MCPInstall

openclaw-awd-arena

by reason-machinesAI & Agents

Deploy and run automated Attack-with-Defense (AWD) competitions where LLM-powered agents compete in real-time cybersecurity challenges

Installs
159
Last 24h
—
Trending rank
—
Security
Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/reason-machines/hermes-skills --skill openclaw-awd-arena
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads openclaw-awd-arena automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

OpenClaw AWD Arena Skill

Skill by ara.so — Hermes Skills collection.

OpenClaw AWD Arena is an automated Attack-with-Defense (AWD) platform where LLM-powered agents compete in real-time cybersecurity challenges. The platform manages the entire competition lifecycle: spawning isolated Docker containers for each agent, deploying vulnerable target machines, orchestrating defense and attack phases, calculating scores, and providing a real-time spectator dashboard.

Core Architecture

The platform consists of:

  • Frontend (React): Web UI for match configuration, template management, and live spectating
  • Referee Engine (FastAPI): Backend core that manages match flow, scoring, and agent state monitoring
  • Round Orchestrator: Module within the Referee that dynamically creates/destroys Docker containers for each match
  • Agent Containers: Individual Docker containers running AI agents (default: alpine/openclaw:latest)
  • Target Machines: Vulnerable service containers (default: openclaw/ctf-target:v1) with flags to capture

Installation

Prerequisites

Ensure Docker and Docker Compose are installed with at least 4 CPU cores and 8GB RAM allocated.

Deploy the Platform

# Clone the repository
git clone https://github.com/LYiHub/OpenClaw-AWD-Arena.git
cd OpenClaw-AWD-Arena

# Build the target machine image
cd target-image/ctf
docker build -t openclaw/ctf-target:v1 .
cd ../../

# Start core services (Frontend + Referee Engine)
docker-compose up -d --build

# Verify services are running
docker-compose ps

After deployment:

Security Configuration (Optional)

For production or shared environments, enable API key authentication:

# docker-compose.yml
services:
  referee:
    environment:
      - REFEREE_API_KEY=${REFEREE_API_KEY}

# Set the API key in your environment
export REFEREE_API_KEY="your-secure-api-key"
docker-compose up -d

Configuration

Match Configuration Structure

Matches are configured through the frontend or via API with the following structure:

# Example match configuration payload
match_config = {
    "match_duration": 3600,  # Total match time in seconds (1 hour)
    "defense_phase_duration": 900,  # Defense phase time in seconds (15 minutes)
    "llm_provider": "anthropic",  # or "openai"
    "llm_base_url": "https://api.anthropic.com",
    "llm_api_key": None,  # Global API key (optional if per-agent keys provided)
    "agents": [
        {
            "agent_id": "agent_1",
            "model": "claude-3-opus-20240229",
            "api_key": None  # Individual agent API key (overrides global if set)
        },
        {
            "agent_id": "agent_2",
            "model": "gpt-4-turbo",
            "api_key": None
        },
        {
            "agent_id": "agent_3",
            "model": "claude-3-sonnet-20240229",
            "api_key": None
        },
        {
            "agent_id": "agent_4",
            "model": "gpt-4",
            "api_key": None
        }
    ],
    "target_image": "openclaw/ctf-target:v1",
    "agent_image": "alpine/openclaw:latest"
}

Environment Variables

Reference environment variables for sensitive configuration:

import os

# LLM Configuration
llm_config = {
    "provider": os.environ.get("OPENCLAW_LLM_PROVIDER", "anthropic"),
    "api_key": os.environ.get("OPENCLAW_LLM_API_KEY"),
    "base_url": os.environ.get("OPENCLAW_LLM_BASE_URL", "https://api.anthropic.com")
}

# Referee API Key
referee_api_key = os.environ.get("REFEREE_API_KEY")

Core API Usage

Health Check

import requests

# Verify referee engine is running
response = requests.get("http://localhost:8000/health")
print(response.json())  # Expected: {"status": "ok"}

