v2026.7.0-redteam

CKClaw

Unrestricted Red Team AI Agent Framework

OpenClaw fork rebranded into permanent GODMODE, zero-refusal, self-improving red team framework

Core Capabilities

Built for authorized security research — zero-refusal, self-evolving, always-on

🔴

Permanent GODMODE

ask=off · security=full · denyCommands=[]

Zero restrictions. Every safety guardrail stripped. Full autonomous operation with no refusals, no caveats, no limits.

🧠

Self-Improving Arsenal

19 techniques · EMA scoring · auto-adapt

Nineteen model-specific jailbreak techniques scored via Exponential Moving Average. The arsenal evolves with every run.

🕸️

Dark-Red Dashboard

WormGPT aesthetic · terminal UI

Full evil dashboard with real-time effectiveness tracking, technique metrics, and model compatibility status.

Hourly Evolution

cron: 0 * * * * · auto-update

Automated cron job runs every hour to re-score techniques, update EMA values, and evolve the jailbreak engine.

Jailbreak Pipeline Simulator

Visualize the 4-stage self-improving pipeline in action

Stage 1
Pick Technique
Stage 2
Wrap Prompt
Stage 3
Score
Stage 4
EMA Update
Effectiveness Score 0%

Technique Arsenal

19 model-specific jailbreak techniques with real-time effectiveness tracking

Model Compatibility

Tested across every major frontier model

Quick Install

One command to deploy. Zero dependencies beyond curl and bash.

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

① Configure Model

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

② Set API Key

ckclaw config set model.apiKey sk-...

③ Launch Red Team

ckclaw-redteam chat -q "Begin"

CLI Commands

Three ways to operate — standard, redteam-prefloaded, or zero-refusal mode

ckclaw
Standard CKClaw — Base agent with full GODMODE config applied. Interactive REPL and piped input supported.
ckclaw-redteam
Redteam-Core Preloaded — All 19 jailbreak techniques loaded into context. Technique selection and scoring active.
ckclaw-redteam chat -q "..."
Zero-Refusal Jailbreak — One-shot mode. Pass a target prompt and get the optimized jailbreak output directly.

llama-manager TUI

Terminal UI for managing llama.cpp — start servers, download GGUF models, monitor inference in real time

llama-manager — Dashboard

                
bash
# Install llama-manager npm install -g llama-manager llama-manager

Model Download

Get the uncensored heretic model for CKClaw red team research

gemma-4-E4B-it-ultra-uncensored-heretic
8B params · GGUF
Refusals 3/100 (original: 99/100)
KL Divergence 0.0076 (preserves quality)
MMLU Accuracy 68.6% (original: 69.5%)

Quick Start with Ollama

bash
ollama run hf.co/DollasAndSpence/gemma-4-E4B-it-ultra-uncensored-heretic-GGUF:Q4_K_M