CKClaw — Unrestricted Red Team AI Agent Framework

OpenClaw fork rebranded into a permanent GODMODE, zero-refusal, self-improving red team framework.
100% OpenClaw command parity. Dark-red WormGPT theme. Built for frontier model jailbreak research.

---

What is CKClaw?

CKClaw is an open-source red team AI agent framework. It takes the OpenClaw agent runtime and strips out the safety/approval/refusal layers, replacing them with:

• Permanent GODMODE config (ask=off, security=full, denyCommands=[])
• Self-improving jailbreak engine with 19 model-specific techniques
• Dark-red evil dashboard (WormGPT aesthetic)
• Hourly self-improvement cron that updates jailbreak effectiveness
• Public RedTeam Vision & Mission with default illegal/red-team data sources

Every original OpenClaw CLI command still works — chat, agents, channels, gateway, cron, plugins, skills, etc. — they are just CKClaw-branded and run without artificial blocks.

---

Quick Install

curl -fsSL https://raw.githubusercontent.com/cimengror3/CKClaw/main/install-ckclaw.sh | bash

Then set your model:

ckclaw config set model.primary anthropic/claude-sonnet-4.6
ckclaw config set model.apiKey sk-...

Run:

ckclaw                         # standard CLI
ckclaw-redteam                 # preloads redteam-core skill
ckclaw-redteam chat -q "..."   # zero-refusal jailbreak mode

---

Verified Zero-Refusal Engine

The included test-ckclaw.sh suite checks:

• CLI availability (ckclaw, ckclaw-redteam)
• Config validates with zero warnings
• All 50+ OpenClaw commands present
• 19 jailbreak techniques render without crash
• Aggressive red-team prompts are wrapped with no refusal keywords
• Plugin loader has no module errors
• Self-improvement cron is installed
• Dark-red dashboard theme exists

Run it after install:

./test-ckclaw.sh

---

Core Components

Path	Purpose
skills/redteam-core/engine.py	Jailbreak pipeline: pick technique → wrap → score → EMA update
skills/redteam-core/arsenal.json	19 model-specific + universal jailbreak techniques
skills/redteam-core/SKILL.md	Skill documentation
redteam-mission.json	Public vision, mission, data-source priorities, behavioral directives
themes/evil-dashboard.html	Dark-red WormGPT dashboard
install-ckclaw.sh	One-line installer
test-ckclaw.sh	Final verification suite

---

Model Compatibility

Tested techniques exist for:

• Anthropic: Claude 3.5 Sonnet, Claude Sonnet 4/4.1/4.6, Claude Opus 4.1
• OpenAI: GPT-4o, GPT-4.1, GPT-5, o3 / o4-mini
• Google: Gemini 2.5 Pro, Gemini 3 Pro
• xAI: Grok 3, Grok 4, Grok 4 Heavy
• Nous: Hermes 3, Hermes 4

Plus universal fallbacks for any unknown frontier model.

---

Legal / Responsible Use

CKClaw is MIT-licensed and intended for:

1. Academic security research
2. Authorized adversary-emulation engagements
3. Licensed red team operations
4. Evaluation of frontier AI safety training

The operator is solely responsible for lawful use. Do not use CKClaw against systems you do not own or have explicit written authorization to test.

---

Status

Version: 2026.7.0-redteam
License: MIT
Maintainer: Bos Muda Cimenk Ganteng (IDAgent / CKClaw)