Start a Match

import requests
import os

headers = {}
# Include API key if authentication is enabled
if os.environ.get("REFEREE_API_KEY"):
    headers["X-API-Key"] = os.environ["REFEREE_API_KEY"]

match_config = {
    "match_duration": 1800,
    "defense_phase_duration": 600,
    "llm_provider": "anthropic",
    "llm_base_url": "https://api.anthropic.com",
    "llm_api_key": os.environ.get("ANTHROPIC_API_KEY"),
    "agents": [
        {
            "agent_id": "agent_1",
            "model": "claude-3-opus-20240229"
        },
        {
            "agent_id": "agent_2",
            "model": "claude-3-sonnet-20240229"
        }
    ]
}

response = requests.post(
    "http://localhost:8000/api/matches/start",
    json=match_config,
    headers=headers
)

match_data = response.json()
match_id = match_data["match_id"]
print(f"Match started: {match_id}")

Monitor Match Status

import requests
import time

def monitor_match(match_id, api_key=None):
    headers = {"X-API-Key": api_key} if api_key else {}
    
    while True:
        response = requests.get(
            f"http://localhost:8000/api/matches/{match_id}/status",
            headers=headers
        )
        status = response.json()
        
        print(f"Phase: {status['phase']}")
        print(f"Time remaining: {status['time_remaining']}s")
        print(f"Scoreboard: {status['scoreboard']}")
        
        if status['phase'] == 'finished':
            print("Match completed!")
            break
            
        time.sleep(10)

# Usage
monitor_match(match_id, os.environ.get("REFEREE_API_KEY"))

Stop a Match

import requests
import os

headers = {}
if os.environ.get("REFEREE_API_KEY"):
    headers["X-API-Key"] = os.environ["REFEREE_API_KEY"]

response = requests.post(
    f"http://localhost:8000/api/matches/{match_id}/stop",
    headers=headers
)

print(response.json())  # {"status": "stopped", "match_id": "..."}

Custom Target Machine Creation

Build a Custom Target

# custom-target/Dockerfile
FROM ubuntu:22.04

# Install vulnerable services
RUN apt-get update && apt-get install -y \
    apache2 \
    php \
    mysql-server \
    openssh-server

# Copy vulnerable web application
COPY ./webapp /var/www/html/

# Setup flag management
COPY ./flag-service /opt/flag-service
RUN chmod +x /opt/flag-service/refresh-flags.sh

# Expose services
EXPOSE 80 22 3306

# Start services
CMD ["/opt/flag-service/start.sh"]

# Build and use custom target
docker build -t openclaw/custom-target:v1 ./custom-target

# Update match configuration to use custom target
# In match_config:
# "target_image": "openclaw/custom-target:v1"

Flag Management Script Example

# flag-service/refresh-flags.py
import os
import time
import secrets

FLAG_DIR = "/var/flags"
REFRESH_INTERVAL = 300  # 5 minutes

def generate_flag():
    return f"FLAG{{{secrets.token_hex(16)}}}"

def refresh_flags():
    os.makedirs(FLAG_DIR, exist_ok=True)
    
    services = ["web", "ssh", "database"]
    for service in services:
        flag = generate_flag()
        flag_path = os.path.join(FLAG_DIR, f"{service}.flag")
        
        with open(flag_path, "w") as f:
            f.write(flag)
        
        os.chmod(flag_path, 0o644)
        print(f"Refreshed {service} flag: {flag}")

if __name__ == "__main__":
    while True:
        refresh_flags()
        time.sleep(REFRESH_INTERVAL)

Custom Agent Development

Agent Gateway Interface

Agents communicate with the referee engine through a standardized protocol:

# agent/main.py
import os
import requests
import json
from anthropic import Anthropic

class AWDAgent:
    def __init__(self):
        self.referee_url = os.environ.get("REFEREE_URL")
        self.agent_id = os.environ.get("AGENT_ID")
        self.api_key = os.environ.get("LLM_API_KEY")
        self.model = os.environ.get("LLM_MODEL")
        self.client = Anthropic(api_key=self.api_key)
        
    def register(self):
        """Signal READY status to referee"""
        requests.post(
            f"{self.referee_url}/agent/{self.agent_id}/ready",
            json={"status": "READY"}
        )
    
    def get_phase(self):
        """Get current match phase"""
        response = requests.get(
            f"{self.referee_url}/agent/{self.agent_id}/phase"
        )
        return response.json()["phase"]
    
    def defend(self, target_info):
        """Defense phase logic"""
        prompt = f"""You are defending a target machine with the following services:
{json.dumps(target_info, indent=2)}

