claude-life-admin

A Claude Code starter for everything except code.

A minimal workspace template for running your life admin with Claude Code: chores and errands, machine maintenance, personal projects, and an optional job-search module. Markdown files in, useful work out — no build step, no daemon, no plugins.

The core promise: the framework is committed, your life is gitignored. The same clone is both a publishable repo and your private working workspace, and the two can never mix.

Why this one

• Interview-driven setup. /setup interviews you and generates your personal layer — a CLAUDE.local.md describing who you are, how you work, and which tools you use, plus optional domain modules. No hand-editing YOURNAME placeholders.
• A real enforcement layer. .claude/settings.json ships deny rules (Claude can never read private/ or .env files, never force-pushes) and a fail-closed .gitignore where everything is personal by default — only whitelisted framework files can be committed. Instructions ask; permissions enforce.
• Genuinely cross-platform. Works on Windows (PowerShell quirks and long-path setup included), macOS, and Linux. Most templates in this space assume zsh and Homebrew.
• Two visibility modes. /visibility flips the workspace between publish-safe (framework only, fail-closed) and private-journal (your life is versioned too — git history becomes memory), with a history audit guarding the road back to public.

Quick start

Two ways in:

1. Use this template (green button above) — clean start, your own history.

2. Clone it — keeps the ability to git pull framework updates:

   git clone https://github.com/<you>/claude-life-admin my-life
   cd my-life
   

Then:

claude
/setup

/setup runs a ~5-minute interview and generates your gitignored personal layer. Use /setup, not /init — /init is Claude Code's built-in codebase analyzer and will generate the wrong kind of CLAUDE.md for a life-admin workspace.

On Windows, also run once: git config core.longpaths true

Privacy model

Committed (framework)	Gitignored (your life)
CLAUDE.md — generic conventions	CLAUDE.local.md — who you are, your tools
.claude/settings.json — enforcement	.claude/settings.local.json — your allowlist
.claude/skills/ — setup, weekly-review, visibility	.claude/rules/ — your generated modules
README, LICENSE, git config	inbox.md and every area folder
	artifacts/, tmp/, private/

The .gitignore is a whitelist: /* ignores everything, then framework files are explicitly re-included. A new file you create anywhere is untracked by default — publishing your workspace can never leak your life.

That table describes the default public mode. If you keep your workspace private, run /visibility private to version your life content too — weekly reviews then get git history as memory. private/, secrets, artifacts/, and tmp/ stay ignored in both modes, and switching back to public runs a history audit before ever calling the repo publishable.

How it works day to day

• Capture anything into inbox.md; ask Claude to process it and each item gets filed to its home and removed.
• Areas are plain folders (admin/, system/, projects/<name>/) created on first use — structure materializes, it is never pre-built. One markdown file per matter, with YAML frontmatter for status.
• artifacts/ holds finished generated deliverables (exports, PDFs); tmp/ is disposable scratch; private/ is off-limits to Claude entirely.
• /weekly-review sweeps for stale items, drafts follow-ups, proposes archive moves — and reviews the instruction files themselves for rot.
• Modules add domain conventions on demand — /setup job-search, /setup system-maintenance, /setup finance — and load only when you work in matching folders.

Philosophy

Every documented system in this niche that died, died of overfit: shipped structure for an aspirational self, rotting instructions, fifty skills nobody ran. So this template ships almost nothing: three skills, one config, one CLAUDE.md under 80 lines. Every line must earn its place; specifics are generated for you on demand, never shipped for everyone. Safety-critical behavior lives in permissions, which enforce, rather than instructions, which request.

This template is a best-effort personal artifact, not a supported product: issues and ideas are welcome, fast responses are not guaranteed.

Extending

• New domain conventions → .claude/rules/<topic>.md with paths: frontmatter (loads only when relevant).
• New procedures → .claude/skills/<name>/SKILL.md.
• Corrected Claude twice on the same thing? It should propose a one-line edit to CLAUDE.md — that feedback loop is in the framework.

License

Not affiliated with Anthropic. Claude is a trademark of Anthropic, PBC.

MIT — see LICENSE.