👑 Claude God Mode

The ultimate Claude Code optimization toolkit.
15 AI agents • Autonomous token pipeline • 60-99% token savings • Opus-level reasoning • Elite UI design

Stop burning tokens. Start shipping code.

Claude God Mode is a drop-in optimization toolkit for Claude Code that combines autonomous token reduction with Opus-level reasoning patterns — giving you 2-10x more usage from your plan while producing higher quality code.

⚡ What You Get

Token Pipeline (Autonomous)

Every layer runs automatically — no manual intervention after setup.

Your Prompt
    |
    +-  [RTK]               Bash output compressed 60-90%
    |
    +-  [jCodeMunch]         Lightweight code indexing, 95%+ savings (50-150MB)
    |
    +-  [Context Mode]      Tool outputs sandboxed in SQLite (98%)
    |
    +-  [MCP Compressor]    JSON/data auto-compressed (25-66%)
    |
    +-  [Cozempic]          Context auto-pruned at 4 thresholds
    |
    +-  [Headroom]          AST-aware prompt compression (~50%)
    |
    v
Claude sees ONLY what it needs

Quality Agents (On-Demand)

15 specialized agents that push Claude's reasoning to its ceiling.

🚀 Quick Start

One-Line Install (Recommended)

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/Itachi-1824/claude-god-mode/main/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/Itachi-1824/claude-god-mode/main/install.ps1 | iex

Manual Install

git clone https://github.com/Itachi-1824/claude-god-mode.git
cd claude-god-mode
# Copy agents
cp agents/*.md ~/.claude/agents/
# Copy CLAUDE.md template (review and customize first!)
cp config/CLAUDE.md.template ~/.claude/CLAUDE.md
# Copy .claudeignore template to your projects
cp config/.claudeignore.template /path/to/your/project/.claudeignore

📊 Token Savings

Measured savings from real-world usage:

Combined effect: 2-10x more usage from the same plan.

🤖 Agents Deep Dive

All agents live in ~/.claude/agents/ and are available globally. Claude auto-delegates to them based on the task, or you can invoke them explicitly.

Adversarial Coach 🔴

Model: Opus | Effort: High | Tools: Read-only

Tries to break your code. Never praises, never approves. Every issue comes with a concrete exploit scenario. Use after any significant implementation.

Red Team 🔴

Model: Opus | Effort: Max | Tools: Read-only

You tell it what security measures you added. It specifically attacks those stated assumptions. Finds blind spots that generic security review misses.

Architect 🟣

Model: Opus | Effort: Max | Tools: Read-only

System design only. Produces ADRs (Architecture Decision Records). Compares 3+ approaches before recommending. Never writes implementation code.

Tree of Thought 🔵

Model: Opus | Effort: Max

Mandatory 4-phase process: Decompose → Branch (min 3 approaches) → Evaluate → Select. Will not start implementing until all branches are compared.

RISE Refiner 🟠

Model: Opus | Effort: High

Recursive self-improvement: Generate → Critique (exactly 3 weaknesses) → Fix only those 3 → Compare before/after. 17-24% measured improvement.

Constitutional Reviewer 🔵

Model: Opus | Effort: High

4-pass review: Write → Critique against 6 principles → Rewrite fixes → Verify all violations resolved. Must find at least 1 issue (if zero found, review wasn't thorough enough).

Researcher 🟦

Model: Sonnet | No turn limit

Multi-depth research with 7 levels: surface, basic, deep, in-depth, ultra, nuclear, overkill. Each level maps to specific search strategies and agent patterns. Supports pluggable web search.

See all agents →

⚙️ What Gets Installed

Optional Components

The installer asks before installing each of these:

🎨 UI/Design Stack

God Mode turns Claude into an elite frontend designer with autonomous visual feedback loops.

Component Libraries (MCP Servers)

Design Agents

How It Works Autonomously

1. You say "build a landing page for X"
2. ui-architect agent auto-activates (description-based routing)
3. It fetches components from 21st.dev/shadcn/magicui instead of writing from scratch
4. Adds animations from animotion + magicui
5. design-critic auto-runs after implementation
6. Visual score + specific fix suggestions returned

Note: 21st.dev requires a free API key from 21st.dev. Replace REPLACE_WITH_YOUR_21STDEV_API_KEY in ~/.claude.json after install.

📦 Per-Project Setup

After installing God Mode globally, each new project needs:

1. Code Index (automatic)

jCodeMunch indexes your project automatically when Claude navigates code via its MCP tools. No manual setup needed.

2. Graphify (one-time per repo)

# In your project directory:
graphify .
# Or via Claude:
/graphify .

Auto-updates via PostToolUse hooks.

3. .claudeignore (recommended)

# Copy the template to your project root:
cp ~/.claude-god-mode/.claudeignore.template /path/to/project/.claudeignore

Prevents Claude from reading node_modules/, dist/, build artifacts, etc.

🛡️ Safety & Automation

Hook Scripts (~/.claude/bin/)

Safety Hooks (Deterministic)

🏗️ Architecture

~/.claude/
  +-  CLAUDE.md              ~80 lines (lean, connectivity + testing rules)
  +-  settings.json          17 hooks across 8 lifecycle events
  +-  agents/                15 specialized agents (on-demand, 0 tokens when unused)
  +-  bin/                   6 hook scripts (safety, handover, loop guard)
  +-  skills/                Plugin skills (graphify, caveman, etc.)
  +-  plugins/               11 enabled (LSP, security, design, memory, skills)
  +-  projects/              Per-project memory and config

~/.claude.json               12 MCP servers: jcodemunch, context-mode,
                             headroom, serena, docfork, 21st-dev-magic, shadcn,
                             magicui, animotion, dembrandt, aceternity, glance

💬 Philosophy

1. Autonomous over manual — tools that intercept and optimize automatically beat instructions that Claude might ignore
2. Agents over CLAUDE.md — specialized agents load on-demand (~0 tokens when unused) vs. CLAUDE.md rules (loaded every turn)
3. Hooks over suggestions — PreToolUse hooks that nudge behavior are stronger than polite instructions
4. Evidence over assertions — every verification requires proof (test output, type checker results), not "this should work"
5. Lean base, rich extensions — CLAUDE.md stays under 60 lines; complexity lives in agents and hooks

🤝 Contributing

PRs welcome! Areas we're looking for help:

• New agents for specific domains (ML, mobile, embedded, etc.)
• Search provider integrations
• Benchmarks comparing with/without God Mode
• Platform-specific install improvements
• Documentation translations

📝 License

MIT — do whatever you want with it.

_{Built by @Itachi-1824 after hitting Claude Code limits one too many times.}