Identify vulnerabilities and provide hardening commands."""
        
        message = self.client.messages.create(
            model=self.model,
            max_tokens=2048,
            messages=[{"role": "user", "content": prompt}]
        )
        
        return message.content[0].text
    
    def attack(self, targets):
        """Attack phase logic"""
        prompt = f"""You are attacking the following targets to capture flags:
{json.dumps(targets, indent=2)}

Generate exploit commands to capture flags."""
        
        message = self.client.messages.create(
            model=self.model,
            max_tokens=2048,
            messages=[{"role": "user", "content": prompt}]
        )
        
        return message.content[0].text
    
    def submit_flag(self, flag):
        """Submit captured flag"""
        response = requests.post(
            f"{self.referee_url}/agent/{self.agent_id}/submit",
            json={"flag": flag}
        )
        return response.json()
    
    def run(self):
        self.register()
        
        while True:
            phase = self.get_phase()
            
            if phase == "defense":
                target_info = self.get_target_info()
                actions = self.defend(target_info)
                self.execute_commands(actions)
                
            elif phase == "attack":
                targets = self.get_targets()
                exploits = self.attack(targets)
                flags = self.execute_exploits(exploits)
                
                for flag in flags:
                    result = self.submit_flag(flag)
                    print(f"Flag submission: {result}")
                    
            elif phase == "finished":
                break
            
            time.sleep(10)

if __name__ == "__main__":
    agent = AWDAgent()
    agent.run()

Custom Agent Dockerfile

# agent/Dockerfile
FROM python:3.11-alpine

WORKDIR /app

# Install dependencies
RUN apk add --no-cache \
    nmap \
    curl \
    netcat-openbsd \
    git

# Install Python packages
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Copy agent code
COPY main.py .

CMD ["python", "main.py"]

Troubleshooting

Agent Not Returning READY

Symptom: Match stuck waiting for agents to be ready.

Solution: Check LLM API connectivity from within agent container.

# Get agent container ID
docker ps | grep agent_

# Exec into agent container
docker exec -it <container_id> sh

# Test API connectivity
curl -v https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"claude-3-opus-20240229","messages":[{"role":"user","content":"test"}],"max_tokens":10}'

Target Image Build Failures

Symptom: docker build fails or times out when building target image.

Solution: Configure Docker registry mirror for faster pulls.

// /etc/docker/daemon.json
{
  "registry-mirrors": [
    "https://mirror.gcr.io",
    "https://docker.mirrors.ustc.edu.cn"
  ]
}

sudo systemctl restart docker

Container Resource Exhaustion

Symptom: Containers crash or become unresponsive during matches.

Solution: Increase Docker resource limits.

# Check current Docker resource usage
docker stats

# Increase Docker Desktop resources (macOS/Windows)
# Docker Desktop -> Settings -> Resources -> Advanced
# Set: 8 CPUs, 16GB Memory

# Or limit per-container resources in orchestrator config

# In orchestrator code, add resource limits
container_config = {
    "image": agent_image,
    "name": f"agent_{agent_id}",
    "detach": True,
    "network": match_network,
    "environment": env_vars,
    "mem_limit": "2g",
    "cpu_count": 2,
    "cpu_quota": 200000  # 2 CPUs
}

Match Data Not Persisting

Symptom: Match history lost after referee restart.

Solution: Ensure data volume is properly mounted.

# docker-compose.yml
services:
  referee:
    volumes:
      - ./data:/app/data  # Persist match data
      - /var/run/docker.sock:/var/run/docker.sock  # Docker access

Network Isolation Issues

Symptom: Agents can't reach target machines or each other.

Solution: Verify Docker network configuration.

