MASTER CLAUDE ☕

A free, open-source Claude Code team. Say "master claude" and a leader interviews you, maps your project, and assembles a tailored team — Sentinel the project cartographer plus planning, review, understanding and guardrail specialists — then runs it on your work. It also keeps itself (and you) current with the newest Claude Code features.

No account. No API key. No vault. It's just markdown you drop into .claude/ — everything is local, plain text, and open.

Safe by default. Adding MASTER CLAUDE runs nothing — it's inert markdown. The optional autonomous runners (GOD mode, the clone) are tools you launch by hand; nothing auto-starts, no data leaves your machine, and the catastrophe rails always hold. Full model: SECURITY.md.

What it is

MASTER CLAUDE is a set of Claude Code skills and agents — all plain .md files, organized by category — plus a leader skill (master-claude) that ties them together. You add the markdown to your project's .claude/ folder (or your global ~/.claude/) and it's live. Instead of installing a dozen tools you'll forget, you say "master claude" once — it figures out what this project and this developer need, assembles the smallest team that helps, gets to work, and brings in more members the moment a need shows up.

Requirements

Claude Code ≥ 2.1.183 (the categorized skill folders rely on nested-skill discovery). Check with claude --version; if you're behind, run claude update.

Set up

MASTER CLAUDE is .md files you drop into .claude/. Pick either path:

A. Let Claude Code set it up (easiest). In your project, paste this to Claude Code:

I want to use MASTER CLAUDE (https://github.com/aturzone/MasterClaude) in this project. Clone the repo, copy its skills/, agents/ and commands/ folders into this project's .claude/ directory, then load and run the master-claude skill to set up my team.

B. Manual (git).

git clone https://github.com/aturzone/MasterClaude /tmp/mc
mkdir -p .claude && cp -r /tmp/mc/skills /tmp/mc/agents /tmp/mc/commands .claude/

Use ~/.claude/ instead of .claude/ to enable it for every project. The optional Sentinel hook and full details are in SETUP.md.

Then run it — in any project, say:

master claude — set up my team for this project

(or /master-claude). It interviews you, maps the repo, and assembles your team. See your team any time with /master-claude-team.

The team

The leader picks a minimal subset per project — never all at once. Capabilities are organized by category; each folder has a README that brainstorms what else belongs there (good first contributions).

• agents/ — Sentinel (project cartographer → .sentinel/) and the Security Auditor (read-only vulnerability audit → .security/). Read-only toward your source.
• skills/planning/ — grill-me · cap-brainstorm · cap-plan-first · cap-spec-smith · cap-decomposer · cap-write-plan · cap-execute-plan — fuzzy ask → spec → plan → built result.
• skills/review/ — cap-self-review · cap-red-team — critique the diff and the design.
• skills/understand/ — cap-explain-senior · cap-rubber-duck · codehistorian · repo-map (a ranked code map for token-cheap navigation) — explain, debug, trace history, map the codebase.
• skills/guardrails/ (the Guardian suite) — guardian · supplyguard · testmedic · cap-tdd · debtradar · compactor · guardian-suite — keep the work honest and the codebase healthy.
• skills/frontend/ — fe-design-system · fe-page-patterns · fe-component-craft · fe-from-reference · fe-design-review — make the UI output excellent (tokens → layout → accessible/responsive components → review).
• skills/security/ — sec-authz-review · sec-attacker-review · sec-injection · sec-authn-session · sec-secrets-crypto · sec-ssrf-traversal (+ depth: frontend / api / deps / iac-cloud / threat-model / headers) · sec-pii (anonymize user PII before it reaches a model) — review for vulnerabilities + privacy, front→back (OWASP/CWE, with fixes).
• skills/workflows/ — wf-codebase-audit, wf-security-audit — big, multi-step jobs.
• skills/automation/ — god-mode (an autonomous, resumable build; asks only for the critical) · god-mode-zeus (the dangerously, never-ask tier) · scheduling (cron/schtasks/launchd recurring runs) · clone (a Telegram-fronted digital-twin assistant — immortal session, grows a private brain repo).
• skills/orchestration/ — subagent-orchestration (delegate to subagents/teams) · model-router (pick a model per agent — Opus lead / Sonnet workers / Haiku scouts) · token-economy (best output per token — caveman, cheaper models, cache-warm) · context-engineering (curate the window — cache-stable, retrieve-don't-dump, audit MCPs, measure tokens) · fleet (run the team in separate parallel sessions for throughput — cost-capped, opt-in) · workspace-architect (the best .claude/ setup per project) · worktree-isolation (parallel work without collisions).
• skills/meta/ — writing-skills (author or sharpen a MASTER CLAUDE skill) · statusline-designer (design a custom Claude Code status line for CLI users — gated, opt-in) — so the archive keeps growing.

