   • codex/AGENTS.md → ~/.codex/AGENTS.md (create ~/.codex/ if it doesn't exist).
6. Report: what got installed, what got copied, and what still needs the user's own action (claude login / API auth, codex login, gh auth login, confirming the projects-root path if this machine's convention differs).

Keeping this skill itself in sync

Like every other user-scope flow, changes here follow the agents-md-sync rule — mirror any edit to this skill into ~/.codex/AGENTS.md's Sync (push / setup) section in the same change.