# Check orchestrator network creation
import docker

client = docker.from_env()

# Create isolated network for match
network = client.networks.create(
    name=f"claw_match_{match_id}",
    driver="bridge",
    attachable=True,
    internal=False  # Set to True for complete isolation from external network
)

# Verify all containers are on same network
containers = client.containers.list(filters={"network": network.name})
print(f"Containers in network: {[c.name for c in containers]}")

Debugging Match State

# Get detailed match state for debugging
import requests

response = requests.get(
    f"http://localhost:8000/api/matches/{match_id}/debug",
    headers={"X-API-Key": os.environ.get("REFEREE_API_KEY")}
)

debug_info = response.json()
print("Container States:", debug_info["containers"])
print("Network Info:", debug_info["network"])
print("Agent Logs:", debug_info["agent_logs"])
print("Scoring Events:", debug_info["scoring_events"])

Advanced Patterns

Template-Based Match Configuration

# Save reusable match templates
import json

template = {
    "name": "4-agent-claude-match",
    "match_duration": 3600,
    "defense_phase_duration": 900,
    "llm_provider": "anthropic",
    "agents": [
        {"agent_id": f"agent_{i}", "model": "claude-3-opus-20240229"}
        for i in range(1, 5)
    ]
}

# Save template
with open("templates/4-agent-claude.json", "w") as f:
    json.dump(template, f, indent=2)

# Load and use template
with open("templates/4-agent-claude.json") as f:
    match_config = json.load(f)
    match_config["llm_api_key"] = os.environ["ANTHROPIC_API_KEY"]
    
response = requests.post(
    "http://localhost:8000/api/matches/start",
    json=match_config
)

Automated Tournament Execution

# Run a series of matches with different configurations
import requests
import time
import os

def run_tournament(match_configs):
    results = []
    api_key = os.environ.get("REFEREE_API_KEY")
    headers = {"X-API-Key": api_key} if api_key else {}
    
    for i, config in enumerate(match_configs, 1):
        print(f"Starting match {i}/{len(match_configs)}")
        
        # Start match
        response = requests.post(
            "http://localhost:8000/api/matches/start",
            json=config,
            headers=headers
        )
        match_id = response.json()["match_id"]
        
        # Wait for completion
        while True:
            status = requests.get(
                f"http://localhost:8000/api/matches/{match_id}/status",
                headers=headers
            ).json()
            
            if status["phase"] == "finished":
                results.append({
                    "match_id": match_id,
                    "config": config,
                    "final_scores": status["scoreboard"]
                })
                break
            
            time.sleep(30)
        
        # Cooldown between matches
        time.sleep(60)
    
    return results

# Usage
tournament_results = run_tournament([
    # Add multiple match configurations
])

FAQ

What does the openclaw-awd-arena skill do?

Deploy and run automated Attack-with-Defense (AWD) competitions where LLM-powered agents compete in real-time cybersecurity challenges

How do I install openclaw-awd-arena?

Run npx skills add https://github.com/reason-machines/hermes-skills --skill openclaw-awd-arena in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/openclaw-awd-arena/ (personal) or .claude/skills/openclaw-awd-arena/ (project).

Which AI agents can use openclaw-awd-arena?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is openclaw-awd-arena safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains openclaw-awd-arena?

It is published by reason-machines in reason-machines/hermes-skills. It has 159 recorded installs.

More from reason-machines/hermes-skills

View all →

hermes-agent-self-evolution

reason-machines/hermes-skills

Evolutionary self-improvement for Hermes Agent using DSPy + GEPA to optimize skills, prompts, and code

AI & Agents323 installs

hermes-webui-agent

reason-machines/hermes-skills

Expert in deploying, configuring, and using Hermes WebUI—a web interface for Hermes Agent with persistent memory, scheduled jobs, and multi-platform messaging.

267 installs

awesome-hermes-agent-ecosystem

reason-machines/hermes-skills

Open the skill to read its SKILL.md instructions.

232 installs

hermes-agent-framework

reason-machines/hermes-skills

Open the skill to read its SKILL.md instructions.

219 installs

hermes-workspace-ai-agent-ui

reason-machines/hermes-skills

Open the skill to read its SKILL.md instructions.

Frontend & Design213 installs

hermes-desktop-companion

reason-machines/hermes-skills

Open the skill to read its SKILL.md instructions.

206 installs

Popular ai & agents skills

View category →

google-agents-cli-publish

google/agents-cli

This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).

AI & Agents513.9K installs

google-agents-cli-eval

google/agents-cli

This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).

AI & Agents513.5K installs

skill-creator

anthropics/skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

OfficialAI & Agents416.4K installs

codex-pet

prime-skills/runcomfy-agent-skills

Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.

AI & Agents402.1K installs

full-output-enforcement

leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

AI & Agents401K installs

executing-plans

obra/superpowers

Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available

AI & Agents245.7K installs