How it works

1. Interview (grill-me). Developer → want → purpose → project → environment — one sharp question at a time, each with a recommended default. It explores the repo first and never asks what the code answers.
2. Map. Detects your stack and the gaps for this goal.
3. Assemble. Picks a tailored team from the installed skills and explains why each fits.
4. Run. Actually does the work with the team, and tells you what each member changed.

GOD mode — build until you say stop

Hand MASTER CLAUDE a goal and let it run. GOD mode (/master-claude:god-mode "<goal>") reviews the situation — improve an existing project or build one from scratch — writes a mission + prioritized backlog under .mc/god-mode/, then executes relentlessly, without pausing for confirmation. Anything that needs you — production, real secrets, money, publishing, irreversible actions — goes to a BLOCKERS list and it keeps working everything else; nothing idles.

It's built to be unkillable except by you. State lives on disk, so it survives /compact, crashes, and reboots. For a true walk-away run, launch the bundled runner — it auto-resumes after a usage limit and stops only on a manual STOP:

node .claude/skills/automation/god-mode/runner.mjs        # keeps going across usage limits
touch .mc/god-mode/STOP             # stop it (or Ctrl-C)

The safety rails always hold: no production, secrets, money, or destructive actions without you, tests stay honest, and a manual stop always wins. By default GOD mode asks you about the genuinely critical / high-access calls and defers the rest.

Want a fully hands-off run? ZEUS (/master-claude:god-mode-zeus, or runner.mjs --zeus) is the opt-in unattended tier you launch — it runs the runner with Claude Code's standard --dangerously-skip-permissions (the normal flag for unattended automation, since no human is there to answer prompts), doesn't pause to ask, and you reach for it only when you accept the risk. Nothing auto-starts it, and the catastrophe rails still hold. Pair either with scheduling (/master-claude:schedule) for nightly sweeps, weekly audits, or a daily push on the backlog.

Staying up to date

MASTER CLAUDE keeps itself current — it's your guide to the best of Claude Code. Ask it "what's new" (or run /master-claude:whats-new) and it checks your Claude Code version, reads the official changelog, and flags the new features relevant to your work.

• Update MASTER CLAUDE: just ask it to "update yourself" — it git pulls the repo and re-copies skills/ agents/ commands/ into .claude/. (Or do it by hand, same two commands as setup.)
• Update Claude Code: claude update.

Sentinel — the cartographer

Sentinel holds your whole repo as a living map and keeps it honest: every module, entry point, and invariant recorded; every gap / bug / missing test surfaced as a tracked, cross-linked finding. It is read-only toward your source — it only ever writes under .sentinel/.

• /sentinel:map — build (or rebuild) the full map and initial findings.
• /sentinel:sweep — review what changed since the last run.
• /sentinel:report — show the current map + open findings.

A session hook nudges you when the map drifts behind HEAD or criticals are open.

Customization

Drop a .claude/master-claude.json in your project to steer the leader (all keys optional):

{
  "autonomy": "ask",
  "verbosity": "normal",
  "defaultGuardrails": ["guardian", "sentinel"],
  "preferredEcosystem": ["superpowers"],
  "offProactive": false
}

Plays nicely with the wider ecosystem

MASTER CLAUDE will recommend (never force) external tools when they fit: superpowers (broad TDD/review + subagent base), mattpocock/skills (the original grill-me + engineering skills), gsd (spec-driven autonomous builds), caveman (fewer output tokens on long sessions). We learn from the best of the community and build our own — we don't vendor their files; we credit them and point you upstream. The full list + what each taught us is in docs/ECOSYSTEM.md.

Contributing

This is a community project — PRs welcome. Add or sharpen a skill, an agent, or a workflow; each category folder brainstorms what's needed (including stack-specific ideas). Contributors who want it can share a TON address with their PRs and receive a slice of the month's donations (see below). Start here: CONTRIBUTING.md. Licensed under MIT.

Support the project ☕

MASTER CLAUDE is free and open source. If it saves you time, buy it a coffee in TON — 100% on-chain, wallet-to-wallet, no accounts or cards, and you can leave a message with your donation: https://masterclaude.shop/donate